@tavon-ai/pi-ai-sdk-bridge

Third-party bridge from @earendil-works/pi-coding-agent AgentSession streams to Vercel AI SDK v6 useChat UI message streams.

Use rich AI SDK/AI Elements chat UIs while running Pi as the agent harness.

Start the bridge

npx @tavon-ai/pi-ai-sdk-bridge --cwd /path/to/project --port 3001

This starts an AI SDK-compatible chat endpoint at:

http://127.0.0.1:3001/api/chat

Options:

• --cwd <dir> — workspace directory Pi should operate in (default: current directory).
• --port, -p <port> — listen port (default: 3001).
• --host <host> — bind hostname (default: 127.0.0.1).
• --tools, -t <list> — comma-separated Pi tool allowlist.
• --base-path <path> — chat API base path (default: /api/chat).

Programmatic usage

import { serve } from "@hono/node-server";
import { createPiChatHandler } from "@tavon-ai/pi-ai-sdk-bridge";

const handler = createPiChatHandler({ cwd: "/path/to/project", basePath: "/api/chat" });

serve({
  port: 3001,
  hostname: "127.0.0.1",
  fetch: (request) => handler(request),
});

Security

Pi tools can execute code on the bridge host. Run this bridge inside your container/VM/workspace isolation boundary and protect it with auth before exposing it beyond localhost.