eyes4ai

Passive AI activity recorder for Git repositories. Tracks usage across OpenAI Codex, Claude Code, and more — all data stays local.

What it does

eyes4ai sits in the background and records how AI coding tools are used in your repos:

• Sessions, turns, and token costs — broken down per tool when you use multiple
• Git commit correlation — which AI sessions led to actual commits
• Yield metrics — session-to-commit rate, cost per commit, abandoned sessions
• Trend comparison — current vs. previous period

It uses OpenTelemetry to passively capture telemetry that Codex and Claude Code already emit. No code changes, no wrappers, no proxies.

Quick start

# Install eyes4ai globally
npm install -g @eyes4ai/cli

# Set up everything (Codex, Claude Code, git hooks, auto-start daemon)
eyes4ai install --global

# Or install for a single repo only
cd your-repo
eyes4ai install
eyes4ai serve

Important: Restart any running Codex or Claude Code sessions after install so they pick up the new config.

Then use your AI tools as usual. When you want a report:

eyes4ai report --days 7

Period: last 7 days

  Sessions:              12  (codex: 8, claude: 4)
  Turns:                 87  (codex: 55, claude: 32)
  AI-active days:        5 / 7
  Estimated cost:        $4.23  (codex: $2.80, claude: $1.43)
  AI-linked commits:     9 / 14 (64%)
  Avg cost per commit:   $0.47

How it works

1. Install configures your AI tools to emit OpenTelemetry logs to a local endpoint
2. A lightweight server receives OTLP/HTTP payloads, normalizes them, and appends to daily JSONL files
3. Git post-commit hooks capture commit metadata and link commits to recent AI sessions
4. Reports aggregate the local data into actionable metrics

All data lives in .eyes4ai/private/events/ — nothing leaves your machine.

Commands

| Command | Description | |---------|-------------| | install [port] [--global] | Configure AI tools, git hooks, and (with --global) auto-start daemon | | uninstall | Remove the background daemon | | serve [port] | Start the OTel ingestion server | | report [--days N] [--json] | Generate an activity report | | record-commit [hash] | Manually record a commit | | reprocess [file] | Re-normalize events (after schema/pricing updates) |

Supported tools

| Tool | Status | Detection | |------|--------|-----------| | OpenAI Codex CLI | Supported | codex.* OTel events | | Claude Code | Supported | claude_code.* OTel events | | Your tool here | Add a provider | Plug & play architecture |

Privacy

• All data is stored locally in .eyes4ai/private/
• Raw prompts are never stored — only hashed fingerprints and short previews
• Sensitive attributes (emails, account IDs) are redacted
• No network calls except localhost OTel ingestion and optional anonymous telemetry (disable with EYES4AI_NO_TELEMETRY=1)

Docs

Full documentation: eyes4ai.selcukcihan.com

Contributing

See CONTRIBUTING.md for development setup, testing, and how to add new AI tool providers.

License

MIT