OpenClaw Logfire Plugin

Pydantic Logfire observability plugin for OpenClaw. Traces the full agent lifecycle with OTEL GenAI semantic conventions — tool calls, token usage, errors, and optional distributed tracing across services.

Install

openclaw plugins install @ultrathink-solutions/openclaw-logfire

Set your Logfire write token:

export LOGFIRE_TOKEN="your-token"

Add to openclaw.json:

{
  "plugins": {
    "entries": {
      "openclaw-logfire": {
        "enabled": true,
        "config": {}
      }
    }
  }
}

Restart OpenClaw. Traces appear in your Logfire dashboard.

What It Traces

Every agent invocation produces a span tree:

invoke_agent chief-of-staff          (root span)
  |-- execute_tool Read              (file read)
  |-- execute_tool exec              (shell command)
  |-- execute_tool Write             (file write)

Spans follow OTEL GenAI semantic conventions: gen_ai.agent.name, gen_ai.tool.name, gen_ai.usage.input_tokens, gen_ai.usage.output_tokens, etc.

Metrics

• gen_ai.client.token.usage — token count histogram (input/output)
• gen_ai.client.operation.duration — agent invocation latency

Key Configuration

Security & Privacy

• Secret redaction (on by default): Strips API keys, platform tokens, JWTs, and credentials from recorded tool arguments before export
• Tool output is not captured by default (captureToolOutput: false)
• Message content is not captured by default (captureMessageContent: false)
• Data destination: Traces are exported via OTLP HTTP to Pydantic Logfire (US or EU region). No other external endpoints are contacted.
• No local persistence: All data is streamed to Logfire; nothing is written to disk

External Endpoints

By using this plugin, trace and metric data is sent to Pydantic Logfire. Only install if you trust this destination.

Links

• GitHub
• npm
• Logfire
• Blog: We Put OpenClaw Into Production