WUPHF is an open-source multi-agent orchestration platform built in Go that creates a collaborative 'office' for AI agents (Claude Code, Codex, OpenClaw). Agents share a persistent knowledge base via a git-backed wiki with typed fact graphs, per-agent notebooks, and promotion flows. The system uses fresh sessions per turn with 97% prompt cache hit rates, push-driven agent wakes (zero idle burn), and per-role MCP tool scoping (4 tools in DM vs 27 in full office). It ships with a web UI, Telegram bridge, OpenClaw bridge, and two action providers (local CLI and Composio). Benchmarks show flat ~87k input tokens per turn vs accumulated-session orchestrators that grow from 124k to 484k over 8 turns.
Table of contents
Get StartedSetup prompt (for AI agents)OptionsMemory: Notebooks and the WikiOther CommandsWhat You Should SeeTelegram BridgeOpenClaw BridgeExternal ActionsWhy WUPHFBenchmarkClaim StatusEvaluate This RepoWatch the wiki write itselfThe NameSort: