Veyra

A local AI agent you shape through conversation. Chat, approve commands, and ask Veyra to build and test improvements. Each version is a Git commit.

Start

Requires Python 3 with venv.

./supervisor/supervisor

Open http://localhost:3000. In Settings, choose a ChatGPT subscription, a Claude subscription (requires the official Claude Code client), or the OpenAI or Anthropic API. There is no default service or automatic switch to API billing. Complete subscription sign-in on the computer running Veyra. Save API keys only in Settings → API keys on localhost, never in chat.

Use

Describe what you want in chat. Veyra proposes commands for your approval, implements changes, tests them, and commits the next version.

For another device on your private network, open http://COMPUTER_IP:3000. The initial network login is veyra / veyra; change it in Settings. HTTP is unencrypted, so use an encrypted tunnel on untrusted networks.

Stop and recover

Stop the foreground supervisor with Ctrl+C. For recovery, open http://127.0.0.1:3001 on the Veyra computer. Recovery works independently of the agent; do not expose its port to the network.

• agent/ — the app and model integrations
• supervisor/ — process lifecycle and local recovery
• state/ — persistent local data and credentials, ignored by Git
• workspace/ — local development area, ignored by Git

Committed agent changes restart the app automatically. Persistent state survives version changes, and open pages wait for unfinished drafts before updating.