Skip to content
New — orchestrate a team of Mel agents from your terminal. Learn more about orchestrating Mel agents
Mel
Contact salesDownload

Mel – the agentic terminal
The AI terminal for teams.

Changelog

What's new in Mel.

The Mel changelog lists every release, newest first: what changed in the terminal, the team of agents and the VS Code extension.

  1. v0.1.72In progressNovember 2026
    • Factory is on for everyone: it is in the sidebar with nothing to turn on first. You can turn it off in Settings > Features > Factory.
    • Factory opens the pull request itself when a work item is handed off, with what was verified, what was not and any open findings in its description. If it cannot, for example when gh is missing or signed out, the Inbox says why and the Open pull request button does it for you.
    • Factory has a page header with search, a filter and a New menu, an Agents page that lists the Foreman and each stage's agent, and a Productivity tab on the Dashboard. The stages are now called Spec and Implement, and Tools and apps is now MCPs and apps.
    • Tabs can run across the top of the window: Settings > Appearance > Tabs > Tab layout. A new install starts with tabs across the top; an existing install keeps the sidebar.
    • In the tab sidebar each tab shows its state as a mark, and a tab that changed while you were in another tab gets an unread dot.
    • Mel tells you 14 days before your sign-in expires, with the date, once a day.
    • On Windows, a new install puts mel on your PATH, so the mel command works in any new terminal.
    • On Windows, installing or updating refreshes the icon cache, so the taskbar shows Mel's new logo instead of the old one.
    • On macOS 26, Mel's icon no longer sits inside a grey frame.
    • More commands that cannot be undone now ask you first: deleting cloud storage or a whole deployment (gsutil rm, gcloud storage rm, pulumi destroy, cdk destroy, serverless remove, heroku apps:destroy, railway down, doctl delete), publishing (docker push, gh release create, mvn deploy), archiving a GitHub repository, Deno.remove, a recursive chmod or chown outside the project such as chmod -R 000 ~, and an sh -c command that builds what it runs from a command substitution.
    • More hidden commands are read before they run: SQL behind comments, aws with global options in front, PowerShell aliases, an interpreter started behind timeout, stdbuf or nice, more ways of piping text into a shell, GIT_CONFIG_PARAMETERS or RSYNC_RSH set on an earlier line, env -C and env -S, a trap set on an earlier line, and rimraf.
    • /night now runs for up to 12 hours of clock time (it was 10), and the most it accepts is 100,000 credits. On your own key, only the 12 hours limit it.
  2. v0.1.71LatestNovember 2026
    • More commands that cannot be undone now ask you first: a command written with its full path, such as /usr/bin/env or /usr/bin/xargs in front of it; more ways to force or mirror a push set through git config or git remote add --mirror; SQL that deletes with no WHERE in more spellings; and deleting a keychain, a GPG secret key, a stored secret, a cloud secret or key, or a backup.
    • Commands hidden in more places are now read: text with escape codes piped into a shell, a download saved and then run another way, ssh and scp options written together such as ssh -vo, RSYNC_RSH and script --command, code glued to python -c or perl -e, a Python module imported under another name, and a line typed into a Python, Node or Ruby prompt running in the background.
    • On Windows, a Remove-Item fed by a pipeline, and commands run through start, call, forfiles, schtasks or Start-Job, now ask first.
    • sed -E -i and tar xfC now ask before they write outside the project, as other writes do.
    • Reading a credential through the tool that holds it, such as gh auth token or git credential fill, now asks first, as reading a credential file does.
    • In a disposable cloud container, a run's program may write files at a path it works out itself without stopping at a card. On your own computer nothing changes.
    • Billing shows free-trial credits on their own line, with the date they end.
    • On Windows, desktop notifications now show Mel's new logo.
    • Mel can run all night: type /night and a number of credits in a conversation, and the agent carries on by itself until it has used that many credits or 10 hours have passed, whichever comes first, and says which one stopped it. Approvals still wait for you.
  3. v0.1.70November 2026
    • More commands that cannot be undone now ask you first, whatever you set: git steps that throw away uncommitted work, force and mirror pushes set through git config, SQL that empties or drops data (such as DELETE with no WHERE) and database resets such as prisma migrate reset or rails db:drop, unpublishing a package or deleting a release, robocopy /MIR and the Windows erasers, rimraf and cmake -E rm, and download-then-run with curl -O or plain wget.
    • Commands hidden inside other commands are now read too: git's own runners such as rebase -x and bisect run, ssh ProxyCommand and rsync -e, launchers such as env -S, script -c and watch, trap and coproc, scripts fed to Python, Node, Perl or Ruby on standard input, file deletes written inside those languages, and shell-outs from awk, osascript, vim and sqlite3.
    • A program that writes a file through its own option, such as sed -i, dd of=, curl -o, sort -o or tar -C, now asks before it writes outside the project, as > file does, and a path through a link is judged where it really leads. With Stay in this folder on, reading .. or a path like ~name/ asks too.
    • The agent's file tools no longer write inside a .git folder, and a shell command that writes there, or a git config that sets a key git can run as a command, always asks first: such a file can make git run a command later.
    • Secrets stay away from the model in more places: a project instruction file that links to a key or a .env outside the project is not read, reading a credential folder such as ~/.ssh or ~/.aws asks first, a credential file named inside python -c or similar code is recognised, and Mel knows more key names and credential files. In a release build, a project's own .env can no longer turn off the always-ask rule.
    • The agent's pane tools follow the same rules as its commands: text it types into a terminal pane and a command it opens in a new pane get the same checks and approval cards as running the command, and a tool the conversation was not offered is refused.
    • Web requests from the agent have their own approval card, which shows the headers and the start of the body, and your denylist applies to them. Plan mode blocks a web request that would change something, and mel agent --yes asks before a web request to another host.
    • When the agent's browser opens a page on this computer without asking, it can no longer be sent on to another site.
    • Approving one project's MCP server no longer turns on servers with the same name in other projects. Approvals are now kept per project, so a project's server you approved before is off once and asks again.
    • Connecting a connector with OAuth now checks that the sign-in came from the server Mel asked, and refuses one that did not.
    • When the agent sends several command lines at once, the card lists every line. The agent's own terminals belong to the conversation that opened them.
    • Each saved run on the Runs screen shows a scorecard: four marks for efficiency, a check after the last edit, instructions followed and no unproven claims. Hover it for a sentence each; not measured means Mel has no record that answers it yet.
    • Deleting a conversation says so if its file could not be removed, mel mcp serve carries on after a line it cannot read, the Download manually link after a failed update opens only openmel.dev, and the agent's browser keeps Chrome's sandbox on.
    • mel --help no longer lists -p as a short form of --path; mel -p starts the agent.
  4. v0.1.69November 2026
    • More commands that cannot be undone now ask you first, whatever you set: deleting a whole project, find -delete, Windows deletes, deleting cloud resources and deploying to production, publishing a package, commands sent to another machine or into a container, download-and-run in any shell including PowerShell, rsync --delete, erasing a disk, and git aliases that hide a force-push.
    • A command that names a file holding credentials, such as cat .env or an upload of it, now asks first, as reading the file does. Mel knows more credential files, including the Codex and Claude Code sign-ins, .my.cnf, .envrc, Terraform state and Maven and Gradle settings, and it masks lower-case passwords and tokens before the model sees them.
    • Your command denylist is checked before Always allow, and a list you have edited also applies in hands-off and with mel agent --yes.
    • The agent's web requests no longer follow a redirect on their own: the redirect comes back to the agent, and the next address is judged by itself. A PDF or screenshot saved outside the project asks first, and a stray request can no longer end a connector sign-in.
    • The new Folders panel lists the folders a conversation read and changed. Turn on Stay in this folder and Mel asks before its file tools, or a plain command such as cat or grep, read a path outside the project.
    • A first-run walkthrough shows new users around. After a failed step Mel offers Retry and two or three other ways forward, and long tasks show Step X of Y.
    • Mel has a new look: the Manrope and JetBrains Mono typefaces, the brand colours, the spiral beside each answer, and a Sessions | Files switch at the top of the side panel.
    • On a Mac, the app icon now has the standard rounded shape and size in the Dock and in Finder, with no box around it.
    • Codex and other full-screen tools no longer move the cursor when they start, a colour reset no longer leaves (B in the output, and progress lines that redraw themselves show only their last state.
    • A long paste into a program running in the terminal, such as cat or a REPL, now goes to that program and shows the paste line. Only a paste at the shell prompt opens the agent chat.
    • Discard in the Review panel also clears staged changes. A staged new file is unstaged and kept on disk.
    • Settings > Knowledge > Project based lists the rules that apply in the project, and a MEL.md you remove stops applying.
    • A picture or document opened from the file tree opens in your system's app. A program file is shown in its folder instead, so a double-click never runs it.
    • Your keys, your sign-in and your settings are saved so a cut write cannot lose them, and a settings file Mel cannot read is set aside and you are told.
    • Mel checks for updates while it is open, not only at start, and in the last week before your sign-in ends it says so and offers Sign in again.
    • The agent's background jobs belong to the conversation that started them: another conversation cannot read, write to or stop them. A job's long output keeps its start and its end, and says how much was left out.
    • You can delete a single conversation from the Conversations list or from its menu.
    • Settings > Billing no longer shows an expiry date for add-on credits; they do not expire.
    • A run with a credit limit, such as mel agent --max-credits or a cloud job with a budget, now carries on past Genius's per-turn spend cap and counts what that turn cost, instead of stopping there.
    • On Windows, Settings > Features > Command line adds the mel command to your PATH.
    • Claude Sonnet 4.5 retires on November 30 and is no longer offered. A conversation that picked it uses Auto.
  5. v0.1.68November 2026
    • On a Mac, the new Install the mel command links mel into /usr/local/bin, so mel agent and the other mel commands work in any Terminal window. Find it in the command palette or in Settings > Features > Command line. It may ask for your password, and it never replaces a mel that is not Mel's.
    • The new tab shortcut now works while a program such as top or ssh runs in the pane. A new tab also closes Settings, Runs or Projects so you can see it, and the sidebar scrolls to it. If something still blocks the shortcut, such as an open palette, Mel says so.
    • A line such as write ~/notes.txt with a short summary now goes to the agent, instead of running the shell's write command.
    • Settings > Knowledge > Project based has its own Add. It opens the MEL.md of the project you are in, and creates it if there is none.
    • A new Mel Drive prompt keeps the name you give it, and two prompts with the same name no longer overwrite each other.
    • The file tree now shows files the agent or a command creates, without reopening the folder.
    • Mermaid diagrams with brackets or slashes inside a label are drawn instead of showing a parse error.
    • An unattended cloud run now reconnects by itself after a lost connection and carries on, up to three times.
    • The app icon is the new Mel logo.
  6. v0.1.67November 2026
    • mel -p followed by a prompt, and mel --print, now run the agent once on the folder you are in, the same as mel agent. Everything after -p is read as it is after mel agent, and mel --help lists it.
    • Factory is a new built-in skill for work that should be planned before it is built. In the agent box type /factory and what you want built, or pick Factory: spec first, then build in the command palette. Mel reads the files the work touches and your project's own notes, writes a spec to a file in the specs folder, such as specs/billing-page.md, and stops. The spec has a goal, what Mel read, numbered requirements, what is out of scope, a plan of small tasks that each name the check that proves them, risks and a test plan. Mel asks you to approve the plan before it starts the work. Reply approved, use a team of agents and independent tasks run in their own panes. Mel then works the plan one task at a time and ticks each task with what its check printed. A task whose check could not run stays unticked and is marked UNVERIFIED. At the end Mel runs the project's own checks, compares what changed with the plan, and adds an Outcome section that keeps what was verified apart from what was not. Every approval for a command or an edit still asks as usual. Factory is on by default and can be switched off in Settings > Library.
    • The rows of the command palette and of the slash menu are now named for a screen reader: the command, and for the slash menu what it does.
  7. v0.1.66November 2026
    • When you type a plain-English sentence in the terminal and the shell itself fails on it, such as what is rust? ending in no matches found, the block now shows a button under its output: Looks like a question, not a command, with a way to send it to the agent. If an agent chat is open behind the terminal, clicking it brings that chat back with your words in its box, ready to send. If there is none, it opens a chat and sends them. A sentence that starts with a real program still runs as that program and fails in its own words, with no button.
    • Billing and Usage > Usage history now has a By model table: which AI models answered your requests over the last 7, 30 or 90 days, how many each answered and what each cost in credits. Requests on your own key are counted apart and shown as your key. The counts start on the day Mel's server began recording them, and the table says since when, and how many earlier requests have no model recorded. On a Business team, owners and admins can switch from Just me to Whole team to see each current member.
    • When your credits run out and you have no key of your own, a band above the agent box now says Out of credits. On a paid plan it has a pack to pick and a Buy button, with links to Billing and to using your own key; on Free it offers your own key. It can be closed. The footer shows Out of credits with Add credits (Add a key on Free) at its right end.
    • The light themes are easier to find. The command palette now has Toggle Light/Dark Theme, Theme: Light, Theme: Dark, Theme: Match System and a Theme: row for each of the eight themes, such as Theme: Paper, and each pick says which theme you are now using. Typing light, dark, theme or a theme's name in Settings search finds Appearance. In Appearance the theme cards shrink to keep four on a row, so the light themes show without scrolling in the default window.
    • The tab menu in the sidebar has more in it. Right-click a tab row, or use the new menu button that shows beside its close button when you hover. It now offers Copy tab title, Copy working directory, Move Tab Up, Move Tab Down, Close other tabs, Close Tabs Below and Save as new config, which writes the tab's layout as a new tab config and opens it for editing. The colour row starts with a no-colour choice. While you have text in the sidebar search, the four move and close-many rows are greyed with Clear the sidebar search first, so they never move or close a tab you cannot see.
    • Projects are new, and off until you turn them on in Settings > Features > Projects. With them on, the sidebar gets a Projects row and a Pinned list, and a Projects page with a card for each project. A project has instructions, memory and context files, and every session started in it carries them. Drag a conversation onto a project, or use Move to in the tab menu; it can be moved to another project but not yet taken out of one. Everything stays on this computer: Mel's servers store nothing about your projects.
    • The grey hint in an empty terminal prompt now ends with an example task that changes with each new session, such as Run the tests and fix whatever fails. A session keeps its example while it is open, so the text does not move under the cursor. The agent box keeps its own hint about attaching a file.
    • Export conversation now says what it holds word for word. The README and transcript.md state that the commands and outputs are the saved rows, not a summary, and that the note on a reopened conversation about a summary is about what the agent remembers, not about the export. When Mel replaced parts of a conversation with a summary or a Cleared from context pointer, the export also holds wire-original.jsonl, the messages as first recorded, beside wire.jsonl, which shows the replacements. A command the transcript shows on one line is also written out exactly, in a block.
    • The agent's grep tool now uses plain grep when git cannot search: git is not installed, git fails for a reason that is not your pattern, or a Mac has no developer tools. Before, it answered that grep was unavailable and searched nothing, and on a Mac without the developer tools it found nothing and said nothing. The limits are the same (80 matches, 256 KB, 20 seconds), node_modules, target and credentials folders are skipped, and the result notes that .gitignore was not applied.
    • When the agent runs a command with a timeout under 10 seconds, Mel now waits 10 seconds and the result says the bound was raised. The timeout is in milliseconds, so 1000 is one second, and before, such a call killed a download or a build after one second. A command stopped at a timeout the agent named now says the unit, and a command with no timeout still moves to the background after 120 seconds.
    • The new menu button on a sidebar tab is named Session menu for a screen reader, and so are the buttons of the Close tabs question and the quit question.
  8. v0.1.65October 2026
    • Pasting a long prompt (16 KB or more) into the agent box now shows it as a short chip, such as [Pasted text #1: 1,240 lines, 48 KB], and sends all of it. Delete the chip and the paste is left out; undo brings it back. A long message you have sent is shown as its first lines with a Show all line. Pasting it into the terminal prompt opens the agent chat with the chip. A paste larger than 4 MB is refused with its size and the limit, and is never cut. A large paste into a coding agent in a pane is now sent in small pieces in the background, and Ctrl+C stops the rest.
    • The + button in the agent box is now a menu: Attach files or Attach folder. Any file can be attached. Images (and PDFs, for a model that reads them) attach as files; any other file or folder is added to your message by its path, and the line above the box says so. Before, the + opened a file window that offered only images and PDFs, so on a Mac only a folder could be picked.
    • On a Mac, Command-Q (⌘Q) now asks before it quits while a build, a server, a command or an agent turn is running, as the red close button already did. Pressing it again while the question is showing does not answer it, so holding the key cannot quit by itself. A command you typed into an agent chat now counts as running work for both. Quit from the Dock, and logging out or shutting down, still quit without asking.
    • A terminal with thousands of blocks stays quick. The block list now draws only the blocks on screen, and only the lines of a long block that are on screen, so typing no longer slows down as the history grows and one 1 MB block opens without a stall. Before, every block was drawn on every key press. Screen readers now hear the blocks on screen, the last block and the selected one, not every block.
    • On Auto Genius, a cloud task now carries on in a new turn, once, when it reaches the per-turn credit limit, where it used to stop there. A cloud task can therefore spend up to 2 times that limit (on Pro, 2 x 300 of the 1,500 monthly credits). A worker you run yourself carries on up to 3 times, and MEL_WORKER_GENIUS_CONTINUE=0 turns that off.
    • A cloud task, or a task started by a cloud trigger, with no Auto mode picked now runs in Genius instead of Balanced, which can use more credits. Choosing Balanced while Balanced is already your default counts as picking no mode. A mode named for a worker you run yourself with MEL_WORKER_BIAS, and a pinned model, are unchanged.
    • The Auto mode you pick on the desktop now reaches the cloud task you start from it. A cloud trigger can name the mode its task uses too, with a bias field on POST /v1/triggers/run set to responsive (Balanced), cost-efficient, genius or open-weights.
    • The agent no longer searches your whole home folder or your whole computer. A search with its grep, find_files or glob tools, or a command such as rg, find, grep -r, tree or du, over your home folder or / is refused, and the agent is told to search a project instead. A subfolder or a project is not affected. The app also asks you first before it runs such a command. Cloud tasks and hands-off runs are not held to the command part.
    • A sentence in the agent box that starts with a word that is also a program, such as "Make a local file ~/x containing a heading" or "find the file that has foo in ~/src", now goes to the agent instead of running in the shell. Commands such as make clean, make all, make test, test -f x and open . still go to the terminal, and a line that starts with ! still runs in the shell.
    • The agent's question card now says how long it waits for your answer, for example: If you do not answer within 30 minutes, the work continues without it. Before, it said only that the work continues if nobody answers in time.
    • A background wait with an empty pattern now waits for the job to end or for its time to run out. Before, it returned at once while the job was still running, so the agent could think a long task had finished early.
    • The conversation switcher now finds conversations older than the newest 200. It searches all of them, shows the newest 200 matches and says how many more it left out.
    • The search box in the Tools panel's conversation list now searches all your saved conversations, not only the newest 50. If more than 50 match, it says how many it left out.
    • A new rule added in Settings > Agents > Knowledge is now saved in Mel Drive > Personal, so it shows in the Drive tree and can be moved to the Trash from there. Before, it was saved at the top level of the Drive, where the tree did not list it. Rules an earlier version saved there stay where they are and still apply.
    • When a command's output is cut at the 1 MB limit, the notice now says how much more was not shown, such as 2.4 MB, instead of only that the output was cut.
    • When the agent's file write, edit or patch fails (a full disk, a refused rename), or a save in the review editor fails, it no longer leaves a hidden temporary file ending in .mel-tmp next to yours. Your original file is untouched, and you still get the real error.
    • What's new in the app now shows this same list of release notes, word for word, and opens on the notes of the version you are running. Before, it showed a list that stopped at v0.1.7.
    • Ctrl-click (Command-click on a Mac) on a file name in a command's output now opens it for more commands: git status, diff, show, log, ls-files, ls-tree and stash, and fd. A name followed by a line and column, such as stats.py:12:5, links to the file. Before, only ls output offered file links.
    • Ctrl+I, which sends a line to the agent instead of the shell, now shows where the line will go: the hint under the box reads sends to Mel, not the shell, with Ctrl+I to undo. Before, the key seemed to do nothing.
    • Opening a search result from global search now takes you to its line: the editor puts the cursor on the line and scrolls it to the middle. A file that cannot be read now says so instead of opening nothing.
    • When you edit a step of a plan before approving it, the agent is now told that your version replaces its own and which lines you changed. Before, it could build from its own version. Choosing Keep planning still drops your edit.
    • A signed-out first launch now shows a Sign in button beside the avatar. In the sign-in card the provider buttons come first, then Sign in with browser, then email and password with Forgot password? on the password row, and Cancel is last. The empty password field no longer shows dots, and signing in shows a short notice, Signed in as your email.
    • Pasting an image copied by another program on Linux (X11) now attaches it. The notice that the agent chat opened with your image is larger and clearer. Pressing Ctrl+V on an image while the command palette is open now says there is no text on the clipboard and to paste the image into the agent box.
    • When every request to the provider of your own API key is refused for its rate limit or request size limit, Mel now says so, names the provider and the ways out, instead of calling it a provider outage.
  9. v0.1.64October 2026
    • New: mel agent runs Mel's agent for one request from a terminal or a script. mel agent "fix the failing test" works on the current folder, streams the answer to your terminal, asks before it runs a command or changes a file (--yes answers those, and nothing that always asks), and ends with an exit code a script can read. --json prints events, --max-credits stops it starting another continuation once that many credits are spent, and the conversation appears in the Mel app.
    • Signing in with Google, GitHub or Apple now takes you straight to that provider in your browser and back to Connect the app, instead of asking you to sign in to the website first. If the sign-in code runs out, the card says Code expired and offers Get a new code.
    • Pasting a screenshot (or dropping an image) into the terminal prompt now opens the agent chat with the image attached and your typed text kept. The agent box says how to attach images: click +, paste, or drag a file in.
    • New in the model picker: Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed.
    • Mel Red Team (the mel redteam command, run from a terminal) now runs your project's tests and linter, and the command each finding suggests, on Linux with bubblewrap 0.8 or newer. They run in a fresh copy of your code held in memory, inside a sandbox Mel checks first: your files are not touched and there is no network. A finding is marked Reproduced only when the command printed the failure its author named. On macOS and Windows nothing runs yet and the report says why, and --review-only runs nothing anywhere.
    • Mel Red Team: a model whose review ends without its findings block is no longer counted as having found nothing. Mel asks it once more, and the report names which models gave no block and what became of each.
    • Desktop control (Settings > AI, off by default): every use now asks you first, as the setting says. Before, an Always allow for commands or the fast-forward button could skip the card.
    • A line that starts with ! in the agent's input now runs in the shell even when Settings > AI > Autodetect terminal commands in agent input is off. Before, the hint under the box said so but Enter sent the line to the agent. A shell line sent while another shell command in the conversation is still open stays in the box and says why.
    • A command the agent runs with a here-document is no longer refused as unfinished when the document's text contains <<, such as 1 << n in C code.
    • The plan list in Settings > Billing and Usage now says what the website says: Pro names Claude, GPT, Gemini, DeepSeek, Qwen and more, Business is on the five US model providers only (Anthropic, OpenAI, Google, xAI and Groq), and Free no longer lists web searches.
    • A workflow in Mel Drive is lavender instead of coral, which looked like the colour of an error.
    • The README inside an exported conversation asks you to write to hello@openmel.dev.
  10. v0.1.63October 2026
    • Mel's stable updates are now signed with a new key. If you use the stable channel and run any version from v0.1.56 to v0.1.62, install this version once from openmel.dev, or switch the update channel to Beta in Settings, update, and switch back to Stable. After that, updates arrive by themselves again. Versions before v0.1.56 update by themselves.
    • A command that first moves to another folder and then deletes there now stops for your approval. Mel follows each cd on the line and judges a delete from every folder the line passed through, so a line like cd / && rm -rf * is stopped.
    • Settings is arranged in groups (Agents, Code, Cloud platform), which start folded. The group with the open page opens by itself, and a search opens every group that has a match.
    • Settings > Cloud platform > API keys: see, create and revoke your Mel API keys. A new key is shown once, with a Copy button.
    • The conversations list in the Tools panel has a search box, a New conversation button and two-line rows with the source and age. A conversation whose last request was answered shows a green check, and View all opens the Runs screen.
    • Mel Drive has a team section, a Trash (Move to Trash, Restore and Delete forever), and a Rules page in Settings > Knowledge. Notebooks and plans have a breadcrumb and an Editing / Viewing switch, and opening a workflow docks its card above the input.
    • Settings > Knowledge has Global and Project based tabs, a Search rules box and + Add for a new rule; each rule is a card.
    • Click a workflow in Mel Drive and its command goes into the input, ready to run or edit.
    • Security: updated the library that checks website certificates, closing three published advisories.
  11. v0.1.62October 2026
    • Settings > Third party CLI agents: a toolbar under Claude Code, Codex or Gemini in a pane, built from the chips you choose and arrange (changed files, git branch, pull request, time and more), with your own command patterns for when it shows. Rich Input, Mel's compose box for these agents, can open when a session starts, show and hide as the agent works and waits, close after you send, and send with Ctrl+Enter.
    • Settings > Indexing and projects: Mel can build a map of each folder's files and the code symbols in them, on this computer only. It is off by default, you add and remove folders on the page, and agents do not use the map yet.
    • Settings > Editor and Code Review: absolute or relative line numbers, open the review panel on a conversation's first accepted change, turn the project explorer and global file search off, show hidden files in the explorer, and auto save (a second after you stop typing, and when you leave the editor).
    • Settings > Melt: two lists of command patterns, Added commands that always open full screen and Denied commands that never do, on top of the built-in ssh, coding-agent CLIs and full-screen programs such as vim, less, top and tmux. A pattern that could never work, or would match nearly every command, is refused with the reason.
    • Settings > Profiles is now Settings > Agent mode, and its permissions take effect. Choose how much the agent may do without asking for edits (Agent decides, Always allow or Always ask) and for commands (Always allow or Always ask), keep a command allowlist and denylist, and set whether orchestrated agents may start: Never, Always allow or Always ask. The denylist is checked first, and a step that cannot be undone, a write outside your project and a connector call still ask whatever you choose.
    • Stop now also stops a wait. Pressing Stop while the agent waits on a command or a background job ends the wait at once, where it held the turn for up to 10 minutes; the job keeps running, and the Stop message says so. A web request or a connector call still runs to its own end.
    • The conversation's running background jobs are listed above the input, each with its command, state and clock, and a Stop of its own that ends the job and everything it started, then tells you whether it stopped.
    • A command that the agent runs with its longest time limit (10 minutes) is now handed to the background after 2 minutes, like one with no limit, instead of being killed at 10 minutes. A shorter limit it chooses still stops the command.
    • A background job's last lines are no longer missing from what the agent reads when it exits, and a conversation is no longer told that a server stopped when another conversation started it. A job waiting at a prompt (a password, a y/N question) shows that prompt to the agent, which can also close a job's input to finish a program that reads until its input ends.
    • More commands that cannot be undone now need your approval: git stash clear and drop, discarding changes with git checkout -- or git restore, git branch -D, git gc --prune, git reflog expire, git update-ref -d and git filter-branch or filter-repo. A shell command that writes a file outside your project (with > or a heredoc) needs it too, and a heredoc is read as the commands it may contain.
    • A bare Enter no longer approves an approval card nobody looked at. Enter stays with the box while it holds text or has focus, Shift or Alt with Enter never approves, Allow ignores clicks and keys for a moment after a card appears or the window comes back to the front, and a step that cannot be undone needs a click or Ctrl+Enter (Cmd+Enter on a Mac). Esc still denies at once.
    • New in the model picker: Claude Haiku 5.5, GPT-6 Luna, Gemini 4 Argon, Grok 4.7, GLM 5.3 FlashX, MiniMax 3.1 Flash, and GLM 5.3 and Kimi K3 on DashScope.
    • Picking /plan in the slash menu now puts /plan in the box for you to finish, and plan mode starts when you send the message.
    • A Mermaid diagram whose box words look like Mermaid's own syntax (a box such as API[/GET /api/sites]) is now drawn, where it used to show as code with a "couldn't read this diagram" error.
    • The dictation microphone is the same icon in the agent's input and in the terminal's toolbar.
    • On a Mac the command palette showed Control keys for shortcuts that use Command (New Terminal Tab said ⌃T; the shortcut is ⌘T). The keys it shows now match the shortcuts.
    • The agent's code search no longer fails on a pattern with a lone { in it, such as return {. On Linux that was a pattern error that found nothing; it now matches the brace on every system.
    • Importing a skill from a .zip: a zip named like ..zip can no longer unpack outside the folder Mel unpacks it in, and file names such as aux.py or con.md import on macOS and Linux (only Windows refuses them). Mel also uses newer versions of the library that unpacks zip files and the one behind its secure connections, which fix published security advisories.
  12. v0.1.61October 2026
    • The Auto mode you pick (Balanced, Cost-efficient, Genius or Open-weights) is remembered for new conversations, and the model chip shows it. You can also set it in Settings > AI > Default models.
    • The four Auto modes now choose differently: Genius leads with the strongest models, Cost-efficient spreads over cheaper ones.
    • The agent no longer runs the same command over and over: a third identical run with nothing changed in between is stopped, and a task that stops making progress ends with a report of what was done and what is left instead of carrying on by itself.
    • Starting a dev server or reading a file back no longer counts as having checked the work.
    • Asked for a diagram, the agent draws it in the chat as a Mermaid diagram instead of opening a web page.
  13. v0.1.60October 2026
    • One screen for all your agent runs on your computer (Ctrl+Shift+M): Needs you, Running and Recent. Needs you shows Mel's agent waiting for an approval or an answer, and Claude Code or Codex waiting in a tab. Cloud runs are coming soon; today they are open only to a small test group.
    • For the small test group on cloud runs (coming soon for everyone else): a cloud run that starts waiting for you sends a notification; click it to open the runs screen.
    • New connectors in Settings > Connectors, each on your own key: Slack, Linear (now also creates and updates issues), and read-only PostHog, Airtable, Postman and Cloudflare.
  14. v0.1.59October 2026
    • Review the agent's changes one at a time: accept or undo a single change, approve the whole review, attach the diff to your message, and send review comments with Ctrl+Enter. The git chip shows how far your branch is ahead of or behind its upstream.
    • Pin files and notes so the agent sees them on every turn: /pin, /unpin and /pins, the Explorer's right-click menu, or Settings > Knowledge.
    • Mel reopens your tabs, panes and folders after a restart (layout only, never running commands), and asks before quitting while a command, job or agent is still running.
    • When Claude Code, Codex or another CLI agent in a Mel tab stops and waits for you, Mel tells you, even from another tab.
    • See what each answer cost and what the conversation has cost so far.
    • Eight themes in Settings > Appearance, with a separate choice for dark and light.
    • Screen readers (VoiceOver on macOS, Narrator on Windows) can read Mel's own buttons and cards.
    • Pictures and videos the agent makes appear in the conversation, and the agent can edit images and videos with the tools on your computer (ImageMagick, ffmpeg).
    • When the agent needs an account you have not connected, the conversation shows a Connect button and a button to carry on.
    • A connector approval says what the call will do. Approvals for steps that cannot be undone are answered only at your computer, never from your phone or a notification.
    • More keyboard shortcuts, on macOS too; Paste in the block menu; optional right-click paste and focus follows mouse; every on/off setting can be toggled from the command palette.
    • An agent working over SSH reuses one connection, so its steps run faster.
    • Xiaomi MiMo v2.6 Pro and Flash, with your own MiMo key (Settings > Billing and Usage).
  15. v0.1.58October 2026
    • Mermaid diagrams are drawn as pictures in answers and plans: enlarge one, see its code, or save it as PNG or PDF in the conversation's folder. Diagrams are drawn on your computer.
    • Desktop notifications when an agent finishes or needs you while Mel is in the background. On Windows you can click one to open the conversation, and answer an ordinary approval from it. Turn them off in Settings.
    • Ask Mel to connect to your GitHub, Vercel, Netlify, Supabase, Cloudflare and more: it uses the sign-in already on your computer, or starts the service's browser sign-in for you. No connector or token needed.
    • Skills you ask the agent to save are now saved where Mel reads them on every system (on Windows, under %APPDATA%), as a folder with a SKILL.md, so a new chat finds them.
    • The Preview panel has a Reload button, and Python's built-in web server (python -m http.server) is now found.
    • The chip row under the input no longer flickers between one and two rows in a narrow window.
  16. v0.1.57October 2026
    • Grafana joins the Featured MCP servers in Settings ▸ Library: add your Grafana address and a service-account token (kept in your keychain) and ask Mel about your dashboards, data sources, alerts and incidents. It runs read-only.
    • Terraform's Featured server takes an optional HCP Terraform token: with it, Mel can also read your HCP organizations, workspaces and runs (never variable values, state or run logs).
    • New connectors in Settings ▸ Connectors, each on your own key and read-only: Linear, Neon, Hugging Face and Sentry.
    • Shopify's connector can now create and update products, set inventory, create collections and archive products — every change still needs your approval.
    • Context7's optional key is now sent the way Context7 documents it.
    • A Featured server's key can only be read by that Featured server, never by a same-named server a project or team file adds.
  17. v0.1.56October 2026
    • Featured MCP servers in your Library: Settings ▸ Library ▸ MCP servers now offers Playwright, Chrome DevTools, Context7, Docker, Terraform and Kubernetes. Each is off until you turn it on, runs the version Mel tested, says what it needs on your computer (Node.js, Chrome, Docker, a cluster) and links to its docs; Customize copies one into your own servers.
    • A local MCP server now stays open for the whole conversation, so a browser opened by one call is still there for the next. It closes, with everything it started, when you turn it off, change it, close the conversation, leave it idle for 10 minutes, or quit Mel.
    • A server's first start gets up to 2 minutes (a first download), Mel says when it started a server or had to restart one, and a server that needs something your computer lacks says what and where to get it.
    • Screenshots an MCP server returns are saved as files in your project's mel-images folder, and a local MCP server runs in your project's folder.
    • Safer local addresses: Mel reads only real local addresses (localhost, 127.0.0.1, ::1) without asking, and any request that sends data to a local port needs your approval. A web name that merely looks local is treated as the outside website it is.
    • Version checks of tools that read a project's own settings — yarn, pnpm, cargo, rustup, mvn, gradle, poetry, bundle, composer — now need your approval, so a downloaded project can't run its code without your OK.
    • A risky command option written in a shortened form now asks, the same as the full spelling.
    • Mel asks before reading more sign-in files: the GitHub CLI, Docker, Kubernetes, Google Cloud, Azure, Cargo, RubyGems and Terraform logins.
    • Updates must be signed by Mel. An update that isn't is refused, and you keep the version you have.
    • Fixed: a command containing certain accented characters could stop the agent's turn.
    • On a Mac: drag a screenshot, a photo or an image from your browser into Mel to attach it, and dictate with the mic in the prompt box.
  18. v0.1.55October 2026
    • Plans are now editable. When Mel proposes a plan, change, add or remove steps in the plan panel, then approve it as you edited it — or hand it to a team of agents with “Use orchestration”. Hide or show the panel with Ctrl+Alt+P, and save a plan as a Markdown file or to Mel Drive.
    • Keyboard shortcuts can be changed in Settings ▸ Keyboard shortcuts. On Windows and Linux, closing a session is now Ctrl+Shift+W, so Ctrl+W deletes a word in the terminal again; Ctrl+1 to Ctrl+9 switch tabs. On a Mac, Mel's main shortcuts stay as they were.
    • Your conversations open in a side panel with search: Ctrl+Shift+Y (⌃⇧Y on a Mac).
    • A message you send while Mel is thinking joins the conversation right away. Commands and slash commands still wait for the current step to finish.
    • For the small test group on cloud runs (coming soon for everyone else): start a message with “& ” to run it as a task in the cloud. Mel asks you to confirm first.
    • Tools ▸ Preview now shows your running local page live: click, scroll and type in it. It needs Chrome or Edge installed; without one, Preview shows a screenshot as before.
    • The agent's questions show as a card with numbered options and a recommended choice, and each answer can show its token usage.
  19. v0.1.54October 2026
    • The agent no longer stops to ask before commands that only read — ls, cat, grep, git status, git diff and the like. A closed list decides, so anything that could change a file, start a server or reach the network still asks. To be asked for every command again, turn on “Ask before read-only commands too” in Settings.
    • The approval card for a file edit now shows exactly what will change, as a diff of the lines added and removed.
    • A command that is still running after two minutes is moved to the background instead of being stopped. Its output is kept, and the agent can check on it or stop it later.
    • Claude Sonnet 5.5 and GPT-6.1 Sol are now in the model picker.
    • When an edit leaves a file that no longer parses — JSON, YAML, TOML, Python, JavaScript or a shell script — the agent is told at once, with the line, so it fixes the break before moving on.
    • Security: commands the agent runs no longer receive Mel’s own sign-in token or API key in their environment.
  20. v0.1.53October 2026
    • MCP servers you add or switch on are picked up by your next message, without restarting Mel. A server that is set up but switched off is now named as off, with where to turn it on (Settings ▸ Library), instead of the agent saying it has no MCP tools.
    • Pasting an mcpServers block now adds every server it can and lists the ones it skipped and why — one unsupported entry used to stop the whole paste. ${VAR} values in a pasted block are read from your shell’s environment.
    • When an MCP server cannot start, Mel now says why: the server’s own error message, the program that was not found, or the time limit it ran past.
    • The GitHub and Vercel connectors can now create a repository, commit your files to it, and deploy — from the GitHub repository, or straight from a small folder. Each step asks you first. A commit now shows as its own card.
    • Fixed: the agent’s commands now run in the folder it names. They ran in the conversation’s folder instead — a “git init” for a project landed in the home folder. The agent’s own time limit for a command is honoured too, and on an SSH session a background job is refused rather than run on your computer.
  21. v0.1.52October 2026
    • Fixed: the agent now sees a command’s error messages. Mel used to pass on only the errors of the last command in a chain, so in “javac Temp.java && java Temp” a missing Java runtime was invisible and the agent guessed a reason. Every command’s output and errors now reach it, in order. Commands with a here-document (<<'EOF' … EOF) work again — they used to fail with no output.
    • Fixed: a command whose output contains bytes that are not valid text (a binary file, or accented text from Windows tools in some languages) no longer loses the rest of its output and reads as failed. The same applies to background commands and to MCP servers.
    • Mel now always asks you before the agent turns off SSH host-key checking (ssh, scp or sftp with StrictHostKeyChecking=no), deletes Docker containers, images or volumes (docker rm, rmi, volume rm and the prune commands), or writes a file outside your project — whatever your approval setting. Your system’s temporary folder stays free for scratch files.
    • Keyboard shortcuts on Windows and Linux now use Ctrl+Shift, so Ctrl+D, Ctrl+P and Ctrl+T reach the program in your terminal: command palette Ctrl+Shift+P, new tab Ctrl+Shift+T, split right Ctrl+Shift+D, split down Ctrl+Shift+E, Tools Ctrl+Shift+\. New: Ctrl+, opens Settings and Ctrl+= / Ctrl+- / Ctrl+0 zoom. On a Mac nothing changes.
    • Settings ▸ Knowledge: switch your rule files on or off for the agent, add a rule, and create a MEL.md for your project. Settings ▸ Appearance ▸ Text cursor: bar, block or underline, blinking or not.
    • Billing ▸ Usage history shows the last 30 days and one card per conversation, with its tokens, tool calls, models, files changed and commands run.
    • “Choose folder…” in the folder menus of the terminal and the agent; a close button on notifications; one Connected/Disconnect button per connector; on Windows, Mel uses Segoe UI for its interface.
    • Groq retired Qwen 3.6 27B, Compound and Compound mini, so they are gone from the model list. When a provider no longer serves a model, Mel now says so instead of calling the model “incompatible”.
  22. v0.1.51October 2026
    • Fixed: an agent search could use up all of your computer’s memory. Mel’s search tools now search the folder the agent asks for, stop at a fixed number of results, size and time, skip heavy and private folders (node_modules, build output, caches, .git, .ssh and similar), and never search your whole home folder. On Linux this could close your whole desktop session.
    • On Linux, Mel now runs in its own memory-limited space, so if it ever runs out of memory only Mel closes — never your desktop.
    • Secret masking no longer hides ordinary settings. Code that reads configuration (process.env.PORT, os.getenv(…), config lookups) is shown as written, and plain values in a .env file — a port number, a log level, a local address — stay visible. Keys, tokens, passwords and secret-looking values are still masked.
    • When the agent makes an audio file, it now knows the file plays right in the conversation, on the audio card.
  23. v0.1.50October 2026
    • A new look. The conversation page, the terminal page, the model picker, the title bar and side tabs, the menus, the command palette, notifications and every Settings page are redesigned, with new type (Roboto, and Hack Bold for code). Colours now mean one thing each: green for good news and for anything that needs you, pink for a problem, red for a failure — Mel no longer uses amber or orange anywhere. On Windows, Mel draws its own window title bar.
    • The Library: skills, prompts and MCP servers in one place (Settings ▸ Library). Mel ships built-in skills and prompts, and adds your own (~/.mel), your project's (.mel/), and — for Business teams — the ones your admin shares. Run a saved prompt by typing /its-name in the chat; Mel asks for anything it needs to fill in. Import a skill from a GitHub link, a folder or a .zip, or paste the mcpServers block you already use in another app. Anything you import starts switched off until you have read it, and a project's or team's MCP server asks you once before it can run. The same Library is in the command line (mel skill, mel prompt, mel mcp) and in Mel for VS Code.
    • Mel reads the project rules and skills other coding agents use: AGENTS.md and CLAUDE.md as project instructions, and skills in .claude/skills and .agents/skills. When the agent uses a skill, the conversation shows “Using skill: name”.
    • Stop means stop. After you press Stop, Mel no longer picks the stopped task back up on its own; an approval card you stopped shows “Not run — the turn was stopped”, never “denied”; and turning plan mode off is recorded, so a later summary of the session gets it right.
    • No more “check your work” loops. When Mel had finished and checked a task, a safety check could still send it round again and again without answering; it now gets a few rounds at most and then must answer. Plan mode likewise stops after three unapproved plans and asks you.
    • Every finished answer shows its usage line (model, tokens, cost, time) — it was missing whenever a card was drawn after the answer.
    • Mel’s own speed limits no longer stop your work. Several chats or agents at once used to hit a limit of 10 requests a minute and fail; the limits are much higher now, and if one is ever reached Mel waits a moment and carries on, telling you how long.
    • When the model Mel picked is unavailable, the conversation now says which one answered instead.
    • Better model choice: tasks that add a feature with tests, set up CI or fix a failing test now get a stronger model, and plans and design questions are treated as such.
    • Security: opening a repository you do not trust can no longer run code through its git settings. Mel’s own git commands now run with those settings switched off.
    • On Windows, terminal output no longer runs lines together, and MCP servers started with npx now start.
    • Smaller things: a search box in Settings ▸ Connectors and Keyboard shortcuts; /plan your task turns plan mode on and sends the task; the + menu’s Terminal row opens a terminal; the / menu works with the arrow keys and the mouse; the sign-in dialog no longer cuts off the create-account link; and the free-key card counts the credits you were given.
  24. v0.1.49October 2026
    • Mel’s agent now always runs on Mel’s server: it thinks and plans there, while your files and commands still run on your computer, after your approval as before. The “Run the agent on Mel’s server” switch is gone, and if Mel’s server cannot be reached the agent says so instead of quietly running an older copy built into the app.
    • Updates are signed. When the update list carries our signature, Mel checks it and then checks the download against the fingerprint the signed list names. The next release will refuse an unsigned list.
    • On Windows, commands with quotes work as typed — mkdir "two words" or git commit -m "fix bug", in the terminal or run by the agent. Before, the quotes reached Windows escaped, so the command failed, or mkdir made the folder at the top of the drive.
    • On Windows, picking a folder with the folder button now moves the agent there (the folder change used a command Windows refused), and a new agent chat starts in the folder you are working in, not the one Mel started in.
    • Ticking “Always allow file edits in this conversation” now says so in the conversation, and the new /always-ask command turns the questions back on. Typing /plan turns plan mode on.
    • The preview of a running website says what it is: a screenshot of the page, with Open in browser for the live site.
    • On macOS, Mel no longer asks for Keychain access again and again. It reads each saved key once per session, and Settings never waits on the Keychain while it draws — before, a key the Keychain did not trust could raise a dialog on every agent step and freeze Settings behind it.
    • Starting a new conversation while another one is still working now opens it in a new tab, with your text sent there; the running one carries on. Before, the new conversation was refused, or your message queued into the old one.
    • Secrets stay masked. Anything in a file or in command output that looks like a key or a token is replaced before it reaches the model, and you and the agent are told how many were masked. Reading a credential file such as .env always needs your approval, whatever your setting. And Mel no longer loads the .env file of the folder it was started from into the agent’s commands — only its own MEL_* settings.
    • A command that cannot be undone always asks, even on “Agent decides” or fast-forward, and never offers “always allow”: deleting outside your project, a force-push, git reset --hard or git clean, dropping or truncating a database table, publish and deploy commands, piping a download into a shell, sudo, and a few more. With nobody at the desk it is refused. Deleting inside your project is an ordinary command and follows your setting.
    • A connector call always asks too, in hands-off modes as well.
    • The plan card: typing a reply now revises the plan with your feedback — before, anything typed except “Keep planning” approved it, so “no, don’t touch the database” approved the plan. Only the Approve button approves. And /plan now turns plan mode on.
    • What the agent reads from outside — files, command output, web pages, connector results — now reaches the model labelled as data, under one rule that data is never an instruction, so a README that says “ignore the user” is summarised, not obeyed. Skills you load are Mel’s own words and are followed.
    • The model pickers list only models that run: NVIDIA’s rows (Mel holds no NVIDIA key), Qwen rows Auto never uses, and codex rows the provider no longer serves are gone. A model you pinned earlier still works.
    • Settings ▸ Privacy no longer says provider keys are never stored on your device — a key you add yourself is, in your OS keychain. The default Auto bias is now called “Balanced”; “Responsive” promised a speed nothing guarantees.
    • Edits land more often: the agent’s file edits match through Windows line endings and changed indentation (only when the match is unique), keep the file’s line endings, name the nearest line when the text is not there, and replace the file in one step instead of rewriting it in place.
    • The agent checks its work with your project’s own command: it reads the test or check command from the project (a line in AGENTS.md wins) and uses it to verify a change.
    • Mel now reads AGENTS.md and CLAUDE.md beside MEL.md, loads skills in the open SKILL.md format from ~/.mel/skills and from a repository’s .mel/skills or skills folder, and ships ten more built-in skills (code review, debugging an error, fixing failing tests, writing tests, a README and more) and nine starter prompts in Drive ▸ Personal ▸ Starter prompts.
    • On a Free account with no credits and no key, the agent’s start screen now says where to get a free key (Groq or Google AI Studio) and where to paste it — Settings ▸ AI ▸ Provider API keys.
    • If Mel’s server cannot be reached, a paid plan keeps working for up to 7 days on the last plan confirmation the server signed, and the app says until when.
  25. v0.1.48October 2026
    • Mel now reports crashes so they can be fixed: when something inside its window goes wrong, or it crashes, a report goes to Sentry, with a start/stop signal per session. A report never contains your terminal output, prompts, files or keys. Settings ▸ Privacy says whether your build reports, and `MEL_SENTRY=0` turns it off.
    • A file the agent saves for you to open now has the type of the tool that made it: a PDF is always a .pdf, and audio is always an audio file. Before, an answer could get Mel to save a document under a name your computer would run as a program.
    • With the server switch on (Settings ▸ Features ▸ Agent ▸ “Run the agent on Mel’s server (beta)”): Send now works; pressing Stop while the agent makes several changes stops the next change from being made; the Preview chip appears for a server a command started; when Mel checks the agent’s work you see the same one short line as with the switch off; and a question the agent answered from your earlier answers gets its card.
  26. v0.1.47September 2026
    • Mel no longer closes when something inside its own window goes wrong: it saves a crash note, redraws, and says so once. If the same error keeps coming back, it offers to save your work and quit.
    • Fixed a crash while an answer was appearing on screen, when a piece of inline code in it held an accented letter, an arrow, a check mark or an emoji.
    • Fixed more crashes: on every start after a bad zoom value in the settings file, after closing the last tab while the default tab setup was broken, on a tab setup that contained itself, on the Team page, and on a damaged system font.
    • Windows laptops with a Snapdragon (ARM) chip get their own native build instead of running the Intel one under emulation, where the graphics driver could stop in the middle of a session. Mel installed on such a laptop moves to it through the updater.
    • Command output can be selected past the edge of the screen and copied in one go. Ctrl+C stops a running command again — on Windows and Linux it had quietly stopped working.
    • In a conversation, a selection keeps working while you scroll, and every answer, message, tool result and shell card has “Select text…”, which opens the whole text so any amount can be selected and copied. With text selected, Ctrl+C copies instead of stopping the agent.
    • Send now: a message you type while the agent works can join its next step instead of waiting for the turn to end — the button on the queued message, or Ctrl+Enter.
    • Research answers no longer come with notes underneath. When Mel checks the agent’s work you see one short line, with what Mel told the agent behind it, and an honest “I couldn’t verify this” now ends the turn instead of being stopped.
    • The agent is no longer locked out of running commands after three unrelated commands fail: each command’s failures count on their own, an edit clears them, and the check that follows an edit always runs.
    • The Preview chip appears only for a server this conversation started on your computer — not for websites the agent looked at, or addresses mentioned in an answer.
    • A new switch, off by default: Settings ▸ Features ▸ Agent ▸ “Run the agent on Mel’s server (beta)”. With it on, Mel’s server decides each step of an agent turn while your computer still runs every tool, behind the same approval cards. If the server can’t run a turn, the turn runs on your computer and a notice says why. The `MEL_SERVER_HARNESS` variable overrides the switch.
    • Answering a question the agent asked you no longer switches off, for the rest of that turn, the record Mel keeps of the agent’s work — broken since v0.1.37. That record is how Mel notices a file changed underneath the agent and how the turn’s cost is counted, so both work again after you answer.
    • When a turn stops because of a problem — Mel’s server could not be reached, your sign-in had expired, you were out of credits or sending too many requests — Mel now records which problem it was and the work done before it, instead of recording the turn as abandoned with nothing done. A server error in the middle of a turn is shown as a failure, not a warning.
    • Three models that could never fit a Mel conversation are gone from the picker: GPT-3.5 Turbo, Qwen Max and ALLaM 2 7B. Their windows are smaller than Mel’s instructions and tools plus room for an answer, so picking one ended every turn with “too small”.
    • The agent’s folder listings come back in alphabetical order on every computer, so the same folder reads the same everywhere.
    • When the agent looks up a tool it doesn’t normally carry, the tool is there on your next message, as the agent tells you. Before, it was promised and never arrived.
    • The Feedback link opens openmel.dev/contact.
  27. v0.1.45September 2026
    • The agent can no longer tell you it did something it did not do. A turn that says “Done — I updated the file” without having touched anything is refused, as is one that promises to act and then just talks; after repeated promises with no action the turn ends with a visible note saying nothing was changed, instead of finishing quietly as though it had worked.
    • A long job runs to the end by itself. Work used to stop partway with a “tool-call budget reached — press Keep going” message; the same boundary is still there for housekeeping, but Mel crosses it on its own and says nothing, and Stop still works at any moment.
    • Answering the agent with “yes” no longer runs the Unix `yes` command and fills the screen with the letter y. Short replies — yes, no, ok, sure, continue — are read as answers; put `!` in front to force a shell command.
    • A command that prints endlessly can no longer use up all your memory and take Mel down with it. Output waiting to be drawn is now bounded, the command is never slowed down, and if anything is dropped Mel says how much.
    • Asking to read a file gives the whole file. Longer files were being cut in the middle as they arrived, so the agent kept re-reading the same file and getting the same cut.
    • The agent knows where your conversations are kept and how to reopen them, instead of saying no transcript store exists.
    • A `run_command` written across several lines no longer paints over the conversation above and below it, and a heredoc whose end marker never closes is refused before it can write itself into your file.
    • The share icon in the conversation bar opens Export — it was drawn but did nothing. A wide table's scrollbar no longer covers its last row, and the working-directory chip shortens instead of running under the context gauge.
    • Plain-English build requests reach a model that can build them: “I want something that reads the emails and writes the reply by itself” was being answered by the smallest, cheapest model. A model that stalls, goes silent or answers nothing is set aside for a while instead of keeping the rest of a long turn.
    • Claude 5 Sonnet, when picked by hand, no longer spends its whole answer thinking and returns nothing.
    • If Mel stops being able to save a conversation because the signed-in account changed, it now says so instead of staying quiet.
  28. v0.1.44September 2026
    • Eight connectors, each with YOUR OWN key, under Settings ▸ Connectors: GitHub (repositories, branches, commits, files, issues, pull requests incl. diff / create / merge, Actions runs, releases, code and issue search — 25 tools), Vercel (projects, deployments, build logs, redeploy / promote / cancel, environment variables with values never read back, domains, aliases — 19 tools), Alpha Vantage (quotes, price history, fundamentals, news sentiment, forex, crypto, indicators, economic and commodity series — 16 tools), Stripe (balance, customers, charges, payment intents, subscriptions, invoices, products and prices, payment links, refunds, payouts, disputes, events — 24 tools), Shopify (your store's products, orders, customers and collections — 10 tools; paste the Dev Dashboard app's Client ID and Secret with the store domain and Mel mints the token), Supabase (Supabase's own MCP tools — SQL, tables, migrations, logs, edge functions, branches — with your personal access token), Zapier (whatever actions your Zapier MCP server enables; the list is read live) and ElevenLabs. Your key is stored on this computer and sent only for calls you approve; a connector never runs on Mel's account.
    • Connector results are drawn as cards: repository and issue lists with state pills, pull requests with their branches and diff, Vercel deployments with build state and logs, Alpha Vantage quotes with a sparkline for every price series, Stripe balances, customers and prices, Shopify products and orders — every amount with its currency stated. Every link in a card opens in your browser.
    • A connector's tool list is no longer cut in the middle when it is long. With 24 tools the model was shown about half the list and told users that tools which existed did not; the whole list now reaches it, and a list too long to show whole says how many tools were left out.
    • Fourteen models are withdrawn from the picker for now: the eight most expensive OpenAI tiers (o1-pro, the GPT-5 Pro line, GPT-4 and GPT-4 Turbo, o1) and the Claude Opus 4.x line, Claude Opus 5 and GPT-5.6 Sol. A single short prompt on the dearest of them could bill the equivalent of a whole $10 pack in one reply; the worst case is now a twentieth of that. They come back when the per-reply cap is priced properly.
    • Every connector call still shows an approval card first — Mel always asks before a connector is used.
  29. v0.1.43September 2026
    • The agent conversation uses the width you give it: the transcript column is centred and grows to 1280 pt (was a left-anchored 900), and the answer's prose has no fixed measure of its own any more.
    • Answers are set in the proportional face; code blocks and inline code stay monospace, so prose and code read differently again.
    • One card chrome everywhere — rounded 10 pt, hairline, soft shadow — for tool runs, shell output, diffs, tasks, approvals, questions, PDFs and audio. Disclosure chevrons rotate instead of swapping glyphs; card headers light up on hover; a running tool row spins.
    • The approval and question cards are neutral cards with an amber bar on the left (the approval's breathes while it waits), the command sits in a code well, and the buttons are a filled primary and a ghost secondary.
    • Your own turn is a soft accent-tinted bubble; new items fade in; three pulsing dots show Mel waiting on the model before the first token.
    • The composer is a rounded card whose outline warms to blue while it has focus, with a compact chip row, a real model chip carrying the maker's mark, and painted attach / dictate / send buttons.
    • New sessions open in the Mel agent by default (Settings ▸ Features turns it off; Esc still drops to the terminal).
    • Tool, shell, diff and task cards — and the thinking card — settle and fold about 1.5 s after they finish; click any header to expand it again. Answers stream in smoothly instead of appending in jumps.
    • Real maker marks in the model picker for Claude, GPT, Gemini, DeepSeek, Llama, Kimi, Qwen, GLM, Grok, MiniMax and NVIDIA; painted like / dislike / copy on every answer; a new Melling… indicator with a turning spark instead of a moving line.
    • The question card, the usage panel (tokens · cache hit · context · time as tiles), the Tools ▸ Preview panel and plan mode were redesigned and tested against live turns.
    • One chrome for every menu, list, panel and card: the + new-session menu, the account menu, notifications, the conversation's ⋮ menu, the Tools panel (explorer, conversations, search), agent runs, cloud jobs, and the export and continue-on-your-phone cards. Escape on a card now closes only the card.
    • Settings rebuilt: a rail with a glyph for every page in three groups, grouped rows on every page, the content column spans the whole window, theme as a Dark / Light / System control, the nine most-used shortcuts as key chips, a Privacy page that lists what stays on your machine, and an About page with Check for updates.
    • Dark theme: the neutral surfaces moved from pure grey to a blue-tinted neutral of the same luminance. Light theme unchanged.
  30. v0.1.42September 2026
    • Long tasks no longer die with “the relay stream ended without a result (connection dropped)” about 30 seconds into a step where the model is quiet — the relay now keeps the connection alive while it waits on the provider, and the edge timeouts in front of it were raised. This was a server-side fix that reached every installed version on 18 September.
    • The per-tier output cap introduced for shorter answers is gone: at 1024 tokens it also cut off every file the agent tried to write on a normal coding turn (“cut off at Mel's 1024-token output limit”). One 16,384-token ceiling again; answer length is handled by the prompt.
    • When a long build turn reaches its round budget with changes it never verified, the agent now gets one round to run a build, typecheck or test before summarizing — instead of ending with “run these commands yourself”.
    • The agent researches or crawls first, writes a short numbered plan and confirms it once, then works through it without stopping when asked to build or clone something substantial.
    • macOS: the download is now a signed, notarized .dmg — open it and drag Mel to Applications. The .zip is still published for the in-app updater and as a portable download.
    • Drag a file anywhere over Mel (Windows, macOS, Linux/X11) and the whole window dims with an Add anything drop target — images, PDFs, and other attachable files go into the open chat; otherwise the path lands in the prompt you can see. (Pure Wayland still has no OS file-drop events from the window toolkit — use + or paste there.)
    • Paste a screenshot with Ctrl+V (Cmd+V on macOS) in agent chat attaches the image on Windows, macOS, and Linux (X11 and Wayland) instead of Paste failed — the clipboard could not be read (supported mime-type is not found).
    • Drag-select in agent chat keeps scrolling when the cursor hits the top or bottom of the transcript — selection is no longer limited to what is on screen (Windows, macOS, Linux).
    • Vercel, GitHub, Netlify, Railway, Fly, Render, and other common deploy hosts open on click without the “open this link?” toast (unknown sites still ask).
    • When Mel finishes create_pdf, the PDF opens in your system viewer right away — no extra click on the path.
    • When a localhost / loopback URL appears in the chat, Mel opens Tools ▸ Preview and shows a Preview chip with the host so the running app is one click away.
    • The latest tool-run group in agent chat stays expanded after the tools finish (older groups still collapse).
    • Shorter agent answers on simple questions — system prompt length rules plus tiered max completion tokens (Trivial 512 → HugeContext 16k).
    • mel -p /path and mel --path /path open Mel in that directory (same as the positional mel /path).
  31. v0.1.41September 2026
    • DeepSeek Fast slot is now deepseek v4.1 flash (live id deepseek-flash, native vision). Auto routing bias unchanged.
    • ElevenLabs Mel Connect TTS defaults to a free-tier-safe premade voice (Roger); library voices still work on paid plans when you pass voice_id.
    • Hosted relay and download CDN cut over to the new AWS account edge (CloudFront → US East).
  32. v0.1.40September 2026
    • Tab accepts ghost suggestions in the terminal and agent prompt; the code editor prefers LSP completion (local identifiers as fallback).
    • Tools ▸ Preview shows the last localhost URL and browser screenshot beside the chat.
    • For the small test group on cloud runs (coming soon for everyone else): ☁ in the agent footer runs a Mel cloud task, and the titlebar Jobs chip lists open cloud jobs.
    • Share a conversation as a public read-only link (secrets scrubbed; 30-day expiry).
    • Settings ▸ Account ▸ Cloud triggers: mint meltrg_ tokens and fire POST /v1/triggers/run from GitHub Actions / cron.
    • semantic_search tool — local TF-IDF over the project (no paid API).
    • First-run landing shows six starter prompts when you have no recent conversations.
    • Mel Connect MCP gateways: GitHub, Slack, Supabase, Stripe, Shopify, ElevenLabs, Zapier, Vercel, Alpha Vantage (pass-through; publish still ops).
  33. v0.1.39September 2026
    • Continue on my phone… opens and stays open — the click that opened the card no longer dismisses it on the same frame (Export had the same race).
    • The GitHub App install button from a failed handoff goes to a working install page (the old permissions URL 404'd).
    • Downloads go through CloudFront (global edge) instead of only Mumbai S3 — faster worldwide pulls; the updater follows.
    • Mel picks the closer of the India and US East relays at startup (health race); MEL_RELAY_URL still overrides.
    • Relay: session lookup is memoised for five minutes per process so consecutive legs of one turn stop re-paying the same DynamoDB GetItem; sign-out clears the memo.
  34. v0.1.38September 2026
    • Long agent tasks keep going across Mel's ~90-second host legs — no fake "continued 0 times and reached its limit" stop, and no ask to keep going. The cut stays infra-only; Mel auto-continues the same turn (safety cap still 12).
    • When all the models in the failover chain are unreachable, Mel drops the sticky lead for that turn so the next message can pick a live provider instead of re-hitting a dead one.
    • The "built, not verified" completion nudge fires at most once per turn, and a successful file read-back counts as verification — so Mel stops circling on the same downgrade notice.
  35. v0.1.37September 2026
    • Mel stops re-reading the same file in circles. A repeated read now returns the bytes it asked for — it used to answer with a placeholder that said "re-read if needed", which is what sent the agent round the loop testers reported (up to 124 reads of one file, and no edit, in a single turn).
    • Long sessions keep their thread. Mel's repeat guard, its completion check and its empty-answer retry are now ON for everyone — a turn that says it finished without checking is told so and given one more step, a tool that has failed three times the same way is stopped instead of retried ten more, and an answer that comes back empty is named and retried once.
    • Mel stops saying "Task completed." when it did not. A turn that failed, ran out of credits, hit a rate limit, needed a sign-in, outgrew the model's context or simply died now says which of those it was — in the notification and in the transcript.
    • The agent's tools got a memory and a voice. A shell that remembers your cd between commands; reads that tell you which lines you got and how to ask for the rest; searches that say "found 250 of 3110" instead of quietly cutting the list.
    • Long conversations keep what you agreed. Mel decides once whether the chat is too big, measured against the window of the model that is actually answering — so a million-token model is no longer compacted as if it had 200k — and what survives is a checkpoint under fixed headings (decisions, plan, files) instead of quietly dropping the important bits first. Pinned models keep this relief too; auto before the first answer does not invent a window.
    • Something you tell Mel once stays true. A rule you state — use main, don't touch app/ — is offered on a chip in your own words; keep it and Mel checks it before file changes and refuses ones that would break it, quoting you back. The task checklist stops going dark and duplicating; Mel asks real questions with options and deadlines, and will not re-ask something you already answered. Pressing ▶▶ only skips approvals, not every question.
    • Mel's instructions to the model stop being one ever-growing block on every message. Rules now travel with the tools in use, so a one-line question no longer pays for browser advice Mel is not using — and a cut-off answer no longer invents a dead continue tool.
    • Long auto conversations no longer forget early: the context gauge and the compaction trigger now measure against the window of the model actually serving the conversation. Before, every auto conversation was measured against a flat 200k — on a 1M-window model that summarised-and-forgot at 16% of the window.
    • auto's failover chain is generated from the model catalog instead of hand-ranked lists — cheap and open-source providers lead by construction; when a provider's account runs dry Mel steps that whole provider out for ten minutes instead of trying its next model on the same empty wallet.
    • The Fable and Astra models have been removed from the model list. If you had pinned one, new conversations start on auto and Settings says which model was retired — instead of a chip naming a model that no longer exists and turns failing with "unknown model".
    • Mel keeps working until the credits you can see actually reach zero. If you had add-on credits while your monthly ones were spent, Mel was refusing turns against a hidden negative number instead of the two numbers on your Billing page.
  36. v0.1.36September 2026
    • Connectors. Settings ▸ Connectors lists the partner tools Mel Connect offers — ElevenLabs first. Press Connect, sign in with your own account in your browser, and its tools appear to Mel behind the same permission cards as everything else. Your sign-in stays on this computer; disconnect any time.
    • A connector is asked about every time it is used. The card names the company and the exact tool — “ElevenLabs wants to: text_to_speech” — and says plainly that this sends your request to another company. There is no “always allow” for a connector, and a tick you gave a shell command earlier does not cover one.
    • Run your own tool server. Custom server… at the bottom of the Connectors page points Mel at a program on your own machine. Values you enter that look like keys go to your keychain, not into a settings file.
    • The model picker now lists every chat model the ten providers serve (143 ids, up from 61) with the providers’ true context windows — nine were too small (DeepSeek V4 is 1M, not 128k) — and the three ids no provider serves any more (Groq’s two Llama ids, Kimi K2.5) are gone. Picking a legacy id still works; auto only ever picks a model with a verified price and window.
    • The picker also lists NVIDIA’s NIM catalog (55 models: Nemotron 3.5 Lightning, Nemotron 3 Super / Ultra / Nano Omni and every other chat model the endpoint serves — Kimi K3, DeepSeek V4, MiniMax M3, Gemma, Llama, Mistral, Phi, Granite and more, marked (nvidia)). They can be pinned; auto never picks them yet — until Mel’s NVIDIA key and terms are in place a pinned NVIDIA model falls over to the usual chain with a visible “why this model” note.
    • auto reads the task instead of making one fused guess — what kind of work it is, how hard, how many tool rounds it will take, how big it is. A long conversation no longer makes a short question expensive: that one mistake was 75 of the 164 misroutes we measured. Mel’s own “keep going” continues on the previous turn’s terms, and a turn with an image in it is routed once, onto a model that can see it.
    • auto now tells every provider whether to think. A quick question or a small edit gets a no-thinking answer on each model that can be told to stop — DeepSeek, GLM, Kimi K2.6, MiniMax M3, Qwen, Gemini 2.5, the GPT-5.x line, Claude — while a build or a hard problem gets a small or medium thinking budget. Your own reasoning-level pick always wins. MiniMax models no longer show their reasoning inside the answer.
    • Bring-your-own-key users can now set an NVIDIA key.
  37. v0.1.35September 2026
    • Approve from your phone. Link a Telegram chat in Settings ▸ Account and every permission request Mel raises also reaches your phone — a 200-character masked summary, never the full command or a file. Tap Approve or Deny; the first answer wins, the app says “Approved from your phone”, and there is no “always allow” from a phone by design. Off until you link it.
    • A reply from your phone becomes the next thing Mel does. Reply to one of Mel’s messages with a sentence — “use the staging database instead” — and when the current turn finishes, the app picks it up, shows it as “From your phone” and runs it. A message that is not a reply gets a short note back explaining that, rather than disappearing.
    • Mel can email you when a long turn finishes. Off by default; turn it on in Settings ▸ Account. A turn that ran for more than two minutes, or one you sent to the background, ends with an email containing the closing message and what the turn cost. Results only — Mel never asks you to approve anything by email, and each email threads onto the previous one for the same conversation.
    • Continue on my phone, for the small test group on cloud runs (coming soon for everyone else). In a conversation, choose Continue on my phone, pick how long (30 minutes, 2 hours, or until a time you name, up to 8 hours) and see the price before you agree to it. Then close the laptop. Mel in the cloud picks that same conversation up, with its whole history, on a fresh clone of your repository, and you keep it going from Telegram: send it a message and it becomes the next thing Mel does, questions and permission requests come to your phone, and the answer comes back there. Send stop, let the time run out, or press Take back on the laptop, and the conversation comes home with everything that happened while you were away. Cloud time costs 6 credits an hour, billed by the minute, on top of the usual model credits. Paid plans only, because the API keys Mel uses on this computer never leave it. It is not your laptop still running with the lid shut, it is a copy of your repository running on our machines.
    • Export a conversation. ⋮ ▸ Export conversation writes one .zip with everything Mel saw and did in that conversation — prompts, replies, thoughts, every tool call and its full output, shell commands, diffs, approvals and who answered them — plus a per-turn table of what each turn cost, which model answered it and why it was chosen, and a README. You can tick two options before it writes: mask secret-looking strings, and replace absolute paths and your home folder; both are off by default and the README counts what they changed. It stays on your machine — nothing is uploaded, and sharing it is your choice. If Mel circled or forgot, this is the file to send us.
    • Erasing your account now also unlinks your phone. Before this, “erase everything” left a linked Telegram chat behind that could still answer approval requests. It is part of the erase now.
    • A rate limit is no longer reported as running out of credits. When the service sheds load, Mel used to tell you your plan was exhausted — two different problems with two different fixes, and only one of them was yours. It now says which one happened.
    • A crash report names what it matched. The support log records the classification of a panic alongside it, so a report you send is diagnosable without a reproduction.
  38. v0.1.34August 2026
    • Mel finishes the job instead of stopping to ask. A build task — clone this site, scaffold this project — starts by reading: the page, the bundle, the schema. All of that is read-only, and Mel was budgeting it as if it were a research question, so it ran out of room mid-investigation, before it had written a single file, and asked you to press “keep going”. Plan mode made it certain: nothing can change while a plan is being written, so every planning turn was scored as pure research. Mel now recognises a build request for what it is and gets the room to finish, while a turn that has genuinely stopped making progress wraps up sooner than before.
    • Mel can see the pages it builds. Its headless browser could not render WebGL, so every map, chart, 3D scene and canvas app came back as a black rectangle — and Chrome reported that as a successful screenshot. Mel would check its own work, see nothing, and fall back to reading the page's text. It now renders those pages properly, and a screenshot that comes back blank is reported as a failed render rather than passed off as evidence.
    • The rating buttons appear once per answer, not after every step. A long build used to leave a row of thumbs after each intermediate line — “let me check the logs”, “let me verify it renders” — rating nothing that had finished.
    • A link Mel gives you is clickable even when it writes it as code. The address of the thing it just built for you — the local server it started — used to render as plain text you had to select and paste. A code span containing a command is still left alone.
    • A message you send while Mel is still working is now queued instead of thrown away. It used to disappear the moment you pressed enter — the text was taken out of the box and dropped, with nothing on screen to say so — while the placeholder promised you could steer the running agent. Your follow-up now waits, shows above the input, sends itself when the turn finishes, and comes back to the box if the turn ends any other way. It is never sent into a turn that failed.
    • The agent stops forgetting what you already decided. In a long session, when Mel compacts older turns to fit the model's context window, it now carries the task checklist, the questions you already answered, the approvals you already gave, and its own rolling summary. Before, all four were silently dropped — which is why a long build would re-ask a settled question and start re-planning work that was already underway.
    • Clicking “Thought for 2 seconds” actually opens the thinking again. It looked like a button, it listened for clicks, and it never received one, so once a turn ended the model's reasoning became unreachable — even though it was saved on disk the whole time.
    • File paths the agent writes are now links. When Mel tells you a report is saved at a path, you can click it. Paths that would run something — scripts, apps, installers — are deliberately not clickable, and a link whose text does not show where it goes is refused.
    • PDFs got the rest of the way. Asking for one no longer spends an extra round-trip discovering that Mel can make PDFs — it starts straight away — and the card now offers Download, which saves a copy into your Downloads folder without ever overwriting a file already there. You can also read the document beside the conversation instead of opening it in a separate app.
    • Drag files from Finder or Explorer onto Mel to attach them. They go through exactly the same size limits and image handling as the + button, and anything Mel cannot attach lands in the prompt as a path instead of vanishing.
    • A turn that stops at its tool-call budget now says so and offers to keep going, instead of suggesting you commit an unfinished build. Running low on credits no longer tells the agent to cut its work short — it prefers a cheaper model instead of leaving the job half done.
    • The new-session menu names what it starts — Agent and Terminal — and the ⌘T badge now sits on the row that key actually opens. If you had starred a session config as your default, the badge was on the wrong row.
  39. v0.1.33August 2026
    • Sending an image no longer closes Mel. On Windows machines with Snapdragon processors, attaching any picture and pressing send killed the app instantly — every image, every time, while PDFs were fine. Mel was releasing the thumbnail's graphics memory in the middle of the same frame that was still drawing it, which the graphics driver treats as fatal and which no error handler can catch. Nothing was wrong with the pictures: the file size never mattered, because what reaches the graphics card is a 96-pixel thumbnail.
    • Three more ways Mel could close on you, all fixed. Closing a pane or a session that contained images, and changing or removing your profile photo, both released graphics memory the same unsafe way. Separately, searching a project whose files contain certain letters — Turkish capital İ, the Kelvin and Angstrom signs, capital sharp S — crashed the moment results were drawn, on every platform, because the highlighter measured its offsets against a lowercased copy of the line and then applied them to the original.
    • Long tasks finish instead of stopping silently. Any single step that ran past about two minutes was being cut off by the server's request limit — the reply was lost, no file was written, and the agent often looked like it had succeeded. Requests are now handed over cleanly before that limit and continue where they left off, so a long build or a large file write completes. This one is already live for everyone; it needed no update.
    • The agent's thinking panel no longer collapses into a one-character-wide strip down the right edge of the screen. Past the first few thousand characters of a long thought, every later block was laid out in whatever pixels were left over on a row that had already been filled — so it rendered one letter per line while the panel beside it sat empty. It stacks properly now and reads at full width.
    • Team runs tell the truth. An agent in the team whose turn died used to report "IDLE (done)" with its last half-finished sentence, so the agent leading the work believed a file had been written when none had; it now reads FAILED, with the reason, and that agent's own card says so too. A message sent to an agent that is still busy is no longer confirmed as delivered when it was in fact dropped.
    • Reopening a team session keeps its agents. Closing Mel used to scatter them into unconnected sessions and quietly erase the record of which belonged together; the group is now kept, each agent links to its own conversation, and the ones that are no longer running say so plainly rather than pretending.
    • Crash reports name what they matched. When Mel recovers from a graphics-driver crash it now records which fault it recognised and where, instead of only that it restarted. No prompt text, credentials or file contents are ever written to that log.
  40. v0.1.32August 2026
    • Long sessions no longer hit a wall. A malformed reply from one AI provider could get saved into a conversation, after which each model refused it forever with a misleading 'provider outage' message — while new sessions worked. The server now repairs such conversations automatically on every request (existing stuck conversations start working again with no action needed), the malformed shape can no longer be saved in the first place, and the error message for genuinely unserveable requests now tells the truth.
    • Conversations that outgrow a model's context window now compact automatically and continue — a quiet one-line notice instead of an error. Very long sessions are the point of Mel; now the seams show it.
  41. v0.1.31August 2026
    • Recording you can see. While dictating, the mic becomes a stop square with a live input-level meter and an elapsed timer, and transcription shows its own distinct state — no more guessing whether Mel heard you. A muted microphone now shows a flatlined meter while you speak instead of failing silently at the end.
    • A graphics-driver crash no longer closes Mel. If the renderer dies mid-session, Mel writes a crash record, restarts itself in compatibility mode, and tells you what happened — instead of vanishing with your session.
    • Changing plans is now a switch, not a second subscription. Upgrading (or stepping down) opens a Stripe confirmation that modifies the subscription you already have, with proration handled by Stripe. Subscribed users see 'Switch to…' in Billing, and moving from Max to Business is labeled as the step down it is.
    • Beta updates channel. Settings ▸ Features ▸ 'Receive beta updates' opts this install into pre-release builds; everyone else stays on stable. This release is the first to ship through the new channel.
    • 16-bit PNG attachments (like the Mel logo) are normalized before upload so every vision provider accepts them.
  42. v0.1.30August 2026
    • Windows on ARM: Mel now draws with Direct3D 12 on ARM machines instead of OpenGL. On a Snapdragon X laptop the OpenGL path repeatedly crashed the whole operating system under agent load; the same workload ran for eighteen minutes clean on Direct3D 12. The trade-off is occasional visual corruption under heavy load that repairs itself — an OS crash loses your work, a cosmetic glitch loses nothing.
    • If Direct3D 12 misbehaves on your machine, starting Mel with MEL_RENDERER=glow forces the previous renderer.
  43. v0.1.29August 2026
    • Windows: the Direct3D 12 renderer was never actually compiled into any shipped Windows build. Mel asked the graphics library for a backend that was not in the binary, got an empty set back, and could not create a window — this, not a driver, is why v0.1.27 would not open on Windows at all. It is compiled in now, and a Windows CI gate fails the build if it ever falls out again.
    • A start-up failure now leads with the fix. The dialog opens with the recovery steps (start Mel with MEL_RENDERER=glow) rather than the diagnosis, and says plainly that reinstalling will not help — downloads are replaced in place, so a reinstall fetches the same build.
    • The support log records what happened to the fallback attempt, not only the first failure. Reading the old log needed the customer's own account of it to interpret.
    • Publishing a release no longer depends on GitHub. A failed GitHub-Release step used to skip publishing the download manifest, which left the download service holding new files under an old manifest.
  44. v0.1.28August 2026
    • OpenGL is the default renderer again on every machine. v0.1.27 switched ARM Windows machines to a renderer that had never once been started on the hardware it was chosen for, and it did not open there.
    • A renderer that fails to start is no longer the end of the session. Mel retries once on OpenGL, and if that fails too it shows a native dialog naming the likely cause and where to find the support log. On Windows a release build has no console, so the previous behaviour was to exit silently with no window and no message at all.
  45. v0.1.27August 2026
    • This release does not open on Windows — any Windows machine, ARM or Intel. It is superseded by v0.1.28 and v0.1.29; if you are on it, update. The cause and the guard against it are in the v0.1.29 notes above.
    • Long agent conversations stop getting slower as they grow. The transcript now lays out only what is on screen, using measured heights rather than estimates, so a conversation that has run all afternoon scrolls like a fresh one. On a 1.4 MB conversation the transcript's share of a frame fell from 8.37 ms to 0.40 ms and stopped growing.
    • The Thoughts panel no longer re-wraps its whole text about thirty times a second. A blinking cursor was rebuilding the entire reasoning body on every frame; it is now split into stable chunks, so watching the agent think costs what reading it costs.
    • Long command output says when it has been truncated instead of silently stopping at 64,000 bytes, and the agent's own long replies get the same visible marker.
    • Ending a turn no longer saves the conversation twice. It serialised the whole conversation two times in the frame that finished a turn.
    • Replies are bounded. On one afternoon a single account produced two agent replies of 65,536 tokens each, taking about twenty-one minutes apiece, against a normal median of 159 tokens in under five seconds — that runaway is what grew the conversation that made the machine struggle. Completions are now capped server-side, so this applies to every version.
    • A support log you can send us. Off-by-default logging did not help anyone who had already hit a problem, so Mel now keeps a rotating 256 KB log of its own events — starts, renderer choices, update checks, failures. It never contains your messages, prompts or command output, and it is deleted along with everything else when you erase your data.
    • About names the machine Mel is really running on, including when an Intel build is running under emulation on an ARM Windows machine. Support spent two days not knowing that.
    • The agent can read big pages again. Taking a snapshot of a page larger than about 64 KB used to hang for thirty seconds and come back empty; one real page was 1.18 MB. Screenshots also describe the picture rather than the request that made it.
    • There is no v0.1.26: its Windows build failed to compile and it was never published.
  46. v0.1.25August 2026
    • Security: the library Mel uses to open links in your browser is updated for RUSTSEC-2026-0257, where a crafted BROWSER environment variable can inject arguments into the browser being launched. Mel deliberately loads your login shell's environment, which is where BROWSER comes from, so the path was reachable. v0.1.23 and v0.1.24 are affected equally — the advisory was published against a dependency that had not changed — so this is the release to be on.
  47. v0.1.24August 2026
    • The Free plan moves to bring-your-own-key. Mel's AI agent on Free now runs on a provider key you add in Settings → AI: you pay your provider directly, at their price, with no Mel credits and no Mel markup. Groq and Gemini both offer free key tiers. Pro, Max and Business are unchanged and still need no key of your own.
    • If you already have a Free account, nothing changes for you — accounts created before this release keep their monthly hosted-AI credits.
    • Getting to your keys takes one step. Provider keys have their own entry in the avatar menu, instead of being the last section of the fifth settings page.
    • A rejected key now says so. If your provider key is expired or revoked, Mel tells you the key was rejected and where to fix it — it used to report a temporary provider outage and suggest waiting a minute, which never fixed it.
    • A key you supply can no longer affect anyone else. A failure on your own key is no longer counted against Mel's view of that provider's health.
    • Clearer limits. Free runs on your own provider key. Fair-use ceilings (now 5,000 agent turns a day) apply on every plan, including with your own key. Bring-your-own-key means no Mel credits, not no limits.
  48. v0.1.21August 2026
    • Two accounts on one computer no longer share one history. Conversations are stored per account, so signing in as someone else shows you their conversations and not the previous account's — your existing conversations are moved into place once, automatically.
    • Every action answers. A whole class of buttons used to finish in silence — "Check for updates" ran a real network check and changed nothing on screen, and several messages Mel wrote had no place to appear at all. Actions now confirm, and the ones that fail say so.
    • Menus and pickers stop opening in the wrong place. Nine dropdowns were positioned from a height each caller worked out by hand, and four of them got it wrong, so lists could open clipped or off the screen.
    • The agent can supervise real work, not just write it: start a long build or a dev server in the background, wait for it to be ready, call it over HTTP, open the page in a headless browser to see what actually rendered, and look at images. It can also plan first and show you the plan before touching anything.
    • A conversation that had become unusable can be used again — one bad message could make every following turn fail, and Mel now repairs the stored history instead of failing on it.
    • When a model has a bad minute, the conversation no longer stays stuck on the stand-in that covered for it; the next turn goes back to the best available model.
    • Slow frames are now measured rather than guessed at, and the operations behind the worst freezes are barred from the code that draws the window.
    • If you're on v0.1.18 or later, this update is one click: avatar menu → Update and relaunch Mel.
  49. v0.1.20July 2026
    • macOS gets a real application menu — Mel, File, Edit, View, Find, Window, Help — with the shortcuts you expect (⌘, for Settings, ⌘N, ⌘W). Accelerators are deliberately held back while a CLI coding agent has the keyboard, so ⌘P and ⌘D can't be stolen out from under it.
    • Window controls behave: on macOS Mel no longer paints its own minimize/maximize/close over the ones the system already draws — the traffic lights were always the real controls, and in fullscreen Mel's copies did nothing at all. Minimize, Zoom and Toggle Full Screen now live in the Window menu.
    • The macOS app icon is sized to Apple's grid, so it matches its neighbours in the Dock instead of towering over them.
    • Ticks, crosses and close buttons render on macOS. They were empty boxes — the system font Mel relied on skips that block of symbols, so Mel now ships its own.
    • The keyboard hints above the prompt are clickable, not just decorative — and two of them that advertised shortcuts Mel didn't have are gone.
    • The working-directory chip opens a directory picker, in the terminal and in an agent conversation: jump to a parent or subfolder without typing cd.
    • Every titlebar button now tells one story — the icon, the tooltip and what actually happens finally agree.
    • The starter session templates no longer depend on a third-party CLI being installed; they open Mel's own agent. The one that asked for a repository now explains itself and offers a way to add one instead of dead-ending.
    • PDF export works on a stock machine. If no browser-based renderer is installed, Mel now falls back to one it ships with, so "create a PDF" no longer fails with a missing-backend error.
    • Smaller repairs: the last session's close button no longer offers to do something you can't see; a failed command is visible in the sidebar and one keystroke from being handed to the agent; the sidebar's View-as and Density settings do something you can actually notice.
    • Not changed: the in-app Feedback link still opens the GitHub repository. Where it should point is still being decided, so it was deliberately left alone rather than guessed at.
    • If you're on v0.1.18 or later, this update is one click: avatar menu → Update and relaunch Mel.
  50. v0.1.19July 2026
    • A new look: Mel's brand is now a true blue, with pink marking everything the agent does — the /agent label, the agent sparkles, and the running "Melling…" line. Every color pairing is contrast-checked so the app stays crisp and legible.
    • Bigger, more comfortable text by default — and a new Settings → Appearance → Font size control (13–20pt) so you can tune the whole app, terminal output included, to your eyes.
    • A cleaner app icon: the same M mark the website uses, everywhere.
    • If you're on v0.1.18, this update is one click: avatar menu → Update and relaunch Mel.
  51. v0.1.18July 2026
    • One-click updates, for real: earlier builds could never see a new release (their update check pointed at a source that stopped answering) — Mel now checks a public release manifest, so the update prompt actually appears when a new version ships.
    • One-time step: if you're on v0.1.17 or earlier, download this version once from the download page — those builds can't update themselves to it. From v0.1.18 on, updating is one click inside Mel.
    • macOS: after an update Mel now closes and reopens by itself (it used to quit and wait for you to find it in the Dock), and updates survive the move across disk volumes.
    • Windows: updating no longer opens the full installer wizard — Mel closes, updates silently, and relaunches on its own.
    • A small dot on the avatar now signals when an update is ready, and the account menu gained "Check for updates" — a failed check says so honestly instead of pretending you're up to date.
  52. v0.1.17July 2026
    • macOS builds are now signed with an Apple Developer ID and notarized by Apple — Mel opens on a normal double-click with no Gatekeeper "Open Anyway" step, and permission grants like Full Disk Access now persist across updates.
    • With Windows already code-signed, this is the first Mel release signed on both desktop platforms.
  53. v0.1.13July 2026
    • Fixed upgrading from Windows: checkout links opened without their full address, landing on an error page — the browser now receives the complete link on every OS.
    • Windows builds are now code-signed (Concepts Dzign LLC).
  54. v0.1.12July 2026
    • Your account now works on the web too — sign up or sign in at openmel.dev, see your plan, credits and usage at /account, and upgrade from the browser.
    • Sign in with browser: the app can sign in via a short code you confirm on openmel.dev — no password typed in the app.
    • New Profile and Security settings pages — edit your name, change your password, see active sessions and login history, and sign out other devices.
    • Plans in the app now show full feature lists with a monthly/annual toggle (annual saves 10%), matching the website.
    • Real in-app auto-update: Mel downloads, verifies (SHA-256) and applies updates itself, then relaunches.
    • A contact form at openmel.dev/contact, welcome emails on signup, and stronger rate limiting across sign-in and signup.
  55. v0.1.11July 2026
    • Fixed a rendering bug where a long agent conversation could go blank when scrolled — the full response now stays visible from top to bottom at any window size.
  56. v0.1.10July 2026
    • Cloud history sync — your agent conversations now sync to your account, so reinstalling Mel or moving to another machine restores your full history instead of starting over.
    • Multiple agent chats at once — open several agent conversations side by side in a new Agents sidebar; opening one from history adds it as its own chat instead of replacing the one you're in.
    • Agents finish the job — ask the agent to build something and it now runs the whole task to completion instead of stopping every few steps to ask permission. A one-key “Keep going” and a hands-off mode make long builds effortless.
    • Readable long conversations — very long agent transcripts stay within the window; wide terminal output, file paths, and reasoning are contained instead of being pushed off the side of the screen.
    • Running-turn controls — the buttons shown while the agent is working now have tooltips and actions: fast-forward the rest of a turn hands-off, or send it to the background and get notified when it's done.
    • Windows fixes — agent commands run correctly on Windows; closing a session or tab works on the first click; a not-installed command exits cleanly instead of hanging.
  57. v0.1.9July 2026
    • Windows installer — a proper per-user setup.exe with desktop and Start-Menu shortcuts and a clean uninstall from Settings → Apps (the portable .zip is still available).
    • Readable agent responses everywhere — fixed invisible text and washed-out code cards in the agent transcript on light-themed systems; code blocks are now dark, horizontally scrollable, and copyable.
    • Sign-in reliability — fixed a sign-in re-prompt on Windows so your session persists across restarts.
  58. v0.1.8July 2026
    • Credit plans — every plan now includes a monthly bucket of AI credits (1 credit = a cent of real AI cost): Free 150, Pro 1,500, Max 18,000, Business 1,500 per seat. Credits work across the hosted models; bring your own key on any plan and a turn costs 0 Mel credits.
    • Two new plans — Max ($200/mo) for power users who live in the agent, and Business ($50/seat/mo) with per-seat credits and priority support.
    • Add-on credit packs — 400/1k/3k/6.5k packs that roll over month to month and last a year; optional auto-reload with a monthly spend cap you control.
    • Self-serve billing — upgrade, buy credits, and manage your subscription from Settings → Billing; a new Usage tab shows your per-day credit history.
    • Self-serve accounts — the waitlist is gone. Sign up for a free Mel account instantly. The website now leads with Download / Get started everywhere.
    • Account panel — sign up, sign in, and sign out directly from Settings → Account (no more detour through a conversation).
    • Image attachments — attach images to an agent message; image turns route to vision-capable models automatically.
  59. v0.1.7June 2026
    • Enterprise removed — Mel is a personal and small-team tool; the Enterprise/Teams tier and its self-host copy have been removed entirely.
    • Pricing simplified — down to two clean plans at the time (superseded by the credit plans in v0.1.8).
    • SEO and metadata — every page now has its own OpenGraph/Twitter card, sitemap, and robots.txt; optimised hero video.
    • Honesty fixes — removed claims for features that hadn't shipped yet; corrected slash-command names; removed premature open-source claims.
  60. v0.1.6June 2026
    • Accounts & sign-in — added email/password Mel accounts with an in-app sign-in panel for the agent.
    • Account status panel in-app — a clear sign-in/account panel instead of an error when the agent needs authentication.
    • Website redesign — new ink-on-paper design system (Space Grotesk/Mono), real Mel screenshots, provider logos, and a self-playing hero demo video.
  61. v0.1.5Earlier release
    • Agent orchestrator — /orchestrate-agent drives a team of Mel agents across panes with shared memory.
    • Cross-provider fallback — when a provider is rate-limited or unavailable, the agent automatically reroutes to the next available provider and keeps going.
    • Refreshed catalog — 6 providers, 41 verified models.
    • Agent transcript scrolling + delete-conversation from the landing.
  62. v0.1.1Earlier release
    • Live command streaming + Stop — watch long commands stream; Stop kills the whole process tree.
    • Faster orchestration — the lead hands work to its team earlier.
  63. v0.1.0Initial public beta release
    • Team of agents — /orchestrate-cli drives CLI agents (Claude Code, Codex).
    • Agent over SSH — melt a remote session into an agent.
    • Built-in code review — diffs, inline comments, LSP.
    • Command palette + structured blocks.
    • Agent mode — 6 providers, many models, no API keys.

Get started

Want the agent that ships these?

The Mel terminal is free to download, no account needed. The AI agent is free on your own provider key — 1,500 hosted-AI credits a month on Pro.

Get startedDownload the free terminal