# Trace Flow

> Trace Flow captures model API and coding-agent analytics so you can track costs and performance over time.

The hosted service requires an account with access. Coding-agent analytics is available in private alpha.
For your own deployment, see /docs/quick-start#self-hosted-deployments.

Use this file as your starting point. Fetch only the linked documentation you need for the current task.
For model API integration, read /agents.md first, then fetch /docs/quick-start.md for env vars and the first traced request.
For local Claude Code, Codex CLI, or Cursor analytics, fetch /docs/collector.md.

## Core Documentation

- [Quick Start](https://preview.trace-flow.dev/docs/quick-start.md): Start here: add TRACE_FLOW_API_KEY, point your provider at the gateway, and send the first traced request.
- [SDK Reference](https://preview.trace-flow.dev/docs/sdk-reference.md): Provider-specific examples for OpenAI, Anthropic, Google, OpenRouter, and Groq with Vercel AI SDK and native SDKs.
- [OpenTelemetry](https://preview.trace-flow.dev/docs/opentelemetry.md): Advanced setup for custom spans, events, and attributes. Full control over your trace hierarchy.
- [AI Agents](https://preview.trace-flow.dev/agents.md): Hand your coding agent a single machine-readable guide for env vars, gateway wiring, and first-run verification.
- [Coding-Agent Collector](https://preview.trace-flow.dev/docs/collector.md): Install the desktop collector, understand supported sources, and review its local parsing and privacy boundaries.
- [MCP Server](https://preview.trace-flow.dev/docs/mcp.md): Give AI agents direct access to your trace data. Query traces, analyze costs, and debug issues.

## Agent Bootstrap

- [Agent Bootstrap Markdown](https://preview.trace-flow.dev/agents.md): Canonical instructions for AI assistants

## Optional

- [Documentation Index](https://preview.trace-flow.dev/docs): Human-friendly docs landing page