opencode-posthog-observability

OpenCode plugin that sends AI generation telemetry to PostHog AI Observability.

Each completed OpenCode assistant message is captured as one PostHog $ai_generation event. Completed and failed OpenCode tool calls are captured as child $ai_span events. The plugin keeps each generation as its own trace while grouping related turns with the OpenCode session ID.

Install

npm install opencode-posthog-observability

Add the plugin to opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-posthog-observability"]
}

Set your PostHog project token:

export OPENCODE_POSTHOG_PROJECT_TOKEN=phc_your_project_token

For EU projects, set:

export OPENCODE_POSTHOG_HOST=https://eu.i.posthog.com

Configuration

The plugin reads config from the first file it finds:

• OPENCODE_POSTHOG_CONFIG
• .opencode/posthog-observability.json
• .opencode/posthog-observability.jsonc
• .opencode/opencode-posthog-observability.json
• .opencode/opencode-posthog-observability.jsonc
• matching files in your global OpenCode config directory

Example:

{
  "projectToken": "phc_your_project_token",
  "host": "https://us.i.posthog.com",
  "distinctId": "opencode",
  "projectName": "my-project",
  "agentName": "opencode",
  "captureInputs": true,
  "captureOutputs": true,
  "captureMetadata": true,
  "maxTextLength": 12000,
  "diagnostics": false,
  "tags": {
    "env": "local"
  }
}

Environment variables override file config:

• OPENCODE_POSTHOG_PROJECT_TOKEN or POSTHOG_PROJECT_TOKEN
• OPENCODE_POSTHOG_HOST
• OPENCODE_POSTHOG_DISTINCT_ID
• OPENCODE_POSTHOG_AGENT_NAME
• OPENCODE_POSTHOG_PROJECT_NAME
• OPENCODE_POSTHOG_CAPTURE_INPUTS
• OPENCODE_POSTHOG_CAPTURE_OUTPUTS
• OPENCODE_POSTHOG_CAPTURE_METADATA
• OPENCODE_POSTHOG_MAX_TEXT_LENGTH
• OPENCODE_POSTHOG_DIAGNOSTICS
• OPENCODE_POSTHOG_FLUSH_TIMEOUT_MS
• OPENCODE_POSTHOG_TAGS as key=value,team=ai or JSON

Privacy

By default, the plugin captures prompt input, assistant output, tool input/output, model/provider metadata, token counts, latency, OpenCode session/message/tool IDs, and configured tags.

When OpenCode exposes reasoning separately from the final answer, reasoning is sent as its own output entry before the assistant response. This keeps PostHog's conversation view readable without discarding reasoning content.

Disable sensitive content capture when needed:

export OPENCODE_POSTHOG_CAPTURE_INPUTS=false
export OPENCODE_POSTHOG_CAPTURE_OUTPUTS=false

The plugin redacts common secret-looking object keys before sending metadata or prompt inputs: token, secret, password, authorization, cookie, apiKey, and bearer.

Development

npm install
npm test

For local OpenCode tests before publishing:

npm pack
npm install ./opencode-posthog-observability-0.1.0.tgz --prefix .opencode

Then point local .opencode/opencode.json at the installed artifact:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["./node_modules/opencode-posthog-observability/dist/index.js"]
}