Most LLM routers decide where to send a prompt by… calling another model.
Wayfinder doesn’t.
It reads the prompts structure - deterministic, offline, microseconds - to keep cheap prompts on your local model and send hard ones to the frontier ones.
Self-hosted, bring your own keys, OpenAI-compatible
@StevenMathern using fable it was pretty easy, and the prompt was fairly basic… you just have to make clear that the current test battery you have is the contract you are implementing against… but I made clear Fable was the orchestrator, and all agents it dispatched should have been Opus :)
I propose a new SWE benchmark: RebuildFromTests
I thought I’d test Fable out on something people always claim the latest frontier model can do.
It’s not merged, but it’s sat there.
When OpenAI drop Sol I’ll run the same exercise, and see what the results are.
Watch this space
You've hit your session limit · resets 10am (Europe/London) ... and im on the Max 20x plan
its 7:30am @DarioAmodei ... what am I supposed to do now?! Touch Grass?!
@andrewmccalip the diligence on this 'side project' is a bit of an inspiration... if only the desktop apps exposed the ability for you to work with them as well, would open up the 'consumer' ecosystem and not just the "hardcore" dev one