Show HN: Start Using Claude Managed Agents Today – Posse
hackernews
|
|
📦 오픈소스
#ai 딜
#ai 에이전트
#anthropic
#managed agents
#web ui
#오픈소스
원문 출처: hackernews · Genesis Park에서 요약 및 분석
요약
Anthropic의 관형 에이전트는 강력한 인프라를 제공하지만 사용자 인터페이스가 없어 CLI에 의존해야 하는 문제를 해결하기 위해 Posse가 개발되었습니다. 이 경량 웹 앱은 브라우저에서 에이전트 생성, 세션 관리, 메모리 편집 및 다중 에이전트 실행을 가능하게 하며, 백엔드 의존성 없이 Next.js로 구현되었습니다. 사용자는 API 키를 서버 측에 안전하게 보관하며 실시간으로 코드 실행과 도구 호출 과정을 시각적으로 확인할 수 있습니다. 현재 핵심 기능이 구현되어 있으며, 향후 스트리밍 개선과 파일 업로드 기능 추가가 계획되어 있습니다.
본문
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. - 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 docker run -p 3000:3000 \ -e ANTHROPIC_API_KEY=sk-ant-... \ ghcr.io/oguzbilgic/posse:latest Open http://localhost:3000. git clone https://github.com/oguzbilgic/posse.git cd posse echo "ANTHROPIC_API_KEY=sk-ant-..." > .env.local npm install && npm run dev 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. - Next.js 16 + React 19 - TypeScript + Tailwind CSS - Zustand for client state - Marked + highlight.js for markdown/code rendering 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. 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 MIT
Genesis Park 편집팀이 AI를 활용하여 작성한 분석입니다. 원문은 출처 링크를 통해 확인할 수 있습니다.
공유