Olivares AI

Olivares AI starts, confines and records the AI coding tools a team already uses: Claude Code, Codex, Gemini CLI, OpenCode and local models through Ollama. Each person signs in each tool with their own account. Olivares applies the team's policy to every session, holds risky actions for approval and blocks the ones the policy forbids.

The access map shows what each agent read and wrote across repositories, databases and chat, next to what it was granted, and flags a write nobody granted. Every decision lands in a signed, tamper-evident ledger that you can filter, verify and export.

It runs on your own servers as one binary with its console built in, as a container, or from a .deb, .rpm or .apk package, on SQLite or PostgreSQL. Community is free and open source under AGPL-3.0 for a person or a team. Business adds what companies need: directory import, several identity providers, runtime security and audit evidence.

We built it with Claude Code and Codex, the same tools it governs. The hard part was keeping a complete record of what every agent did that anyone can verify, while Olivares itself never holds a tool's password.