Skip to main content
GitHub

Getting Started

Get up and running with Risicare in minutes.

Welcome to Risicare! This section will help you get started with observability and automated error diagnosis for your AI agents.

What is Risicare?

Risicare is an observability and error-diagnosis platform for AI agents. It provides:

  1. Automatic Instrumentation - Trace LLM calls with zero code changes
  2. Decision-Level Observability - Track THINK/DECIDE/ACT phases
  3. Error Diagnosis - Classify failures using a 154-code taxonomy and generate fix recommendations

Roadmap: automatic fix application

Stages 1–3 above (detect → diagnose → generate a fix recommendation) ship today. The next stages — applying fixes automatically with statistical validation, and safe deployment via canary releases / A/B testing — are on the roadmap and not yet connected to the runtime. Generated fixes are surfaced as recommendations; they are not auto-deployed.

Progressive Integration

Risicare uses a tiered approach that lets you start simple and add depth as needed:

TierWhat You AddWhat You Get
0Environment variableAll LLM calls traced
1import risicareExplicit configuration
2@agent()Agent identity tracking
3@sessionUser session grouping
4@trace_think/decide/actDecision phase tracking
5@trace_message/delegateFull multi-agent support

How It Works

Your Agent Code
      ↓
  SDK Instrumentation (automatic)
      ↓
  Rust Gateway (batching, validation)
      ↓
  ClickHouse (traces) + PostgreSQL (configs)
      ↓
  Diagnosis Engine (4-stage LLM pipeline)
      ↓
  Fix Generation (ships today)
      ↓
  Hypothesis Testing → Deployment (roadmap — not yet connected)

Next Steps

Install the SDK

Run pip install risicare and set your API key.

Start Tracing

Set RISICARE_TRACING=true and run your agent.

View Dashboard

Open app.risicare.ai to see traces.

Review Diagnoses & Fix Recommendations

Errors are diagnosed automatically and surfaced with fix recommendations. (Automatic fix application and deployment are on the roadmap — see below.)