TraceRoot

Description

TraceRoot is an open-source observability and self-improving layer for AI agents. It detects failures in production, root-causes them against your source code and GitHub history, opens verified fix PRs, and evaluates every fix so your agent gets better with every release. It is backed by Y Combinator.

A detector scans production traces for failures that metrics miss, such as hallucinations, wrong tool calls, broken logic, and dropped user intent. When one fires, TraceRoot runs root cause analysis automatically and notifies you over Slack and email with the finding and the fix. You can also kick off a TraceRoot agent that runs in an isolated sandbox with full access to your traces, GitHub history, and production source code, opening a fix PR and resolving GitHub issues and Linear tickets automatically, using a hosted model or your own key.

TraceRoot works across your stack with native SDKs for Python and JS/TS, plug-and-play support for common frameworks, and full OpenTelemetry compatibility, including auto-instrumentation for the OpenAI Agents SDK and LangChain/LangGraph that captures agent runs, steps, tool calls, handoffs, LLM calls, tokens, cost, and latency. Pricing starts free with no credit card, including 50k events/month, 15-day retention, and monthly chat, RCA, and detector runs, with Starter, Pro, and Enterprise plans adding more events, retention, rate limits, and SLA support. Bring-your-own-key is supported on all plans.

TraceRoot's Core Features

• Detector that scans production traces for agent failures

• Automatic root cause analysis with Slack and email alerts

• Sandboxed agent that opens verified fix PRs

• Resolves GitHub issues and Linear tickets automatically

• Evaluation of every fix

• Native SDKs for Python and JS/TS with OpenTelemetry compatibility

• Auto-instrumentation for OpenAI Agents SDK and LangChain/LangGraph

• Hosted LLM or bring-your-own-key on all plans

How to use TraceRoot?

1. 1

   Instrument your agent: Add the Python or JS/TS SDK, or use OpenTelemetry compatibility.

2. 2

   Send telemetry: Let traces from your agent runs, tool calls, and LLM calls flow in.

3. 3

   Detect failures: Have the detector scan traces for hallucinations, wrong tool calls, and broken logic.

4. 4

   Get root cause: Receive automatic RCA with the finding and fix over Slack and email.

5. 5

   Ship the fix: Let a TraceRoot agent open a verified fix PR and resolve related issues and tickets.

TraceRoot's Use Cases

• AI agent observability
• Failure detection
• Automated fixes
• Root cause analysis
• Continuous improvement

FAQ from TraceRoot

• What is TraceRoot?

  TraceRoot is an open-source self-improving layer for AI agents that provides observability, detects failures in production, root-causes them against your code and GitHub history, opens verified fix PRs, and evaluates every fix.

• What kinds of failures does it detect?

  The detector scans production traces for failures that metrics miss, including hallucinations, wrong tool calls, broken logic, and dropped user intent.

• How does it fix issues?

  You can run a TraceRoot agent in an isolated sandbox with access to your traces, GitHub history, and production source code, which opens a fix PR and resolves GitHub issues and Linear tickets automatically, using a hosted model or your own key.

• What does it work with?

  It offers native SDKs for Python and JS/TS, plug-and-play support for common frameworks, and full OpenTelemetry compatibility, with auto-instrumentation for the OpenAI Agents SDK and LangChain/LangGraph.

• Is there a free plan?

  Yes. The free plan needs no credit card and includes 50k events/month, 15-day retention, and monthly chat, RCA, and detector runs, with Starter, Pro, and Enterprise plans adding more capacity and support. BYOK is supported on all plans.