posse

A web UI for Anthropic Managed Agents.

Anthropic ships the best agent infrastructure — sandboxed environments, persistent memory, tool use, multi-agent orchestration — but no UI. You're stuck with curl and the API console.

Posse gives you the missing interface: create agents, run sessions, manage memory stores, and watch your agents work — all from a browser.

What you get

• Agent management — Create, edit, archive, and version agents with different models (Sonnet, Opus, Haiku)
• Sessions — Start sessions, send messages, watch tool calls and code execution in real time
• Memory stores — Browse and edit persistent memory that agents carry across sessions
• Environments — Switch between execution environments with different tools and configurations
• Multi-agent — Run multiple agents side by side, each with their own sessions and memory

Quick start

docker run -p 3000:3000 \
  -e ANTHROPIC_API_KEY=sk-ant-... \
  ghcr.io/oguzbilgic/posse:latest

Open http://localhost:3000.

From source

git clone https://github.com/oguzbilgic/posse.git
cd posse
echo "ANTHROPIC_API_KEY=sk-ant-..." > .env.local
npm install && npm run dev

How it works

Posse is a lightweight Next.js app that proxies requests to the Anthropic API. Your API key stays server-side — never exposed to the browser.

The entire app is ~13k lines of TypeScript with zero backend dependencies beyond Next.js. No database, no auth layer, no state server. It reads everything from the Anthropic API and renders it.

Stack

• Next.js 16 + React 19
• TypeScript + Tailwind CSS
• Zustand for client state
• Marked + highlight.js for markdown/code rendering

Why

Anthropic's managed agents are powerful but invisible. You define agents in JSON, create sessions via API, and parse events from streams. It works, but you can't see what's happening.

With posse, you get a real workspace: watch an agent write code, see it execute bash commands, check what it stored in memory, start a new session with a different environment. The kind of feedback loop you need when building with agents.

Status

Early. Shipping fast. Core features work — agents, sessions, chat, memory, environments. What's next:

• SSE streaming (currently polling)
• File upload / media attachments
• Session event filtering and search
• Agent skill configuration UI
• Callable agent orchestration view

License

MIT