Triage at the codebase, not just the ticket
Your agents listen where work happens—Jira, Linear, Telegram, Slack—and answer with grounded, repo-level insight, not generic filler. One core; plugins and config define the rest.
Quick start (npm)
Requires Node.js 24+. Install globally, scaffold config, then run — see installation.
npm i -g agent-detective
agent-detective init
agent-detective doctor Press 1–5 to jump sections (desktop)
Self-hosted Fastify app, JSON + env configuration, Zod-typed options. Configuration hub · Installation
Exhibit A — case visualization
One core. Many sources.
Adapters and plugins turn webhooks and APIs into a single task shape—so the agent always runs the same way, no matter which tool rang the bell.
Case notes
Built for real incidents, not slide decks
“One process from signal to story—tickets in, diffs and comments out.”
- 01
Source-agnostic events
Webhooks and adapters become one task model—Jira, Linear, chat, or your own plugin. The agent runner does not care who knocked.
- 02
Repo-grounded analysis
Local repositories and matching connect incidents to the right tree. Output stays honest to the code you run.
- 03
Operator-ready
Fastify, health, metrics, structured logs, JSON and env. Install via npm on the host; build from source when you extend the core.
Docket
From webhook to well-grounded write-up
The same path every time: normalize, analyze, answer. You keep control of config, secrets, and where the model runs. Embed AI into your engineering workflow as a governed, extensible automation service.
- 1 Event arrives: Jira, Linear, Slack, or your adapter—normalized into a single task model.
- 2 The core agent uses local repo context and your matching rules to analyze and reason.
- 3 Output where you need it: issue comments, optional PR flow, and full observability trail.
Ready to wire your first incident?
The docs walk through plugins, config/default.json, deployment paths, and production checks—no guesswork.
