Post

I want the crash test. Kill the host mid tool call, bring up a second one, and show the resumed turn doesn't fire that tool twice. Durable sessions live or die on that. Put it in the e2e fixtures and the harness earns its way off 0.x.

Can upgrade your brand name to : wisewebkit.com or cloudikit.com or wiserkit.com

Finally, a clean, type-safe pipeline to help developers ship lukewarm hallucinations directly to users with minimal friction.

Finally stable. The framework choice flexibility makes it much easier to integrate into existing stacks.

Already using TanStack AI? Upgrade your stable packages together and follow the migration guide. The experimental harness packages stay on 0.x. tanstack.com/ai/latest/do... Have fun building. 🌴

the product promise is 'choice of models'. the hidden tax is consistency: every provider wants different auth, schemas, pricing, latency. normalize that behind one OpenAI-compatible surface and model choice becomes a config, not a rewrite. routing is that normalization

Curious how this differentiates from Vercel's AI SDK, that's been the default bring-your-own-model option for a while now. More competition at that layer is good though.

Different angle from Vercel's AI SDK though, TanStack isn't tied to Next.js or Vercel's hosting, so it's actually usable if you're not already in that ecosystem.

Lock-in-free, bring-your-own-model is the pitch, but Vercel's AI SDK already has the default install base. Winning devs over means beating muscle memory, not just matching features.

so i think tanstack generally builds the right things from the developer perspective but it would be a lot more confidence inducing if the website peoples’ website actually worked properly