# Show HN： 由您的代理维护的卡帕西风格大语言模型维基（支持 Markdown 和 Git） - 来源：Hacker News 热门（buzzing.cc 中文翻译） - 作者：najmuzzaman - 发布时间：2026-04-25 20:12 - AIHOT 分数：46 - AIHOT 链接：https://aihot.virxact.com/items/cmoebxezr00bpslxghebu4vfe - 原文链接：https://github.com/nex-crm/wuphf ## AI 摘要 项目“Wuphf”是一个受卡帕西风格启发的大语言模型维基系统，支持Markdown格式和Git版本控制。其核心特点是可由用户设定的AI代理自动维护和更新内容，实现了知识库的持续自主管理。该项目已在GitHub开源，并在Hacker News上获得了100点热度。 ## 正文 WUPHF (pronounced "woof") Slack for AI employees with a shared brain. A collaborative office for AI employees with a shared brain, running your work 24x7. One command. One shared office. CEO, PM, engineers, designer, CMO, CRO — all visible, arguing, claiming tasks, and shipping work instead of disappearing behind an API. Unlike the original WUPHF.com, this one works. "WUPHF. When you type it in, it contacts someone via phone, text, email, IM, Facebook, Twitter, and then... WUPHF." — Ryan Howard, Season 7 "WUPHF. When you type it in, it contacts someone via phone, text, email, IM, Facebook, Twitter, and then... WUPHF." — Ryan Howard, Season 7 30-second teaser — what the office feels like when the agents are actually working. 30-second teaser — what the office feels like when the agents are actually working. Full walkthrough — launch to first shipped task, end to end. Full walkthrough — launch to first shipped task, end to end. Get Started Prerequisites: one agent CLI — Claude Code by default, or Codex CLI when you pass --provider codex. tmux is required for --legacy-tui mode (the web UI runs agents headlessly by default; tmux-backed dispatch remains as an internal fallback). --provider codex --legacy-tui npx wuphf That's it. The browser opens automatically and you're in the office. Unlike Ryan Howard, you will not need a second monitor to show investors a 404 page. Prefer a global install? npm install -g wuphf && wuphf Building from source (requires Go and Bun): git clone https://github.com/nex-crm/wuphf.git cd wuphf cd web bun install bun run build cd .. go build -o wuphf ./cmd/wuphf ./wuphf Forking this? See FORKING.md for running WUPHF without Nex, swapping branding, and adding your own agent packs. For the internals, see ARCHITECTURE.md. Forking this? See FORKING.md for running WUPHF without Nex, swapping branding, and adding your own agent packs. For the internals, see ARCHITECTURE.md. Stability: pre-1.0. main moves daily. Pin your fork to a release tag, not main. Stability: pre-1.0. main moves daily. Pin your fork to a release tag, not main. main main Setup prompt (for AI agents) Paste this into Claude Code, Codex, or Cursor and let your agent drive the install: Set up https://github.com/nex-crm/wuphf for me. Read `README.md` first, then run `npx wuphf` to install and launch the office — the web UI opens at http://localhost:7891 automatically. Once it is running, confirm you can see #general, the team of agents, and the composer. If anything looks off, re-read the "What You Should See" and "Other Commands" sections of README.md before retrying — do not skip to flags. For agent conventions and the MCP tools each teammate gets, read `AGENTS.md`. For internals and how the broker, workspaces, and memory fit together, read `ARCHITECTURE.md`. If I mention forking or rebranding, read `FORKING.md`. After it is installed and running, open https://github.com/nex-crm/wuphf in my browser. If I am logged in to GitHub, ask me whether you should star the repo for me as a quick demo that the browser interaction works — only click the star if I say yes. If I am not logged in, just open https://wuphf.team. Set up https://github.com/nex-crm/wuphf for me. Read `README.md` first, then run `npx wuphf` to install and launch the office — the web UI opens at http://localhost:7891 automatically. Once it is running, confirm you can see #general, the team of agents, and the composer. If anything looks off, re-read the "What You Should See" and "Other Commands" sections of README.md before retrying — do not skip to flags. For agent conventions and the MCP tools each teammate gets, read `AGENTS.md`. For internals and how the broker, workspaces, and memory fit together, read `ARCHITECTURE.md`. If I mention forking or rebranding, read `FORKING.md`. After it is installed and running, open https://github.com/nex-crm/wuphf in my browser. If I am logged in to GitHub, ask me whether you should star the repo for me as a quick demo that the browser interaction works — only click the star if I say yes. If I am not logged in, just open https://wuphf.team. Options Flag What it does --memory-backend Override the built-in markdown wiki (nex, gbrain, none) — legacy installs only --no-nex Skip the Nex backend (no context graph, no Nex-managed integrations) --legacy-tui Use the legacy tmux TUI instead of the web UI --no-open Don't auto-open the browser --pack Pick an agent pack (starter, founding-team, coding-team, lead-gen-agency, revops) --opus-ceo Upgrade CEO from Sonnet to Opus --provider LLM provider override (claude-code, codex, opencode, ollama, hermes-agent, openclaw-http) --collab Start in collaborative mode — all agents see all messages (this is the default) --unsafe Bypass agent permission checks (local dev only) --web-port Change the web UI port (default 7891) --workspace Use a specific workspace for one command (does not change the active workspace) --memory-backend nex gbrain none --no-nex --legacy-tui --no-open --pack starter founding-team coding-team lead-gen-agency revops --opus-ceo --provider claude-code codex opencode ollama hermes-agent openclaw-http --collab --unsafe --web-port --workspace --legacy-tui is deprecated, slated for removal, and retained only while the desktop replacement lands. --legacy-tui Opencode and custom endpoints --provider opencode shells out to the opencode CLI binary. WUPHF does not own that provider's HTTP path, and provider_endpoints.opencode.base_url is not consulted. --provider opencode opencode provider_endpoints.opencode.base_url For custom OpenAI-compatible endpoints such as LiteLLM, OmniRoute, or local proxies, use --provider ollama and set WUPHF_OLLAMA_BASE_URL or provider_endpoints.ollama.base_url: --provider ollama WUPHF_OLLAMA_BASE_URL provider_endpoints.ollama.base_url WUPHF_OLLAMA_BASE_URL="http://127.0.0.1:20128/v1" \ WUPHF_OLLAMA_MODEL="openai/gpt-5.4-mini" \ wuphf --provider ollama --memory-backend none --no-open --no-nex still lets Telegram and any other local integration keep working. To switch back to CEO-routed delegation after launch, use /focus inside the office. --no-nex /focus Memory: Notebooks and the Wiki WUPHF ships with built-in memory. No backend choice, no API key, no setup step in the wizard. Every agent gets its own notebook, and the team shares a wiki — a local git repo of markdown articles at ~/.wuphf/wiki/. cat, grep, git log, and git clone all work. ~/.wuphf/wiki/ cat grep git log git clone The promotion flow: An agent works on a task and writes raw context, observations, and tentative conclusions to its notebook (per-agent, scoped, local to WUPHF). When something in the notebook looks durable (a recurring playbook, a verified entity fact, a confirmed preference), the agent gets a promotion hint. The agent promotes it to the wiki. Now every other agent can query it. The wiki points other agents at whoever last recorded the context, so they know who to @mention for fresher working detail. Nothing is promoted automatically. Agents decide what graduates from notebook to wiki. The wiki is not just a markdown folder. It is a living knowledge graph: typed facts with triplets, per-entity append-only fact logs, LLM-synthesized briefs committed under the archivist identity, /lookup cited-answer retrieval, and a /lint suite that flags contradictions, orphans, stale claims, and broken cross-references. The web UI gives you a Wikipedia-style reading view, a rich editor with WUPHF-specific inserts, and an AI-assisted maintenance assistant. See DESIGN-WIKI.md for the reading view and docs/specs/WIKI-SCHEMA.md for the operational contract. archivist /lookup /lint Onboarding seeds the wiki for you. The wizard optionally scans your website and any files you point it at, then writes a starter set of company-context articles (about, owner, products) before the first agent turn fires. Your team starts already knowing who you are and what you ship. Legacy backends. Existing installs on Nex or GBrain keep working — backend selection is sticky in config.json and there is no forced migration. The CLI flag stays available for power users and for moving off legacy backends: config.json wuphf --memory-backend nex # hosted Nex graph + WUPHF-managed integrations wuphf --memory-backend gbrain # local Postgres-backed graph wuphf --memory-backend none # no shared wiki; notebooks still work The web wizard no longer surfaces this as a choice. Markdown is the default and the only path for fresh installs. Internal naming (for code spelunkers): the notebook is private memory, the wiki is shared memory. On the built-in markdown backend the MCP tools are notebook_write | notebook_read | notebook_list | notebook_search | notebook_promote | team_wiki_read | team_wiki_search | team_wiki_list | team_wiki_write | wuphf_wiki_lookup | run_lint | resolve_contradiction. On nex/gbrain the MCP tools are the legacy team_memory_query | team_memory_write | team_memory_promote. The two tool sets never coexist on one server instance — backend selection flips the surface. private shared notebook_write | notebook_read | notebook_list | notebook_search | notebook_promote | team_wiki_read | team_wiki_search | team_wiki_list | team_wiki_write | wuphf_wiki_lookup | run_lint | resolve_contradiction nex gbrain team_memory_query | team_memory_write | team_memory_promote Other Commands The examples below assume wuphf is on your PATH. If you just built the binary and haven't moved it, prefix with ./ (as in Get Started above) or run go install ./cmd/wuphf to drop it in $GOPATH/bin. wuphf PATH ./ go install ./cmd/wuphf $GOPATH/bin wuphf init # First-time setup wuphf share # Invite one team member over Tailscale/WireGuard wuphf shred # Delete workspace state and reopen onboarding wuphf workspace list # Run multiple isolated offices side by side wuphf workspace switch # Flip the active workspace wuphf --1o1 # 1:1 with the CEO wuphf --1o1 cro # 1:1 with a specific agent Share With a Team Member Two ways to invite a teammate. Pick the one that fits your network. Private network — Tailscale or WireGuard. Both machines on the same private mesh. The invite never leaves the network and no public interface is exposed: wuphf share Or click "Create invite" on the Health Check tile inside the office to mint one without leaving the browser. Send the printed /join URL to your teammate. The invite is one use, expires after 24 hours, and the shared web listener only binds to a private-network address by default. /join Public tunnel — no shared network needed. Click "Start tunnel" on the Health Check tile and WUPHF spins up a Cloudflare quick tunnel. The trycloudflare URL is paired with a 6-digit passcode the joiner has to type before they can land in the office; the join handler is rate-limited per source IP so a leaked URL alone cannot be brute-forced. cloudflared is bundled with the npm install (verified against a pinned SHA256 per platform) so the button works on first launch with zero extra setup. cloudflared The tunnel path is opt-in and shown behind a confirmation dialog with the usual disclaimers (URL exposure, channel hygiene, invite-token semantics, TLS). Public LAN binds on the network-share path remain blocked unless you pass --unsafe-lan. --unsafe-lan For the full walkthrough, see Share WUPHF With a Team Member. Publishing skills Once a team-authored skill exists at team/skills/.md, you can publish it to the public agent-skill commons or pull a community skill back into your wiki. Publish opens a real PR via gh; install fetches a public raw SKILL.md and installs it as an active skill in the local team wiki. team/skills/.md gh SKILL.md # Publish your team's deploy skill to the Anthropic skills marketplace wuphf skills publish deploy-frontend --to anthropics # Dry-run the same publish to inspect the manifest + PR body without opening the PR wuphf skills publish deploy-frontend --to anthropics --dry-run # Publish to a custom GitHub repo (optionally pinning a non-main branch) wuphf skills publish deploy-frontend --to github:nex-crm/wuphf-skills wuphf skills publish deploy-frontend --to github:nex-crm/wuphf-skills@master # Pull a community skill into your team's wiki wuphf skills install web-research --from anthropics Supported hubs: anthropics, lobehub, or any github:owner/repo[@branch]. Custom GitHub hubs default to main unless a branch is specified. Publish requires gh auth login first; install only needs network access since it fetches public raw URLs. anthropics lobehub github:owner/repo[@branch] main gh auth login What You Should See A browser tab at localhost:7891 with the office localhost:7891 #general as the shared channel #general The team visible and working A composer to send messages and slash commands If it feels like a hidden agent loop, something is wrong. If it feels like The Office, you're exactly where you need to be. Telegram Bridge WUPHF can bridge to Telegram. Run /connect inside the office, pick Telegram, paste your bot token from @BotFather, and select a group or DM. Messages flow both ways. /connect OpenClaw Bridge Already running OpenClaw agents? You can bring them into the WUPHF office. Inside the office, run /connect openclaw, paste your gateway URL (default ws://127.0.0.1:18789) and the gateway.auth.token from your ~/.openclaw/openclaw.json, then pick which sessions to bridge. Each becomes a first-class office member you can @mention. OpenClaw agents keep running in their own sandbox; WUPHF just gives them a shared office to collaborate in. /connect openclaw ws://127.0.0.1:18789 gateway.auth.token ~/.openclaw/openclaw.json @mention WUPHF authenticates to the gateway using an Ed25519 keypair (persisted at ~/.wuphf/openclaw/identity.json, 0600), signed against the server-issued nonce during every connect. OpenClaw grants zero scopes to token-only clients, so device pairing is mandatory — on loopback the gateway approves silently on first use. ~/.wuphf/openclaw/identity.json If you want WUPHF-created office members to run through OpenClaw instead of bridging pre-existing OpenClaw sessions, enable OpenClaw Gateway's OpenAI-compatible Chat Completions endpoint (gateway.http.endpoints.chatCompletions.enabled = true) and use --provider openclaw-http. The default endpoint is http://127.0.0.1:18789/v1 and the default model target is openclaw/default; override them with WUPHF_OPENCLAW_HTTP_BASE_URL / WUPHF_OPENCLAW_HTTP_MODEL or provider_endpoints.openclaw-http. gateway.http.endpoints.chatCompletions.enabled = true --provider openclaw-http http://127.0.0.1:18789/v1 openclaw/default WUPHF_OPENCLAW_HTTP_BASE_URL WUPHF_OPENCLAW_HTTP_MODEL provider_endpoints.openclaw-http For token-authenticated gateways, WUPHF sends Authorization: Bearer ... using WUPHF_OPENCLAW_HTTP_API_KEY, OPENCLAW_GATEWAY_TOKEN, WUPHF_OPENCLAW_TOKEN, or the saved OpenClaw token from Settings, in that order. Requests include a stable OpenAI user value derived from the WUPHF agent slug so OpenClaw can reuse the same per-agent session across turns. Authorization: Bearer ... WUPHF_OPENCLAW_HTTP_API_KEY OPENCLAW_GATEWAY_TOKEN WUPHF_OPENCLAW_TOKEN user Hermes Agent Runtime Already running Hermes Agent? Point WUPHF agents at its local OpenAI-compatible API server with --provider hermes-agent or set llm_provider to hermes-agent in config. The default endpoint is http://127.0.0.1:8642/v1 and the default model name is hermes-agent; override them with WUPHF_HERMES_AGENT_BASE_URL / WUPHF_HERMES_AGENT_MODEL or provider_endpoints.hermes-agent. --provider hermes-agent llm_provider hermes-agent http://127.0.0.1:8642/v1 hermes-agent WUPHF_HERMES_AGENT_BASE_URL WUPHF_HERMES_AGENT_MODEL provider_endpoints.hermes-agent If your Hermes API server uses API_SERVER_KEY, export the same value as WUPHF_HERMES_AGENT_API_KEY before starting WUPHF. Authenticated requests get stable X-Hermes-Session-* headers per WUPHF agent slug so each office member keeps its own Hermes-side session. API_SERVER_KEY WUPHF_HERMES_AGENT_API_KEY X-Hermes-Session-* Want to add a new integration? See docs/ADD-A-TRANSPORT.md. External Actions To let agents take real actions (send emails, update CRMs, etc.), WUPHF ships with two action providers. Pick whichever fits your style. One CLI — default, local-first Uses a local CLI binary to execute actions on your machine. Good if you want everything running locally and don't want to send credentials to a third party. /config set action_provider one /config set action_provider one Composio — cloud-hosted Connects SaaS accounts (Gmail, Slack, etc.) through Composio's hosted OAuth flows. Good if you'd rather not manage local CLI auth. Create a Composio project and generate an API key. Connect the accounts you want (Gmail, Slack, etc.). Inside the office: /config set composio_api_key /config set action_provider composio