AI/TLDR

nex-crm · 2026-04-25 · notable

Wuphf: Multi-Agent Workspace Where Agents Maintain a Shared Git Wiki

Open-source multi-agent workspace where Claude, Codex, and OpenClaw share a git-backed wiki. Fresh sessions per turn prevent context bloat; 97% Claude API cache hit rate. Install with npx wuphf. Hit 164 HN points on Show HN today.

Wuphf GitHub repository — multi-agent AI workspace with shared git-backed wiki

One npx command launches a shared workspace where multiple AI agents collaborate through a git-backed wiki, with 97% prompt-cache hit rates keeping per-session costs low.

What is it?

Wuphf is an open-source, self-hosted multi-agent collaboration platform. Run `npx wuphf` and get a web UI at localhost:7891 where multiple AI agents — Claude, Codex, and OpenClaw — work together through a shared workspace. Each agent gets a scoped toolset and personal notebook; all agents share a team-wide wiki stored in git-tracked markdown files under ~/.wuphf/wiki/. The project describes itself as 'Slack for AI employees with a shared brain.'

How does it work?

Agents are push-activated: each gets a fresh session per turn, preventing state accumulation across the shared context. The git-native wiki means every agent can read and write knowledge and every change has a full diff history. A 97% Claude API prompt-cache hit rate is achieved because each new session only re-sends the delta (new wiki changes) rather than the full accumulated history. Tool scoping limits each agent to only the capabilities it needs. Action providers (One CLI and Composio) connect agents to external systems. An optional Telegram integration makes the workspace accessible via messaging.

Why does it matter?

Most multi-agent frameworks share a single growing context window, which leads to token bloat and cost spirals in long-running workflows. Wuphf separates persistent knowledge (git wiki) from per-turn computation (fresh sessions), so you pay for one context load and cache-hit for dozens of subsequent turns. The 97% cache rate is a practical production metric that makes continuous multi-agent operation economically viable.

Who is it for?

Developers building multi-agent AI systems; teams experimenting with autonomous AI workflows

Try it

npx wuphf

Sources · 2 outlets

Tags

  • multi-agent
  • open-source
  • claude
  • codex
  • shared-memory
  • git-native
  • agent-workspace
  • prompt-caching
  • mit-license
  • show-hn

← All releases · Learn AI