Ask in the thread, get answers from your code

Every @overcut mention kicks off a three-step flow: it identifies the relevant repositories, clones them for full code access, then opens a multi-agent session that parses your question, gathers evidence from the actual codebase, and replies with a concise, cited answer.

The edge cases, covered

Built to stay accurate and safe under real-world use — grounded answers, read-only access, and a session that keeps your full context alive across follow-ups.

Conversational follow-ups

The session stays open and listens for further @overcut mentions in the same thread, so you can keep asking without restarting the workflow or losing prior context.

Right agent for the question

A coordinator routes each ask to the best-suited specialist — RCA expert for debugging, senior developer for code analysis, technical writer for clear summaries.

Read-only by default

Agents browse code, view diffs, and trace data flows but never push changes or alter settings unless you explicitly ask, keeping your repositories safe.

Every answer cited

Responses point to exact files, lines, commits, or diffs — no hallucinated guesses, and a clear note when information is missing rather than invented.

Secrets stay protected

Tokens and secrets are never echoed back, and answers quote only the minimal relevant snippets instead of dumping large files or raw logs.

Lives where your team already works

No new tool to learn. It answers right inside your issues and pull requests, pulling context straight from your connected repositories.

Running from day one

Fully automatic — it responds to any @overcut mention in an issue, PR, or comment with no setup beyond connecting your repos. The session stays open for follow-ups until you say /done or thanks.