Now in open beta — orchestrate a team of Mel agents from your terminal. Learn more

Agentic debugging

Mel for bug investigation

Turn a vague bug report into a reproduction and a fix. Mel reads your code, reproduces the issue by running real commands, captures the evidence, and traces it back to the line — without you ever leaving the terminal.

Why it helps

Find the root cause, not just the symptom.

Mel does the legwork of reproducing and isolating a bug so you can spend your time on the fix.

Faster repros

Describe the symptom; Mel runs the commands, reads the output, and converges on a reliable reproduction.

Real evidence

Every step is a clean command block with full output — the stack trace, the failing test, the log line, all navigable.

Traced to the line

Mel follows the trail through your codebase with grep, file reads, and LSP jumps until it can point at the cause.

How it works

From "it's broken" to a pinned root cause.

Mel investigates the way you would — just faster, and it never loses the thread.

  1. 01

    Describe the symptom

    Tell Mel what's wrong in plain English. It plans the investigation and starts running, reading the output as it goes.

  2. 02

    Reproduce and capture

    Mel runs your tests, scripts, and commands — each as a clean block — until it can reproduce the failure on demand.

  3. 03

    Trace and propose a fix

    It walks the call path with grep and go-to-definition, explains the cause, and offers the fix as an approvable diff.

Built in

Reproduce, inspect, and trace — all in one place.

Mel investigates with the same tools you'd reach for, and keeps a record of every step.

Runs real commands

Mel executes tests and scripts in your shell, reads the output, and decides the next step — with a timeout so it never hangs.

Navigable command blocks

Every command is a structured block — jump, filter, and re-run instead of scrolling a wall of text.

Search & jump

Fast code search (git grep) and LSP go-to-definition let Mel follow a bug across files in seconds.

Reaches the remote

Melt an SSH session into an agent and Mel investigates on the box where the bug actually happens.

Navigable output

Every command is a searchable block.

Mel's output isn't a wall of text. Each command is a block you can filter and search — regex or case-sensitive — with the match count live, and where every file:line is a link. Ctrl+Click a backtrace frame to open it at the exact line.

  • Filter & regex search within a block
  • Highlighted matches with a live count
  • Clickable file:line links
  • Jump straight to the failing line
mel · bug-investigation
/agent bug investigation
Filter & regex search within a block
Highlighted matches with a live count
Clickable file:line links
✓ done · reviewed & approved
Bug investigation

Reproduce it once. Fix it for good.

Hand Mel the bug report and get back a reproduction, the root cause, and a fix you can review — instead of a day of guesswork.