vercel-labs/py-ai
3 skills68 installs210 starsGitHub
Install
npx skills add https://github.com/vercel-labs/py-aiSkills in this repo
1Vercel Python Servicesvercel-python-services documents how indie builders deploy Python backends on Vercel using the experimentalServices API, often paired with a JavaScript frontend in one repo. The skill emphasizes correct project shape: a single root vercel.json, per-service dependency folders, and backend route definitions that omit the `/api` prefix because the platform removes it before the request hits FastAPI or another framework. Local development uses `vercel dev -L` from the project root so every service shares port 3000 and consistent routing—frontends should call `/api/...` without embedding backend hostnames. Validation mistakes are costly: each service entry needs entrypoint and routePrefix, and extra unknown keys can break preview deploys. The reference flow targets Python 3.12 with FastAPI and permissive CORS for iterative full-stack work. Use when you are actively scaffolding or debugging this split-stack layout on Vercel, not for generic Python scripting off-platform.28installs2Vercel Ai SdkVercel AI SDK is an agent skill for the Python vercel-ai-sdk package, aimed at solo builders who want Claude or other models wired into apps with streaming, tools, hooks, and structured output without relearning the runtime model each session. It documents the core contract: ai.run() owns the Runtime in a context var, starts your root coroutine, drains an internal step queue, and yields Message objects—everything else must execute inside that boundary. The skill walks through decorating async functions with @ai.tool, optionally injecting ai.Runtime parameters excluded from the LLM schema, and composing stream_loop with make_messages for agent loops. GatewayModel examples show how to point at provider routes like anthropic/claude-opus-4.6. Use it during Build when you are shipping an agent backend, CLI assistant, or SaaS feature that needs dependable tool execution rather than one-shot chat completions. It is procedural knowledge for Cursor, Claude Code, and Codex when generating Python agent code, not a hosted inference product itself.25installs3AiThe ai skill documents the Python `ai` SDK from Vercel Labs—the same family of concepts as the TypeScript AI SDK, packaged for async Python agents. Solo builders shipping Claude, Anthropic, or gateway-routed models in APIs or side projects use it to pick providers, register tools, run multi-turn agents, consume stream events, and add structured or media outputs without reverse-engineering the repo. It assumes you are already in active implementation: you install with uv, optionally add provider extras, and follow copy-paste patterns for `get_model`, message helpers, and `agent.run` loops. It matters because agent features are easy to get wrong (stream lifecycle, tool schemas, default model env vars); this skill compresses official usage into agent-ready procedural knowledge for Cursor, Claude Code, and Codex sessions focused on backend agent code.15installs