jmclark-lab drafts a full engine swap to Sakana's Fugu, then parks it
A complete, deployable blueprint for standardizing this fork on a single new AI model - opened for review, then closed without merging.
jmclark-lab laid out exactly what it would take to rip out this fork's current AI engine and run everything on Fugu, a model from the lab Sakana AI - no toggle, no fallback to the old provider. It's a clean, ready-to-ship package that bundles three things:
- A new connector that routes both quick one-shot answers and live, streaming replies through Fugu.
- An audit trail: every answer the assistant saves now carries a stamp of which provider and model produced it - even on cancelled or failed requests.
- A startup check that refuses to boot the server without valid credentials, so a missing key fails loudly up front instead of surprising you in production.
The catch: jmclark-lab flagged it "do not merge" from the start and it was closed a few days later, so none of it reached the base repo.
So what Worth a look for anyone weighing a single-vendor AI backend - and a reminder that the model behind your answers is a decision to record, not a default to forget.
Spotted something wrong? Or know the PR text has fresher detail than the writeup above?