Image: Wikimedia Commons — "Crew" (Unsplash), CC0 (illustrative; no license-safe official TanStack product visual exists — the TanStack logo is TanStack's trademark)
TL;DR: TanStack AI 1.0 release, explained: on 23 October 2026 the open-source AI toolkit shipped its first stable release — 23 provider adapters, 11 sandbox providers and eight UI frameworks — plus an experimental agent harness for building your own coding agents. Upgrading from 0.x is one codemod plus a type check.
📌 KEY HIGHLIGHTS
- 1.0 = stable line: core APIs are frozen and production-ready — the first formal API commitment since the December 2025 alpha.
- Alpha → 1.0: providers 4 → 23, UI frameworks up to eight, coding agents from none to five, sandbox providers 11.
- Experimental agent harness: @tanstack/ai-harness ships as 0.x — durable sessions, subagents, and a runCli route to build a terminal coding agent.
- Upgrade: one codemod (body → forwardedProps, conversationId → threadId), move all stable @tanstack/ai* packages together, then fix type-checker errors with the guide.
What happened on 23 October
TanStack — the open-source team behind TanStack Query, Router and Start — announced TanStack AI 1.0 on its official blog on 23 October 2026, tagged "Still Your AI, Still Your Way". Less than a year ago, it notes, TanStack AI was a single chat() function, four providers, and one promise: no lock-in, ever. Since then: beta in June (which won AI Project of the Year at the 2026 Open Source Awards), an architecture-locking RC in August, and sister project TanStack Start hitting 1.0 on 13 October. The announcement is explicit: core APIs are now stable and production-ready; the new agent harness is deliberately experimental.
TanStack AI 1.0 Release, Explained: What's Actually Stable
The 1.0 commitment covers the core toolkit: chat, tools, agents, media generation, MCP, persistence and sandboxes — with 23 provider adapters (plus an openaiCompatible catch-all), eight UI frameworks (React, Vue, Solid, Svelte, Preact, Angular, Octane, Remix, vanilla JS) and 11 sandbox providers. New adapters include Cloudflare Workers AI, Lovable AI Gateway, Azure OpenAI, World Labs Marble, TypeSafe and Ollaya. Also new: subagents in chat()/useChat, portable SKILL.md skills, chat compaction, cross-provider computer use, automatic prompt caching, a unified Files API, bring-your-own-key adapters, sign-in with ChatGPT / GitHub Copilot / OpenRouter / Claude OAuth, createMCPServer for MCP hosts, two-way WebMCP and fakeText() for offline testing. The docs were rewritten for 1.0 with a five-minute quick start — and like Google's open-source TorchTPU, this is another bet that the software layer is where AI lock-in gets broken.
| Area | Alpha (Dec 2025) | 1.0 (23 Oct 2026) |
|---|---|---|
| Provider adapters | 4 | 23, + openaiCompatible for any OpenAI-style API |
| UI frameworks | React, Solid, vanilla JS | Eight — adds Vue, Svelte, Preact, Angular, Octane, Remix |
| Coding agents | None | Claude Code, Codex, OpenCode, Grok Build + any ACP agent |
| Sandbox providers | None | 11 — incl. E2B, Blaxel, Upstash Box, boxd, Railway |
Image: Wikimedia Commons / Tsinkala, CC BY-SA 4.0 (illustrative)
The experimental agent harness — build your own coding agent
The one part of 1.0 that is not stable is @tanstack/ai-harness, shipping as 0.x on purpose: one conversation that stays open, takes new messages while it works, runs other agents, and survives a deploy. Sessions are durable, approval/sign-in waits persist across restarts, and you get typed subagents with budgets, plugins, multi-user access rules and background workers. The headline demo: drop the first-party plugins into a harness, hand it to runCli, and you have a terminal coding agent with approval controls, /undo, todo lists and /usage — able to hand work to Claude Code, Codex, Grok Build or any ACP agent, each in its own sandbox. Two honest caveats: recovery still has limits (in-memory child-agent waits don't resume across hosts), and the permissions plugin is not a sandbox. Against this week's news that GitHub Copilot flips unconfigured features on by default, an open, self-hosted agent runtime is a timely alternative. AI agents also need purpose-built infrastructure — see the daytona 60 million series b 1 billion valuation explained story.
Upgrading from 0.x: what breaks
If you already run TanStack AI, the blog says 1.0 has no new concepts — every deprecated 0.x API already had its replacement shipped. The path: run the official codemod (body → forwardedProps, conversationId → threadId), update all stable @tanstack/ai* packages together, remove the discontinued -react-ui/-solid-ui/-vue-ui packages (rebuild with createChatHook), then fix type-checker errors with the guide. Heads-ups: threadId is now required on both sides, addToolApprovalResponse() is gone, servers must end runs with RUN_FINISHED, and stored chats need the saved-data migration first.
Why it matters for India readers
India's developer economy — agency teams, SaaS startups, freelancers — runs on exactly this kind of tooling. A framework-agnostic, MIT-licensed toolkit where swapping OpenAI for Anthropic (or a local Ollama model) is a one-adapter change lowers the cost of experimenting, and fakeText() means tests run without burning API credits. Pair it with cheap typed decision APIs like OpenAI's Decisions API at $0.10 per million tokens and the economics of shipping AI features keep improving. One caution: treat the harness as what it is — experimental. Build on the stable 1.0 core first.
Should you care?
✅ Yes — if you ship AI features in web apps. 1.0 is a genuine API commitment from a team whose libraries already run in some of the world's largest apps, and the provider-swap story is the best anti-lock-in answer in the ecosystem. ⏳ Hold off on the harness in production — it's 0.x by design; prototype with it, ship on the stable core.
FAQs
What is TanStack AI 1.0?
The first stable release of TanStack's open-source, framework-agnostic AI toolkit for the web (announced 23 October 2026): building blocks for chat, tools, agents, MCP, sandboxes and media, with 23 AI provider adapters and eight UI framework integrations, MIT-licensed.
Is TanStack AI 1.0 stable and production-ready?
The core APIs are stable — 1.0 is a formal commitment not to change them under you. The @tanstack/ai-harness agent runtime is the deliberate exception: experimental 0.x, APIs can still change.
What is the TanStack AI agent harness?
An experimental runtime for long-running agents: durable sessions that survive restarts, persistent approval waits, typed subagents with budgets, plugins, and a route to build your own terminal coding agent via runCli, served over standard AG-UI events.
How do I upgrade from TanStack AI 0.x to 1.0?
Run the official codemod, upgrade all stable @tanstack/ai* packages together, remove the discontinued -ui packages, then fix type-checker errors with the upgrade guide. Add threadId everywhere, replace addToolApprovalResponse(), and run the saved-data migration before deploying.
Takeaway: TanStack AI 1.0 turns a year of alphas and betas into something you can build a business on — stable, portable, aggressively anti-lock-in. The experimental harness is the part to watch: a credible open route to self-hosted coding agents.
Follow World Tech on TechByCountry for the next hourly update.
📸 Get tech in the last 24 hours on Instagram — follow @_techinlast24hrs for daily posts (9 AM–9 PM) and reels at 12:30 PM & 8:30 PM.
Sources
- TanStack Blog — "TanStack AI 1.0: Still Your AI, Still Your Way" (official announcement, 23 Oct 2026)
- AI Agent Store — AI Agent News Today, October 23, 2026 (independent confirmation of the 1.0 release)
- Wikimedia Commons — "Programmer at work (Unsplash)" by Crew, CC0 (featured image)
- Wikimedia Commons — "Software Developer at Work" by Tsinkala, CC BY-SA 4.0 (in-article image)
.jpg){kind=link}
Comments
Post a Comment