Implements provider-agnostic AI streaming, tool calling, and structured output with TanStack AI.
npx skills add https://github.com/firstaxel/neon --skill tanstack-ai-firstaxel
Please help me install this Agent Skill. Skill: tanstack-ai Source: https://github.com/firstaxel/neon/tree/main/.agents/skills/tanstack-ai Command: npx skills add https://github.com/firstaxel/neon --skill tanstack-ai-firstaxel
generate() API with tree-shakeable adapters for OpenAI, Anthropic, Gemini, and Ollama, with runtime adapter switching. - Streaming & Structured Output: Stream responses with async iteration and enforce Zod-schema-typed structured output, including partial JSON parsing during streaming. - Tool Calling & Agentic Loops: Define tools with approval workflows and bounded agent loops via maxIterations. - Framework Hooks: Build chat and completion UIs with useChat/useCompletion for React, createChat for Solid, or a framework-agnostic vanilla client. - Use Case: Build a TanStack Start chat endpoint that streams GPT-4o responses through a server function, keeps API keys server-side, and renders the conversation with useChat in a React component. ## Quick Start Ask the AI to create a streaming chat endpoint using @tanstack/ai with the OpenAI adapter and wire it to a React useChat component.High-intent search queries and answers about installing and using this skill.
Call generate() with an adapter and messages, then iterate the result with for await to process each chunk. You can also use createReplayStream to serve chunks as Server-Sent Events from an endpoint.
Import the matching adapter such as openaiText or anthropicText and pass it to generate(). Because the API is provider-agnostic, you can select the adapter at runtime based on an environment variable without changing other code.
Yes, convert a Zod schema with convertZodToJsonSchema and pass it as the schema option to generate(). The result is typed as z.infer of your schema, and parsePartialJson handles incomplete JSON during streaming.
Yes, use @tanstack/ai-client for a framework-agnostic vanilla client with subscribe, send, and clear methods, or @tanstack/ai-solid for Solid.js. PHP and Python packages are also available.
Set the maxIterations option on generate() to cap agent loop iterations. For destructive tools, set requiresApproval: true and handle confirmation through ToolCallManager's onApprovalRequired callback.
Client-side code exposes keys to anyone inspecting the browser bundle. The recommended pattern is calling generate() inside a TanStack Start server function so credentials stay on the server.
NovaAI-innovationFetch current model names from AI providers and classify them into performance tiers.
Growth4U-systemsAutomate metric plan setup, analytics tool connections, and dashboard generation.
MichaelFriezeCentralize TanStack Query, Router, and AI debugging in a plugin-based devtools panel.
evanfang0054Integrate OpenAI, Anthropic, Google GenAI, Azure OpenAI, and AWS Bedrock chat models in LangChain.js.