TanStack AI reaches 1.0 with a stable API for 23 providers
The TanStack AI SDK locks its core chat API for production use, while a new experimental harness package handles long-running, persistent agent sessions.
TanStack AI hit 1.0 this week, moving its core chat API out of alpha and committing to API stability for production use. The release now covers 23 provider adapters and eight UI framework integrations, up from four providers and three frameworks at alpha.
The headline addition is @tanstack/ai-harness, a separate package for persistent, long-running agent sessions, distinct from the stateless chat() function that handles one request at a time. A harness session stays open, accepts new messages while it works, can run other agents, and survives a deploy. It ships at 0.x on purpose: its APIs are still experimental and can change in a minor release, so TanStack kept it out of the stable 1.0 core. Upgrading from 0.x means running the provided codemod, updating every @tanstack/ai* package together, adding a required threadId parameter by hand, and handling tool-approval interrupts differently.
Why it matters: anyone building chat or agent features on top of an LLM API now has a vendor-neutral SDK with a committed stable surface, rather than pinning to a 0.x release that could rename things under them. The harness split is also an honest signal about where the hard problems still are: single-turn chat is solved, persistent agents are not.