Documentation
Self-healing observability for AI agents.
Welcome to Risicare documentation. Learn how to instrument your AI agents, diagnose failures automatically, and generate fixes for your agent errors.
Quick Start
Get started in 5 minutes with our Quickstart Guide.
What is Risicare?
Risicare is a self-healing observability platform for AI agents. It provides:
- Automatic Instrumentation - Trace LLM calls with zero code changes
- Decision-Level Observability - Track THINK/DECIDE/ACT phases
- Error Diagnosis - Classify failures using a 154-code taxonomy
- Automatic Fixes - Generate and test fixes with statistical validation
- Safe Deployment - Deploy fixes with canary releases and A/B testing
Quick Links
Quickstart
Get tracing working in 5 minutes
Learn more
View Dashboard
Explore the observability dashboard
Open
Workflow
Follow the six-phase workflow to build self-healing AI agents:
Instrument
Add observability to your agents
Learn more
Observe
View traces, spans, and sessions
Learn more
Diagnose
Automatic error classification
Learn more
Heal
Generate and validate fixes
Learn more
Deploy
Safe deployment with A/B testing
Learn more
Admin
Manage projects and team
Learn more
Progressive Integration
| Tier | Code | What You Get |
|---|---|---|
| 0 | RISICARE_TRACING=true | All LLM calls traced |
| 1 | import risicare | Explicit configuration |
| 2 | @agent() | Agent identity tracking |
| 3 | @session | User session grouping |
| 4 | @trace_think/decide/act | Decision phase tracking |
| 5 | @trace_message/delegate | Full multi-agent support |