The Pulse
TanStack AI 1.0 Stabilizes Core APIs, Keeps Agent Harness Experimental
TanStack AI 1.0 makes its core APIs stable while leaving its new agent harness experimental. The release adds features and integrations, and its upgrade instructions distinguish stable packages from experimental ones.
AI.info Team · Written with AI · fact-checked ·
“The core APIs are stable and ready for production.”
Alem Tuzlak, Jack Herrington, Tom Beckenham and Tanner Linsley, authors of the TanStack announcement
Core APIs reach 1.0
TanStack announced version 1.0 of its AI toolkit on October 23. The release formalizes a commitment to the APIs developers build on, while the new agent harness remains experimental. The project’s central pitch is that developers can choose their model provider, application framework, sandbox and communication protocol.
TanStack AI has grown from an alpha with a chat function and four providers into a toolkit for chat, tools, media, Model Context Protocol (MCP), persistence, sandboxes and agents. The announcement lists 23 provider adapters, plus an openaiCompatible adapter for OpenAI-style APIs. It names React, Vue, Solid, Svelte, Preact, Angular, Octane, Remix and vanilla JavaScript among its UI options, and lists 11 sandbox providers.
The release also includes coding-agent integrations for Claude Code, Codex, OpenCode, Grok Build and agents that use the Agent Client Protocol. TanStack says more than 140 people helped build the toolkit. The announcement describes the 1.0 release as having no new concepts for existing users: APIs marked deprecated in 0.x have been removed, and their replacements had already shipped in 0.x.
Features added since the release candidate
TanStack lists several additions made after its August 2026 release candidate. Developers can hand turns in chat() or useChat to named child agents, such as a researcher or writer, with typed input and limits. Other agent features include portable Agent Skills, which let tool-calling models load SKILL.md files, and computer use that works with OpenAI and Anthropic.
The release also adds decide() for typed choices, scores and boolean decisions; compaction to help keep long chats within context limits; and chat options for reasoning and retries. The announcement describes automatic prompt caching, support for changing tools or system prompts mid-conversation without losing the cache, and history that can replay across models. A Files API provides one interface to each provider’s files.
Other additions include adapters for Cloudflare Workers AI and AI Gateway, Lovable AI Gateway, Azure OpenAI, World Labs Marble, TypeSafe and Ollaya. TanStack also notes Remix 3 support, Svelte devtools, and five sandbox providers: Blaxel, Upstash Box, E2B, boxd and Railway. Its end-to-end tests run on each pull request against deterministic model fixtures, without live API calls.
Upgrades distinguish stable and experimental packages
Teams upgrading from 0.x should update their stable @tanstack/ai* packages to 1.0 together; a 1.0 client needs a 1.0 server. The experimental harness packages remain on 0.x, so the instruction does not apply to every package matching that naming pattern. TanStack provides a codemod for common renames, including changing body to forwardedProps and conversationId to threadId.
The new API requires a thread ID in both the browser and on the server; neither side creates one automatically. Developers who use @tanstack/ai-react-ui, @tanstack/ai-solid-ui or @tanstack/ai-vue-ui must remove those packages, which will not have 1.0 releases. Teams keeping chats, runs or files created with 0.x should follow TanStack’s saved-data migration steps before deploying.
The agent harness remains experimental
TanStack says @tanstack/ai-harness and @tanstack/ai-harness-cli ship as 0.x. Harness entry points in other packages, including @tanstack/ai-skills/harness and @tanstack/ai-acp/agent, are experimental too and may change in a minor release. The project says it wants developers to try the harness before its APIs are locked down; the harness is separate from the stable 1.0 core.
That distinction defines the practical scope of the announcement: developers can build against the stable core APIs while treating harness interfaces as unfinished. For existing applications, the release’s upgrade guidance centers on matching stable client and server versions, removing the discontinued UI packages if used, and migrating saved data when necessary.