OPI — Open Prompt Initiative

Track, redact, and share your AI coding tool usage across a team.

OPI extracts prompt sessions from tools like Claude Code, Codex CLI, and Gemini CLI, strips sensitive data, and exports them as JSON to a shared directory in your git repo. A pre-commit hook keeps everything in sync automatically.

Why

AI coding tools generate thousands of prompts, but that usage data stays siloed on each developer's machine. OPI makes it visible:

• See what models your team is using — shields.io badges in your README show token counts per model
• Share prompt history — exported sessions live in opi-prompts/ alongside your code
• Keep secrets out of git — built-in redaction strips API keys, PII, file paths, and URLs before export

Install

# With uv (recommended)
uv tool install opi-cli

# Or with pip
pip install opi-cli

Or run without installing:

uvx opi-cli init

Quick start

cd your-repo

# Set up pre-commit hook and opi-prompts/ directory
opi init

# Run the pipeline: extract, redact, export, generate badges
opi update

That's it. On every commit, the pre-commit hook runs opi update to keep your prompt exports current.

Commands

Multi-user

Each developer runs opi update independently. Sessions export to opi-prompts/ which is committed to git. The generated README aggregates token usage from all users' exports, so badges reflect the full team's usage.

Redaction

By default, OPI redacts:

• API keys and tokens (OpenAI, GitHub, AWS, etc.)
• PII (email addresses)
• File paths
• URLs with credentials

Configure via opi config --edit or ~/.opi/config.toml:

[redact]
secrets = true
pii = true
file_paths = true
urls = true
org_terms = ["acme-corp", "internal-project"]

Supported tools

• Claude Code
• Codex CLI
• Gemini CLI

License

MIT