Architecture

The Autonomous Incident Agent (AIA) is a modular, event-driven system designed to automatically detect, analyze, and create fix suggestions for application failures. It uses OpenTelemetry for observability, AI for analysis, and GitHub for delivering fixes.

#System Overview

AIA is composed of several specialized microservices that coordinate to handle the incident lifecycle.

Core Services

1. Agent (`apps/agent`)

The OpenTelemetry receiver running on port 4318. Receives OTEL traces and logs from your application, detects errors using built-in detectors (HTTP 5xx, exceptions, latency spikes, crashes), and forwards incidents to the Router.

Key Features:

OTEL HTTP endpoint (/v1/traces, /v1/logs)
Built-in error detectors
Trace ID-based deduplication (30-second window)

2. Router (`apps/router`)

The incident orchestrator. Receives detected incidents from the Agent, enriches them with context (snapshots, file paths), stores them in the State service, and triggers the Autopsy service for analysis.

Key Features:

Incident enrichment with code snapshots
State persistence coordination
Autopsy triggering

3. Autopsy (`apps/autopsy`)

The AI-powered analysis engine. Uses You.com API to analyze incidents and generate:

Root cause analysis
Code patches (git diff format)
AI fix prompts for developers
Step-by-step manual remediation instructions

Key Features:

AI reasoning (You.com API)
Patch generation with validation
Detailed fix instructions
Confidence scoring

4. State (`apps/state`)

The persistence layer using PostgreSQL (Neon). Stores all incident data, autopsy results, and PR metadata.

Database Schema:

incidents table: Core incident data
autopsy_results table: AI analysis results
Indexed by incident_id for fast lookups

5. Git (`apps/git`)

The GitHub integration service. Handles all git operations:

Clones repository using GitHub token
Creates incident branch (aia/incident-{id})
Attempts to apply AI-generated patch
Commits changes (or saves failed patch)
Pushes to GitHub
Creates Pull Request with fix details

Key Features:

Automatic patch application
Failed patch handling (patch_failed_*.diff)
PR creation with detailed context

6. Dashboard (`apps/dashboard`)

The web UI for monitoring incidents. Provides:

Real-time incident list
Autopsy results viewer
AI fix prompt (copy-to-clipboard)
Manual remediation steps
GitHub PR links
PDF report generation

Tech Stack: Bun + HTML templates

/trigger endpoint to cause intentional errors
OTEL instrumentation
Error simulation

#Data Flow

Terminal

1. Application Error
   ↓
2. OTEL Trace → Agent (port 4318)
   ↓
3. Detector identifies incident
   ↓
4. Router enriches with context
   ↓
5. State stores incident
   ↓
6. Autopsy analyzes with AI
   ↓
7. State stores autopsy result
   ↓
8. Git creates PR
   ↓
9. Dashboard displays results