TanStack AI 1.0 Ships Stable Agent Toolkit for Web Apps

TanStack has shipped version 1.0 of its AI toolkit, stabilizing the core APIs developers use to build chat interfaces, tool-calling agents, and coding-agent workflows across providers and sandboxes. According to the official announcement, the release turns a promise made less than a year ago — no lock-in, ever — into a formal commitment: the core APIs are now stable and ready for production.

The toolkit began life as a single chat() function with four providers. It has since grown into a full stack covering chat, tools, media, MCP, persistence, sandboxes, and agents, with 23 provider adapters and eight UI frameworks, built with contributions from more than 140 people. For teams shipping agent features inside web apps and editors, 1.0 means a portable, framework-agnostic runtime they can rely on without rewriting when they swap model providers or change infrastructure.

What TanStack AI 1.0 includes

The 1.0 release stabilizes 76 packages and formally supports agents, subagents, coding-agent workflows, sandboxes, provider adapters, and computer-use tooling, according to coverage in the AI agent news roundup. New additions in the stable line include portable agent skills, subagents, computer-use tooling, and MCP server support, alongside upgrade guidance for existing 0.x users.

Alongside the stable core, the team introduced a new experimental agent harness for running agents in multiple environments. The harness APIs are explicitly marked experimental and may change in minor releases, so early adopters should treat that surface as a preview rather than a commitment.

The upgrade path is designed to be boring in the best way: the team reports that 1.0 introduces no new concepts, removes every API that 0.x marked as deprecated, and keeps every replacement that already shipped in 0.x. Developers who kept up with deprecation warnings are described as most of the way to a clean upgrade.

Why a portable agent toolkit matters

The deeper story is architectural. AI agents that live inside real products need a runtime that is not welded to one model provider, one framework, or one hosting vendor. TanStack AI’s adapter-based design lets teams swap providers and move agents between sandboxes and environments without re-plumbing the application — a direct answer to the integration tax that hits teams every time a model leaderboard reshuffles or pricing changes.

That portability shows up in how the ecosystem is already treating TanStack’s stack. Netlify has made TanStack Start the default framework for new projects in its Agent Runners AI app builder, with a Vite plugin that emulates Netlify’s serverless and edge environment inside the dev server. When a hosting platform standardizes on a framework for agent-built apps, the runtime underneath becomes shared infrastructure for agents, not just a developer library.

The release also lands in a week of fast-moving agent-infrastructure news. For context on how quickly the tooling layer is being funded and adopted, see how Google’s Agent Compute Engine is giving every AI agent its own computer, and how Anthropic and OpenAI donated agent skills to the Agentic AI Foundation. Stable, open toolkits are the substrate all of that investment builds on.

Two things to know before upgrading

The maintainers flag two practical notes. First, upgrade everything together: a 1.0 client needs a 1.0 server, so stable packages should move in lockstep. The experimental harness packages intentionally stay on 0.x while their APIs settle. Second, the rewritten documentation and quickstart are the recommended first stop — running the 1.0 quickstart in a staging environment and validating provider adapters and sandbox paths is the safest way to confirm the harness behaves before touching production.

For builders deciding between toolkits, independent comparisons note that both TanStack AI and Vercel’s AI SDK now support tool approvals, MCP, and cross-provider use; the differentiator is architecture, with TanStack’s framework-neutral clients and adapter model aimed squarely at teams that refuse to be locked in.

Sources

TanStack AI 1.0: Still Your AI, Still Your Way — TanStack Blog
AI Agent News Today — October 23, 2026 — AI Agent Store
TanStack Start Hits 1.0, and Netlify Makes It the Agent Default — Firerun