autonomous issue triage
Reads every issue.
Triages in seconds.
RepoPilot reads each new GitHub issue the moment it lands — classifies it, finds duplicates with cited evidence, asks for missing repro info, routes it to an owner — then drops concrete actions into a queue you approve. You stop triaging. You start shipping.
the problem
Issues pile up faster than anyone can read them.
Every new issue is a context switch: bug or feature? Duplicate? Severe? Who owns it? Does it even have enough to act on? Multiply that by a busy repo and triage quietly becomes a second job — the one nobody volunteers for.
RepoPilot does the reading the instant an issue opens. It arrives already classified, deduplicated, owner-routed, and attached to concrete next actions — so the human decision is a quick yes, not an hour of investigation.
what it does
Eight jobs it takes off your plate.
Type & severity, with reasons
Every issue is labelled bug / feature / question / docs and scored 1–4 — each call backed by a short, auditable rationale, not a black-box guess.
Duplicates, with evidence
Hybrid retrieval over your whole issue history finds likely duplicates and quotes the matching lines from both issues, so you confirm in seconds.
Chases missing repro info
Detects when a bug report is missing steps, versions, or logs — and drafts the exact comment that asks for them.
Points at the likely files
Maps the report to the code most likely at fault via retrieval over your source plus path-term matching — a head start before anyone opens the editor.
Suggests an owner
Routes the issue to a maintainer from CODEOWNERS and area history, so nothing sits unassigned.
You approve. It acts.
Every action lands in a review queue. Approve, edit, reject — or raise the autonomy level and let low-risk labels apply themselves.
Secure by design.
Four-layer HTTP defense: per-route rate limits with Retry-After, tenant-scoped data so you only see your own repos, body-size caps, HSTS + CSP on every response, and a per-installation daily abuse budget — all in front of the agent, not bolted on after.
Watch every run.
Live observability dashboard: p50/p95/p99 latency, per-node LLM cost waterfall, system health tiles (DB · Redis · LLM), dead-letter queue, and traces pushed to Grafana Cloud — so you see exactly what happened on every triage, down to the cent.
how to use it
Live in three steps. No config files.
Install the GitHub App
Add RepoPilot to a repo or your whole org in two clicks. It reads issues and code, writes labels and comments — nothing more.
It triages every new issue
The moment an issue opens, RepoPilot runs the eight-stage pipeline: classification, duplicates, missing info, owner, and proposed actions.
Review the queue — or automate
Open the dashboard, skim the proposals, approve with a keystroke. Raise the autonomy level to let safe actions run on their own.
under the hood
Eight deterministic stages. One bounded graph.
No free-running agent loop. Each stage is timed, injection- screened, and traced with exact token cost.
See it triage a real issue, start to finish.
A captured run on a live repository — every stage, every score, every fraction of a cent it cost.