Supported Tools

eyes4ai uses OpenTelemetry as its ingestion protocol. Any AI coding tool that can export OTel logs can work with eyes4ai. Currently, two tools have built-in support with automatic configuration.

Codex

OpenAI’s Codex has native OTel log export. eyes4ai configures it via .codex/config.toml.

Events captured:

Event	What it records
codex.user_prompt	Prompt hash, redacted preview
codex.sse_event (response.completed)	Token counts, model, estimated cost
codex.tool_result	Tool name, duration, success/failure
codex.tool_decision	Tool approval/denial
codex.conversation_starts	Session start, model, MCP servers
codex.websocket_*	Transport-level events

Claude Code

Anthropic’s Claude Code exports OTel logs when configured with the right environment variables. eyes4ai sets these in .claude/settings.json.

Events captured:

Event	What it records
claude_code.user_prompt	Prompt hash, redacted preview
claude_code.api_request	Token counts, model, cost, duration
claude_code.tool_result	Tool name, duration, success/failure
claude_code.tool_decision	Tool approval/denial

Adding a new tool

eyes4ai’s ingestion server accepts standard OTLP/HTTP JSON at POST /v1/logs. To add support for a new tool:

1. Configure the tool to export OTel logs to http://127.0.0.1:4318/v1/logs
2. Add a normalizer in packages/ingestion/src/ that maps the tool’s event attributes to the eyes4ai.event.v1 schema
3. Register the normalizer’s detect + normalize functions in normalize-dispatch.ts
4. Optionally add pricing entries in pricing.ts
5. Optionally add an install config module

The provider detection uses OTel resource attributes (service.name) and event name prefixes to route to the correct normalizer. Unknown events are stored as raw type and can be normalized later via reprocess.