jmclark-lab bets the entire fork on one obscure AI engine

This fork tears out Mike's ability to choose an AI model and hardwires every request to a single, little-known provider.

infrastructurecompliance

Where the original Mike can route work to several AI vendors and switch between them, jmclark-lab has stripped that out entirely. Every chat, summary, and answer now goes to "Fugu," a model family from the AI lab Sakana - with no fallback, no toggle, and no way back to the other engines. The app won't even start unless the Fugu credentials are set, and the team documents the running cost at roughly $5 per million words in and $30 per million out.

There's a genuinely useful side effect: every AI-generated message is now stamped with a record of exactly which model produced it - a clean audit trail of what said what. Worth noting the team's own caveats: the Fugu connection is unproven against a live service, so its tool-driven features may quietly fail, and it's unclear the provider's endpoint is even live yet.

So what A cautionary read for anyone weighing a single-vendor AI bet - but the per-message model-provenance trail is an idea worth stealing for any regulated legal deployment.

View this fork on GitHub →

Spotted something wrong? Or know the PR text has fresher detail than the writeup above?

Commits in this thread

11 commits from jmclark-lab/mike, oldest first. Source extracted verbatim from the harvested git log.

SHA Subject Author Date
108d290d feat: types.ts - Provider=sakana, UserApiKeys.sakana Julio G. Martinez-Clark 2026-06-23 ↗ GitHub
f18e6ea6 Add Fugu models and update model provider function Julio G. Martinez-Clark 2026-06-23 ↗ GitHub
1ace046e Update index.ts Julio G. Martinez-Clark 2026-06-23 ↗ GitHub
303b65d0 Update index.ts Julio G. Martinez-Clark 2026-06-23 ↗ GitHub
89735836 Update .env.example Julio G. Martinez-Clark 2026-06-23 ↗ GitHub
3e3b6946 Create README.md Julio G. Martinez-Clark 2026-06-23 ↗ GitHub
fe2ca9ed Create sakana.ts Julio G. Martinez-Clark 2026-06-23 ↗ GitHub
f9ddb40a Add tests for streamSakana and completeSakanaText Julio G. Martinez-Clark 2026-06-23 ↗ GitHub
45e56f21 feat(chat): persist Sakana Fugu provider_metadata on assistant messages Julio G. Martinez-Clark 2026-06-23 ↗ GitHub
882fb97e feat(db): add provider_metadata column to chat_messages Julio G. Martinez-Clark 2026-06-23 ↗ GitHub
This migration adds a new column 'provider_metadata' to the 'chat_messages' table to store information about the LLM provider and model used for each assistant response.
bfdadaa4 Merge pull request #1 from jmclark-lab/feature/llm-provider-switch Julio G. Martinez-Clark 2026-06-23 ↗ GitHub
feat: replace LLM provider with Sakana Fugu (straight swap)

Capture this thread into my fork

Download a single Markdown prompt that tells Claude how to port every commit above into your working tree — adapting paths and structure to match your repo. Run it via claude -p < capture-thread-681.md from inside the repo you want the changes in.

⬇ Download capture-thread-681.md