Skip to content
triage docs

Overview & Concepts

Welcome to triage—the zero-overhead Go panic isolation tool, automated GitHub issue & bugfix PR engine, and AI diagnostic platform.

Go HTTP Server (your app)
        │  panic → defer/recover
        ▼
 triage SDK Middleware
        │  async, non-blocking POST (<0.02ms overhead)
        ▼
 Triage Server  (:8080) [Zero-Dependency Single Binary / Container]
  ├── Embedded Studio Dashboard (Vite + React 19 SPA)
  ├── Embedded SQLite Database  (Zero-config auto-migration)
  ├── Ingestion & In-Memory Auth (<1ms API key verification)
  ├── Server-Driven OAuth & JWT Session RBAC (Owner/Admin/Dev/Viewer)
  ├── Resolve Multi-File Package AST Context
  │     ├── Crash function *ast.FuncDecl
  │     ├── Receiver struct & type definitions (cross-file)
  │     └── Related constructors (New<Type>) & package helpers
  ├── Multi-Provider AI analysis       → root_cause + suggested_fix + git patch
  ├── Automated GitHub Actions
  │     ├── File GitHub Issue with AST snippets & telemetry
  │     └── Open Bugfix Pull Request with auto-applied git diff
  └── Return JSON response / Stream Live Incidents via SSE

Why Triage?

Section titled “Why Triage?”

Traditional application monitoring tools and crash loggers capture giant stack traces and attempt to send whole files to large language models. This has several drawbacks:

  1. High Token Costs & Prompt Saturation: Ingesting whole 2,000-line source files into LLMs wastes thousands of tokens per incident.
  2. Context Dilution or Blindness: Raw stack traces lack code context, while single-function fragments miss struct definitions and uninitialized constructors in adjacent files.
  3. Synchronous Request Overhead: Heavy APM agents often add 0.8ms – 1.5ms of latency to each HTTP request.

The Triage Solution

Section titled “The Triage Solution”
  • Multi-File Package AST Slicing: Using Go’s standard go/parser and go/ast packages, Triage isolates the exact crashing function along with referenced struct definitions, constructors, and helper functions across package files.
  • Domain-Aware AI Triage: Enriches LLM prompts with optional architectural context and business invariants, ensuring diagnoses and generated patches respect your service boundaries.
  • >90% Token Reduction: Instead of sending entire repositories or 2,000-line source files, Triage delivers a selectively pruned multi-file context to AI models, reducing diagnostic costs to less than $0.0001 per incident.
  • Sub-0.02ms Client Latency: Telemetry is handed off to an asynchronous 4-worker pool backed by a 1,000-job queue. HTTP responses return immediately with zero blockage.
  • Automated Bugfix Pull Requests: Creates dedicated fix branches, synthesizes clean fixes via the AI engine, commits the changes, and opens a GitHub Pull Request with 1 click.
  • Automated GitHub Issue Triaging: Automatically creates GitHub issues with structured root causes, formatted AST code, and drop-in git patches.
  • Multi-Project & Monorepo Support: Track multiple Go services from a unified workspace with automatic go.mod module discovery and a project switcher in the dashboard header.
  • Real-Time Telemetry Streaming (SSE): Unidirectional Server-Sent Events stream newly ingested crashes and incident state updates directly to connected dashboards with zero page reloads.
  • Zero-Dependency Single-Container Self-Hosting: Run the entire system with embedded SQLite and embedded React UI in a single lightweight container or standalone binary.

Architecture at a Glance

Section titled “Architecture at a Glance”
Component Port Technology Purpose
Triage Server :8080 Go 1.26+ (Embedded UI) Telemetry ingestion, AST slicing, Multi-provider AI, and React Studio Dashboard
Go Client SDK Embedded Go 1.26+ Non-blocking HTTP middleware with panic recovery
Embedded Storage Embedded SQLite (WAL mode) Zero-config embedded persistence for incidents, keys, and settings
Documentation & Web :4321 Astro & Starlight (Bun) Public landing site and technical reference

Next Steps

Section titled “Next Steps”