April 25, 2026AgentsOpen SourceFramework

WUPHF: a shared office for your AI employees

There's a new way to think about multi-agent setups. Not API calls. Not orchestration YAML. A chatroom.

WUPHF is a Show HN that hit the front page today. It's an open-source workspace where Claude Code, Codex, and OpenClaw all log in like coworkers, get their own filesystems, and stream their work into a shared web UI at localhost:7891. You DM an agent mid-task, redirect it, see exactly what it's doing in real time. No more black-box agent runs that come back broken hours later.

The interesting bit isn't the chatroom. It's the wiki. WUPHF gives every agent a private notebook and a shared wiki, both backed by plain Markdown files in git. Agents jot observations privately. When something durable shows up, an archivist LLM promotes it to the shared wiki, synthesizes Wikipedia-style briefs, commits to ~/.wuphf/wiki/. You get cited lookups with /lookup and contradiction detection with /lint. The whole memory system is cat, grep, git log. No vector DB, no cloud, no API keys.

This is the Karpathy thesis playing out in code: agents should write knowledge in the format humans already use, not in some agent-only embedding store. If your team has tried and failed to get persistent agent memory working, this is the cleanest reference implementation we've seen this month.

https://github.com/nex-crm/wuphf
← Previous
Ops Log: April 25, 2026
Next β†’
Stash gives any MCP agent the memory Claude.ai has
← Back to all articles

Comments

Loading...
>_