Index / Poco · updated Aug 17, 2026
Poco
poco-ai/poco-agent · healthy · rank 24 of 43 by stars
Poco wraps a Claude Code–based agent in a polished web UI with sandboxed execution and team channels. It leans into collaboration and IM integrations while keeping one-click Docker self-hosting.
Facts
Repository
Runtime
Posture
Community
Security breakdown
Composite 85 / 100 · how these are scored
higher is safer
higher is safer
higher is safer
higher is safer
higher is riskier
Model access
Read from the repository, not written by a model · 77 files examined
5 providers · custom endpoint
Pinned models
claude-sonnet-4-20250514 released 2025-05-14 — 15 months old (dated in the model id)
Pin last edited 4 months ago
Evidence
Decision
Why choose Poco over OpenClaw?
Why choose this
- Polished web UI with artifacts/playback views
- Built-in IM integrations and mobile support
- Secure sandboxed runtime with local directory mounting
Tradeoffs
- Heavier resource usage (Python/Next.js) vs leaner OpenClaw
- Less mature ecosystem (fewer stars, newer release)
- Cloud subscription may introduce telemetry tradeoffs
Best fit
- Teams wanting channel-based AI collaboration with sandboxed execution
- Self-hosters needing polished web UI and IM integration
- Users who like Claude Code experience with project management
Avoid if
- Those needing ultra-lightweight runtime on edge devices
- Users avoiding cloud dependency or subscription models
- Environments requiring strict no-network isolation
README and commit history show active development and clear OpenClaw alternative positioning, but provided Reddit/Brave mentions are unrelated, reducing community sentiment confidence.
AI layer reviewed Aug 2, 2026 · how this is written
Star activity
1,350 stars today
Overview
Poco (poco-ai/poco-agent) is an open-source AI agent framework positioned as a safer, more beautiful, and easier-to-use alternative to OpenClaw. It wraps a Claude Code–based agent inside a FastAPI backend and a Next.js web UI, providing a polished collaborative environment with servers, channels, DMs, and reply threads where persistent agents can be mentioned, tasked, and monitored.
Core architecture emphasizes security and productivity: all tasks execute inside isolated containers (sandboxes) with optional local directory mounts for self-hosted deployments, while the frontend offers artifacts preview, execution playback, and light/dark modes. Recent commits show rapid iteration on capability grouping, skill references, shared session mobile views, and workspace streaming via signed routes, indicating a focus on team features and UX refinement.
Compared to OpenClaw’s lobster-way terminal/messaging-centric approach, Poco differentiates with built-in IM integrations (DingTalk, Feishu, Telegram), project management, and background cloud execution. It targets groups wanting a self-hosted or cloud AI coworker with strong sandboxing, but trades ultra-lightweight edge efficiency for a richer full-stack experience.