pi-sentry-monitor

Sentry AI Monitoring extension for pi coding agent sessions and tool calls.

Mirrors what opencode-sentry-monitor does for OpenCode — tracks agent sessions, tool executions, token usage, and errors as Sentry traces.

Trace Hierarchy

gen_ai.invoke_agent (session span)
├── gen_ai.execute_tool (per tool call)
│   ├── gen_ai.tool.input  (if recordInputs enabled)
│   └── gen_ai.tool.output (if recordOutputs enabled)
└── gen_ai.request (per assistant message)
    └── token usage attributes (input, output, cached)

Installation

Global (applies to all projects):

pi install npm:pi-sentry-monitor

Project-local (checked into the repo, shared with teammates):

pi install npm:pi-sentry-monitor -l

Then run /reload in pi to activate without restarting.

A companion skill is included — once installed, ask pi to "set up Sentry monitoring" and it will walk through creating a Sentry project, configuring the DSN, and verifying traces are flowing.

Configuration

Create .pi/sentry-monitor.json (or .jsonc) in your project:

{
  "dsn": "https://[email protected]/456"
}

Config File Locations (searched in order)

1. Path in PI_SENTRY_CONFIG env var
2. <cwd>/.pi/sentry-monitor.json[c]
3. ~/.pi/agent/sentry-monitor.json[c]

Environment Variable Overrides

| Variable | Description | |---|---| | PI_SENTRY_DSN / SENTRY_DSN | Sentry DSN | | PI_SENTRY_TRACES_SAMPLE_RATE | Sample rate (0-1) | | PI_SENTRY_RECORD_INPUTS | Record tool inputs (true/false) | | PI_SENTRY_RECORD_OUTPUTS | Record tool outputs (true/false) | | PI_SENTRY_ENABLE_METRICS | Enable metrics (true/false) | | PI_SENTRY_TAGS | Custom tags (format: key:value,key:value) | | SENTRY_ENVIRONMENT | Environment name | | SENTRY_RELEASE | Release version |

Full Config Schema

{
  "dsn": "https://...",
  "tracesSampleRate": 1,
  "environment": "production",
  "release": "1.0.0",
  "debug": false,
  "agentName": "my-agent",
  "projectName": "my-project",
  "recordInputs": true,
  "recordOutputs": true,
  "maxAttributeLength": 12000,
  "includeMessageUsageSpans": true,
  "includeSessionEvents": true,
  "enableMetrics": false,
  "tags": {
    "team": "platform"
  }
}

Event Mapping

| pi event | Sentry span/action | |---|---| | session_start | Create root gen_ai.invoke_agent span | | session_shutdown | End session span, flush Sentry | | model_select | Track model/provider changes | | tool_execution_start | Start gen_ai.execute_tool child span | | tool_execution_end | End tool span, capture errors | | message_start (assistant) | Open gen_ai.request span (captures real LLM latency) | | message_end (assistant) | Close gen_ai.request span, attach token usage + content | | agent_end | Flush pending data to Sentry |

Development

git clone https://github.com/sergical/pi-sentry-monitor && cd pi-sentry-monitor
npm install
npm run build

# Install local copy for testing
pi install ./path/to/pi-sentry-monitor
# or test without installing (ephemeral)
pi -e ./dist/index.js

License

MIT