Documentation

Risicare traces your AI agents — their LLM calls automatically, and their tool calls, decisions and hand-offs when you add the SDK's decorators — and gives an error code from a 154-code taxonomy to an exception that leaves a traced call. These docs show you how to instrument your agents, read their traces, and see how their failures are classified.

What is Risicare?

Risicare is an observability platform for AI agents. It provides:

• Automatic Instrumentation - Trace LLM calls with import risicare and two environment variables in Python, or with init() and a patchX() call in JavaScript
• Decision-Level Observability - Track THINK/DECIDE/ACT phases
• Error Classification - An exception that leaves a traced call gets a code from a 154-code taxonomy. The SDK assigns 19 of the codes, the same in Python and JavaScript (details)
• Diagnosis and Fix Recommendations (held) - Root-cause analysis and fix recommendations across 7 fix types are built and held for the beta; no new diagnosis or fix recommendation is produced today (details)
• Self-Healing (not enabled) - Fix deployment (canary rollout with automatic rollback) and A/B testing are built but not enabled in this release; a fix cannot be promoted and a deployment cannot be created. The learning stage is not connected (details)

Quick Links

Workflow

The docs follow the product's workflow. Instrument and Observe work today, and every failure is classified; root-cause diagnosis and fix recommendations are held for the beta; fix deployment is built and gated off:

Progressive Integration