TanStack AI now has subagents π€
Define agents once, pass them to chat(), and each one streams its own card in your UI.
Jev routes every message to the right agents, in parallel or in orderβ‘
π To no one's surprise, Jev support just landed in TanStack AI via `decide()`, our new API for typed choices, scores, and booleans.
We're excited to use this new API with more models soon, but for now the "jevolution" is in full swing!
Jev on, ya'll.
Generate dynamic worlds with TanStack AI! π
Use the new generateWorld function and generate worlds with realtime edits.
Check the docs here:
tanstack.com/ai/latest/docs/β¦
We officially support @remix_run v3 as a TanStack AI adapter!
Same API you already know: createChat in a clientEntry island, typed tool components, working example included.
npm i @tanstack/ai-remix
π Celebrating 6 months of partnership with @Railway by making the TanStack Start -> Railway path even better with faster prompt-to-prod, and an end-to-end experience on sandboxes.
Start here! π
docs.railway.com/guides/tansβ¦
π¨ New partner alert!
π @Render and TanStack are joining forces to help you ship better, faster apps with zero ops. Whether that's your first user or your billionth, Render has everything you'll ever need to get to production.
Start building now π
tanstack.com/partners/render
TanStack AI now supports Cloudflare as an official adapter!
Connect to your gateway via bindings, api key or expose it to the OpenAI adapter directly!
npm i @tanstack/ai-cloudflare
ποΈ + β² = π
TanStack and @Vercel are officially teaming up!
Weβre working together to make Vercel a first-class home for TanStack Start with better support for deploying and running anything you can dream up with TanStack on Vercel.
Start here! π
vercel.com/kb/guide/deploy-aβ¦
Your chat options are now your UI types. π₯
TanStack AI ships with headless UI kits!
Every tool, part and interrupt widget completely typesafe. Forget a component and it won't compile.
React, Solid, Vue, Svelte, Angular, Preact, Octane.
Check it out π
Your context window is fixed. Your chat is not. π
TanStack AI shrinks what the model sees and leaves your transcript alone allowing for infinite compacting chats.
You pick your compaction strategy!
npm i @tanstack/ai-compaction
Pasting every skill into the system prompt is a tax you pay on every request. πΈ
withSkills() ships a one-line catalog instead. The model reads it, calls load_skill, and pulls the full instructions only when the task needs them.
Works with any provider.
TanStack AI now supports @Lovable AI Gateway. π
Six modalities, one adapter shape:
chat β lovableText
image β lovableImage
video β lovableVideo
embed β lovableEmbedding
speech β lovableSpeech
transcribe β lovableTranscription
npm i @tanstack/ai-lovable
Stop paying for your users' tokens. πΈ
Headless BYOK just landed in TanStack AI: users bring their own API key, it never touches your server, and 15 provider adapters got a /byok entry in one pass.
Two calls to wire up. π
TanStack AI now has first-party generic interrupts. π₯
The server can stop a run and ask the user for typed data, no fake tool call, no hand-wired resume.
One definition types the payload going out and the answer coming back.
Four boundaries:
- beforeModel
- afterModel
- beforeTools
- afterTools
WebSockets just landed in TanStack AI. π
One socket for the whole conversation.
Full-duplex. And if it drops mid-stream, the client reopens at the last offset and replays only the gap, no second model call.
Works on Node (`ws`) and Cloudflare (`toWebSocketResponse`)
TanStack AI officially added the ByteDance adapter! π
Generate videos with Seedance 2.5 today.
Check out what we generated through TanStack AI below:
You might've seen recently that @Docker released support for microVM sandboxes for agents.
We just shipped an adapter for it directly in TanStack AI!
Run your agents inside of Docker Sandboxes with TanStack AI today.
Still leaving TanStack AI to embed and rerank? π€
Not anymore. Both are first-class activities now. Fuse text + image into one vector. Rerank your own objects, get them back typed.
Two new Code Mode isolates landed too: QuickJS on Bun, Daytona.
Approvals were custom events. Not anymore. π
Interrupts in TanStack AI got rebuilt on AG-UI. Bound items, typed originalArgs, and decisions that can carry data
tanstack.com/ai/latest/docs/β¦
It forgot your name. Again. π§
Server-side memory landed in TanStack AI.
One middleware, two verbs: recall pulls what matters into the prompt before the model, save persists the finished turn after the stream.
Injected memories shown in devtools.
tanstack.com/ai/latest/docs/β¦
Durable sandbox runs in TanStack AI. π
A disconnect now detaches the run instead of killing the sandbox.
The agent writes to a journal inside the box, and a later request picks up at the exact byte.
tanstack.com/ai/latest/docs/β¦
Four minutes into a video render. Someone hits reload. π¬
Generation persistence in TanStack AI:
The run is recorded, and a reload repaints status, result and error like the run just finished.
tanstack.com/ai/latest/docs/β¦
Reload the page mid-answer. The reply keeps streaming. π
Durable chat just landed in TanStack AI.
One middleware on the server, one option on the client, and a refresh restores the transcript and rejoins the run that's still going.
Docs: tanstack.com/ai/latest/docs/β¦
TanStack Table V9 is shipping with a record number of framework adapters from us, with refreshed support for React, Vue, Angular, Solid, Svelte, and Lit, but also brand new support for Preact, Octane, Ember, and Alpine.
π΄Announcing TanStack Table V9π΄
Now Stable!
After more than two years of development, TanStack Table V9 lays a new foundation for years to come, with a tree-shakable plugin architecture, state management powered by TanStack Store, new features, major performance improvements, hundreds of bug fixes, and more.
TanStack Devtools now officially support rspack! π
And you can also make your trigger a floating bubble which you can move across your screen!
Or just play with it and throw it around!
Want to queue up messages to your LLM provider?
TanStack AI's useChat now queues it and auto-sends when the stream settles.
Pick between FIFO, batch, or interrupt queues. π
We're starting to add adapters for even more JavaScript frameworks and libraries. In the past couple of weeks, we've introduced our first Alpine JS, Marko, and Ember JS adapters.
All thanks to community contributions!
When TanStack Table V9 launches as stable (really soonβ’), it will be offering a record number of adapters.
Preact, Ember, and Alpine are new! And we're also still exploring potential Marko and Qwik V2 adapters.
Writing our libraries in 95%+ vanilla JS makes this maintainable.
Adding voice messages to your AI chat app? ποΈ
useAudioRecorder() β record β get a ready-to-send audio part. That's it.
One API across React, Vue, Solid, Svelte & Angular.
TanStack AI runs coding agents in a sandbox now π¦
Agents: Grok build, Claude Code, Codex, OpenCode, or any ACP agent
Sandboxes: Cloudflare, Docker, Local, Daytona or Vercel sandboxes
Fully configurable sandboxes with skills, mcps and more! π
tanstack.com/blog/run-codingβ¦
MCP tools have only ever returned text for the model to read.
Now they return interactive UI. πͺ
TanStack AI's new MCP Apps support renders ui:// widgets right in the chat - static or fully interactive, with a client/server bridge.
How do you trace an AI image or video call generation?
Same way you trace chat.
You already wrote otelMiddleware() for chat.
Drop the same value into media generation hooks and
get full tracability.
No new API required. π
TanStack AI π€Angular
Today we have released an official @angular adapter for TanStack AI! π
Try it out in your Angular apps today and find the docs here:
tanstack.com/ai/latest/docs/β¦
TanStack AI is now in Beta!
TanStack AI is no longer just a text-generation library with extras bolted on. TanStack AI has first-class dev experience for:
β Text and Streaming Structured Data
β Tool Calls
β Summarization
β Image Generation
β Video Generation
β Audio Generation
β Realtime voice chat
β Middleware
β Lazy tool discovery
β Code Mode
β Host-side MCP
β Orchestration (still experimental)
β Pluggable debug logging
β Isomorphic devtools
All with per-model type-safety that actually matters!
A lot of these improvements were directly inspired by TanStack Form.
Here's the full blog post giving some of the backstory on the development of V9 up until this point: tanstack.com/blog/tanstack-tβ¦
TanStack Table V9 is now in Beta!
A foundational release that introduces:
β State management with TanStack Store - (yes, compatible with React Compiler finally)
β Advanced re-render management with state subscriptions
β Lower memory usage at scale
β Tree-shakable features with type-safety
β Better feature/plugin extensibility
β New utilities for writing re-usable/composable code
β Rewritten devtools and debugging options
TanStack Table V9 is available for the latest versions of React, Preact, Angular, Solid, Vue, Svelte, and Lit.
Read the full blog post down below π
MCP support in TanStack AI is here! π
π΄ One MCP server or many
π΄ Standalone clients or pooled servers
π΄ CLI for typegen
π΄ Managed lifecycle via chat or complete control
Read the full blog post below π
A tool returns an image of a secret number. The model reads it back. ποΈ
TanStack AI now supports multimodal tool results: return an Array<ContentPart> with an image part and the model sees the actual pixels - not a stringified blob.
tanstack.com/ai
Your AI chat vanishes on every page reload? π
TanStack AI just shipped a persistence option: pass one adapter and conversations hydrate on load, save on change, clear on reset. localStorage, IndexedDB, anything.
tanstack.com/ai
One adapter. Every OpenAI-compatible provider. π
openaiCompatible({ baseURL, apiKey, models })
Connects DeepSeek, Kimi, Together, Fireworks, Qwen, local LM Studio/vLLM, and more.
Type-safe model unions, no package per vendor.
tanstack.com/ai π
We just released a new version of TanStack AI! π
In today's release we have added client/server context for tools and middleware.
Plug it into your useChat/chat and have it available, fully typesafe out of the box. π₯
TanStack AI now runs on react-native! π
Stream down to mobile devices with ease using fetch + polyfills or XHR transports!
Try it out with the latest releases!
Your AI no longer forgets its structured outputs π§ β¨
TanStack AI now supports multi-turn structured output that remembers across turns.
Ask for a recipe π
Refine it to vegan π±
Make it cheaper πΈ
Add a salad π₯
And every typed result stays preserved in chat history.π΄β‘
New in TanStack Start: Deferred Hydration.
Choose when selected parts of your app hydrate, like when they become visible, get interacted with, or the browser goes idle.
Your AI stack just stopped being closed. π
TanStack AI now speaks AG-UI client-to-server.
- The same client hits any AG-UI server.
- Any AG-UI client hits a TanStack server.
Zero breaking changes.
Migration guide in the tweet below π
New in TanStack Router: more control over dynamic route params.
Use params.parse to accept or skip a candidate, with priority for overlapping dynamic params.
New in TanStack Start: route-aware CSS inlining.
Inline initial route CSS during SSR to reduce blocking stylesheet requests and improve FCP.
tanstack.com/start/latest/doβ¦
30 seconds of spinners is not great UX. π
@tanstack/ai now streams structured output straight into useChat. Pass a Zod schema, get a typed `partial` and `final` for free. β¨
Blog post link below π
TanStack Devtools just migrated to @OxcProject parser + magic-string!π
The results:
π Per-file transform: 1.65 ms β 0.46 ms
π Full pass: 75.73 ms β 21.25 ms
π Speedup: 3.56Γ
Your AI agents are running blind in prod?
One line of otelMiddleware() and every chat, iteration, and tool call lands in your OTel backend with full GenAI
semconv attributes.
Vendor-neutral. Optional peer dep. Already shipped on @tan_stack ai.
tanstack.com/ai/latest/docs/β¦
Every TanStack package used to share the same generic OG image.
Not anymore. π₯
- 17 libraries. 34 themed cards.
- Brand color per package.
- Rendered on demand.
Live now β tanstack.com π’
Every AI audio provider ships a different SDK.
TanStack AI just unified them.
Gemini Lyria, fal MiniMax, Stable Audio. Same streaming call, any model.
Typed. Tree-shakeable. Hooks for React, Solid, Vue, Svelte.
Try it β tanstack.com/ai
Blog π