TanStack AI
A framework-agnostic, provider-agnostic AI toolkit from the TanStack team — the Switzerland of AI tooling.
About
TanStack AI is the TanStack team's take on AI infrastructure: a set of open-source libraries for building AI features that refuse to lock you into a framework or a cloud. The alpha ships server support for JavaScript/TypeScript, Python, and PHP with full agentic flows and tool calling, plus provider adapters (OpenAI, Anthropic, Gemini, Ollama in TypeScript) and client libraries for vanilla JS, React, and Solid.
Two design choices stand out. First, an open, published protocol between server and client — any language, any transport, as long as you speak the protocol. Second, isomorphic tools: define a tool once with a schema, then provide isolated server and client implementations, with type safety across the boundary. There is also an AI devtools panel for inspecting what the model is doing on both sides of the connection.
The honest caveat: it is alpha software. Python and PHP packages were not yet on their registries at announcement, and the ecosystem around it is thin compared to the Vercel AI SDK. But the team behind React Query and TanStack Router has a long record of shipping boring-in-the-best-way infrastructure, and the anti-lock-in positioning is a real philosophical difference, not marketing.
What it can do
- Framework-agnostic: works with any frontend/backend stack
- Provider adapters for OpenAI, Anthropic, Gemini, Ollama (more coming)
- Full agentic flows with tool calling, streaming, structured outputs
- Isomorphic tools: define once, implement per server/client with type safety
- Open published client/server protocol; AI devtools inspection panel
Setup
- Install the core package (
@tanstack/ai) plus the adapter for your provider (@tanstack/ai-openai,@tanstack/ai-anthropic,@tanstack/ai-gemini,@tanstack/ai-ollama). - Define tools once with schemas (isomorphic tool definitions).
- Run the agentic loop server-side with
chat({ adapter, messages, tools })and stream via SSE. - Add a client library (
@tanstack/ai-react, Solid, or vanilla) and the AI devtools panel for debugging.
Prerequisites
A JS/TS project (Python/PHP support maturing); API keys for your chosen model providers.
Use cases
Works with
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Adoption 6: provider-tier estimate — thinner evidence (alpha, no public adoption metric); TanStack brand is strong but the AI package is new; Maintenance 7: alpha announced Dec 2025, still alpha; team ships steadily but package surface is evolving; Openness 10: MIT per TanStack convention (0.7*10) + official docs verified at tanstack.com/ai (0.3*10); Manual 7: anti-lock-in design corroborated by announcement and technical analyses; docked for alpha status and thin ecosystem vs Vercel AI SDK. License: MIT (TanStack standard).
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Adoption 6: provider-tier estimate — thinner evidence (alpha, no public adoption metric); TanStack brand is strong but the AI package is new
- · Maintenance 7: alpha announced Dec 2025, still alpha; team ships steadily but package surface is evolving
- · Openness 10: MIT per TanStack convention (0.7*10) + official docs verified at tanstack.com/ai (0.3*10)
- · Manual 7: anti-lock-in design corroborated by announcement and technical analyses; docked for alpha status and thin ecosystem vs Vercel AI SDK
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.