AI Decision System
Replayable, offline-verifiable records for AI-assisted go/no-go decisions: experiment promotions, typed model answers and human-approved actions.
翻譯準備中
尚未偵測到更新。GitHub 發布版本後會顯示在這裡。
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
在專案資料夾中執行 /nomorevibe verify,即可驗證所有權並自行管理介紹。
一分鐘完成驗證您的自托管 AI 行銷部門——一個 CMO 代理,將工作委派給內容、社群、電子郵件、SEO 與分析專家。開源、先審批、以檔案為基礎。基於 OpenClaw 構建。
AI 翻譯
Your self-hosted AI marketing department — a CMO agent that delegates to specialists for content, social, email, SEO, and analytics. Open-source, approval-first, file-based. Built on OpenClaw.
取自儲存庫 README
正在載入完整 README。 以下為已儲存的摘錄。
Your self-hosted AI marketing department — a CMO agent that delegates to specialists.
MarketingClaw is an open-source, self-hosted AI marketing team you run on your own
hardware. Instead of a single chatbot, you get a small org: a CMO agent that owns
strategy and delegates, plus specialists for content, social, email, SEO, and
analytics. They plan campaigns, draft copy, schedule posts, run the newsletter, and
report on the numbers — and they answer you on the channels you already use.檢視完整 README ↗Quick start · Website · Docs · The team · Security · Vision
MarketingClaw is an open-source, self-hosted AI marketing team you run on your own hardware. Instead of a single chatbot, you get a small org: a CMO agent that owns strategy and delegates, plus specialists for content, social, email, SEO, and analytics. They plan campaigns, draft copy, schedule posts, run the newsletter, and report on the numbers — and they answer you on the channels you already use.
It is for founders, small teams, and in-house marketers who want a marketing team that runs on their own infrastructure, keeps every decision as a file they can read and diff, and never ships anything public without approval.
setup-marketing provisions a flat team of six role agents — one orchestrator and five
specialists — each with its own workspace, skills, and schedule.
| Agent | Role | What they do |
|---|---|---|
| Morgan | CMO · orchestrator | The default agent for every DM. Owns the campaign plan and calendar, breaks work down, and delegates to the specialists. |
| Sasha | Content | Long-form and copy — blog drafts, landing pages, post copy. Writes to files and hands off; never publishes directly. |
| Riley | Social | Schedules and publishes across your social accounts via Postiz, and triages mentions. |
| Jordan | Runs newsletters and lifecycle email with a draft → test → approve → send discipline, and watches deliverability. | |
| Quinn | SEO | Keyword research, on-page audits, and the blog pipeline into your CMS or git. |
| Alex | Analytics | Pulls the numbers from GA4, Search Console, and your platforms into weekly reports and flags what changed. |
sessions_spawn, specialists own their lane.setup-marketing installs a default cron set: a weekly content
calendar, a daily publish-queue reconcile, a weekly analytics report, a monthly SEO
audit, and a weekly email-health check.BRAND.md, the campaign plan, the content calendar, the post
log, and the report archive live in a git-tracked directory you own.Runtime: Node 24 (recommended) or Node 22.19+. MarketingClaw is a pnpm workspace;
use pnpm for source checkouts.
git clone https://github.com/promisingcoder/marketingclaw.git
cd marketingclaw
pnpm install
pnpm build && pnpm ui:build
# First run: guided setup of the gateway, workspace, and model provider
pnpm marketingclaw onboard
# Provision the marketing team: a short brand interview writes BRAND.md and
# creates the six-agent roster (Morgan, Sasha, Riley, Jordan, Quinn, Alex)
pnpm marketingclaw setup-marketing
# Start the gateway, then open the dashboard
pnpm marketingclaw gateway
pnpm marketingclaw dashboard
setup-marketingis the one new command MarketingClaw adds on top of the engine. If you skip it,onboardalone still seeds a marketing-flavored workspace with the CMO — runsetup-marketingat any time to provision the full team; it's idempotent. The full walkthrough lives in docs/start/marketing-quickstart.md.
Bring your own model API key (Anthropic, OpenAI, Google, and others); onboarding prompts for it. Prefer a current flagship model for the CMO and copy work.
MarketingClaw connects to real messaging surfaces, so it treats inbound DMs as untrusted input.
marketingclaw pairing approve <channel> <code>. Approved senders are added to a local
allowlist.dmPolicy="open" plus "*" in the
channel allowlist). Run marketingclaw doctor to surface risky DM policies.main sessions (group chats, shared
channels), set agents.defaults.sandbox.mode: "non-main" so tools run inside a sandbox.
The specialists that publish are approval-gated regardless of sandbox.Before exposing anything remotely, read the security docs: Security, Exposure runbook, Sandboxing, and Configuration.
The Gateway is the control plane: it runs on your machine (or a server), holds the
sessions, and connects every channel at once. On top of it, MarketingClaw adds a flat
team of role agents and a shared, file-based workspace under ~/.marketingclaw/marketing/
— BRAND.md, the campaign plan, the content calendar, and the report archive. Everything
the team decides is a plain file you can open, edit, and put under version control, and
nothing goes public without an approval.
Ask Morgan in chat — "Draft this week's newsletter," "Audit our landing page SEO," "Plan next week's content calendar" — and she breaks the work down, hands drafts to the right specialist, and brings the result back for your sign-off.
MarketingClaw is a fork of github.com/openclaw/openclaw (MIT), created by Peter Steinberger and its community. The gateway, messaging channels, agent runtime, and companion apps are the upstream project's work; MarketingClaw adds the marketing team, its skills, and the onboarding that provisions them. Engine-level fixes belong upstream. Full third-party attribution and the retained upstream release history live in THIRD_PARTY_NOTICES.md and UPSTREAM-CHANGELOG.md.
MIT — see LICENSE. Contribution guidelines are in CONTRIBUTING.md; the project direction is in VISION.md; report vulnerabilities via SECURITY.md.