Getting started

1. Pick an agent

Any capable coding agent works: Claude Code, Codex, Cursor, Gemini CLI, … BAIT ships entry points for all of them (CLAUDE.md for Claude Code, AGENTS.md for the others).

2. Turn off training — before touching data

This is the one step you must not skip. In your agent’s privacy / data-controls settings, disable model training / enable zero-data-retention. BAIT walks you through it; the principle is in Privacy. When in doubt, prefer API/enterprise tiers with zero data retention over consumer tiers.

3. Install BAIT — one prompt

In your agent, just say:

install https://github.com/DeepWaterIMR/BAIT

The agent takes it from there and walks you through:

  1. Privacy gate — confirms model training is off (and shows you how to turn it off if needed).
  2. Clones BAIT to a sensible location — and refuses to install at a drive root like C:\ for security; it suggests a safe place in your home folder.
  3. Installs the skills globally, so BAIT works in every project without re-cloning.
  4. Sets up the database (asks first — it’s a big download). You’ll need IMR network access (VPN / HI-Adm Wi-Fi) and >2 GB of disk; the download runs in the background with a log and can take hours. Don’t run it over mobile internet or Starlink. The database lands at ~/IMR_biotic_BES_database/ — outside any repo, on purpose. Intranet is only needed for downloading/updating; querying is fully offline.
  5. Verifies the connection and shows you an example.

4. Ask your first question

“What is the largest cod in the database?”

Browse more in Example queries.

Keeping BAIT current

Say “update BAIT” — the agent runs git pull, re-syncs the global skills, and can refresh the database with updateDatabase() (changed years only in normal operation; intranet required). If the database schema changed, BioticExplorerServer safely performs a full rebuild.

How it learns

When you ask something new, the agent offers to save the solution as a cookbook recipe. Review it, commit it, and the next person’s agent benefits. The cookbook is IMR’s shared, version-controlled memory — BAIT gets better the more it’s used.