by 5hirish
Open-source AI marketing agent: a growth analyst with every login. Reads Google Ads, GA4, Search Console, Stripe & more together, tells you why a number moved, and fixes it when you approve.
Available after the catalog security scan passes.
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
# Add to your Claude Code skills
git clone https://github.com/5hirish/ductGuides for using ai agents skills like duct.
See how duct compares with popular alternatives.
See the whole picture. Fix it in the same moment.
Duct is an open-source AI marketing agent for product and growth teams. It connects Google Ads, GA4, Search Console, Mixpanel, Stripe and seven more, reads across all of them at once, remembers what it learns about your business, and, with your approval, changes things in those tools for you. The approve tool does not exist in the list the model can call; you press the button.
⬇️ Download the desktop app and Duct runs the service for you; your model keys never leave your machine. Prefer to run every part of it yourself? Self-host it. Same code, MIT either way.
Free during beta · No credit card · About 10 minutes to connect your tools
| Before you install | Straight answer |
|---|---|
| Where do my keys live? | Model keys: your OS keychain, never stored on our servers. Tool connections: encrypted at rest, one project at a time, and on your own disk if you self-host. |
| What can the agent do without me? | Read, compare and propose. By default nothing in your accounts changes until you press Apply. How that is enforced. |
| What does it cost to run? | Duct is free during beta. The model is yours: your own API key, or your ChatGPT plan in the desktop app. |
| Can I run all of it myself? | Yes. Self-host it: SQLite on your disk, no account, nothing leaving the machine. |
| What if Duct goes away? | It is MIT. The self-host build is the same code and needs nothing from us. |
📊 Insights. Ask "why did signups drop last week" and get an answer that reads Google Ads, GA4 and Stripe together, not one at a time. The kind of thing it finds: ROAS up 14%, but Android 7-day retention down 2.4×. You are paying to acquire churners. It pauses to ask when it needs a decision, a verification sub-agent re-checks every number before it reaches you, and the brief it writes is saved, versioned and cites its sources.
🌱 SEO audit. Paste a URL and Duct crawls the site, reads the technical and content signals, and hands back a prioritised report. This is the free front door at getduct.ai: no account needed to run one.
🎬 Content Studio. A plan on a board, a post per card, slides and images generated for each one, and one click to publish to TikTok, Instagram, LinkedIn, YouTube, Threads, X, Facebook, Pinterest and Bluesky. Results come back onto the card, so the next plan learns from the last one.
Who it is for today: PMs and growth teams, SEO and content teams, and performance marketers chasing attribution gaps, budget shifts and creative fatigue.
🚧 Coming next: Sales & RevOps (pipeline risk and conversion signals from CRM, calls and product usage), E-commerce & DTC (LTV-versus-ROAS blind spots across Meta Ads, Stripe and RevenueCat), and Customer Success (at-risk accounts from NPS, usage and support signals). Follow along in Issues and at getduct.ai.
Ask Claude Code, Codex or Cursor "is my site ready to launch?" and it reads your live pages and the repository behind them, checks what the page promises against what the code does, scores the launch out of 100 and gives you three ranked fixes, each as a diff it applies only when you say yes. No Duct account, nothing sent to us.
npx skills add 5hirish/duct
Sign in with your own ChatGPT plan in the desktop app and Duct runs on it. No API key, no second bill. Or bring keys for Anthropic, OpenAI, Gemini, xAI or anything on OpenRouter, and assign three tiers, Heavy, Standard and Light, so deep analysis and cheap summaries do not pay the same price. Hit a rate limit and Duct steps down a tier instead of stopping.
An agent can propose a change and show you a preview; a person applies it. Pause a campaign, add negative keywords, set a budget or a bid, create a GA4 key event or audience, publish a Tag Manager version, annotate Mixpanel. Every change set is previewed, checked against guardrails you set, applied only after approval, and can be rolled back. The agent can propose and inspect. It cannot approve or apply. That decision lives in code the model never sees, and there is a test that keeps it that way.
Duct keeps a memory of your business, your projects and the work it has done: the CPA target, the campaign you told it to ignore (and that you were the one who said so), the finding from last month's brief. Every fact knows when it was true and what replaced it, so an answer can say why it believes something. You can read the whole timeline, correct it, pin what matters, pause it or reset it.
Google Ads · Google Analytics 4 · Search Console · Tag Manager · Meta Ads · Apple Search Ads · OpenAI Ads · Mixpanel · Microsoft Clarity · GrowthBook · Stripe · RevenueCat · GitHub, with HubSpot next and new ones landing as the agents need them. What each one reads and can change: getduct.ai/integrations.
Google connects with OAuth, read-only unless you grant more; the rest take a key you paste. Credentials are encrypted at rest, scoped to one project at a time, and Duct tells you exactly which permissions it has and which it is missing.
Projects hold the context. Invite collaborators, share a report with them by link, and every brief, audit and plan stays in the project's library, versioned, for the next person who asks.
Native on macOS, Windows and Linux. Model keys live in the OS keychain, sign-in runs in your own browser, updates are signed, and it says hello with a mosaic. Download it.
The same app can be built with the backend inside: SQLite on your disk, loopback only, no account, nothing leaving the machine. Or run the backend on your own server:
git clone https://github.com/5hirish/duct.git && cd duct
make setup
cp backend/.env.example backend/.env.local # add one model provider key
make serve-backend # then, in a second terminal:
make serve-app
CONTRIBUTING.md has the full setup, and desktop/README.md covers the self-host build.
Python and FastAPI on the backend, Next.js in the browser, a Tauri shell on the desktop, one repository. docs/architecture.html is the interactive map: open it in a browser, pick the job you came to do, and it walks you to the files. The Duct Doctrine is the short version of what the project believes, each tenet linked to the code that enforces it.
| Path | What | Read first |
|---|---|---|
backend/ |
API, agents, connectors, database | backend/AGENTS.md |
app/ |
The product UI | app/AGENTS.md |
desktop/ |
The native shell | desktop/AGENTS.md |
site/ |
getduct.ai | site/AGENTS.md |
docs/ |
Engineering records and references | docs/README.md |
Start with CONTRIBUTING.md. Not touching code? CONTRIBUTOR-PLAYBOOK.md is for design feedback, writing and product proposals. Everyone who helps is credited in CONTRIBUTORS.md. Agent-written contributions are welcome; the PR template asks where an agent helped so a reviewer knows where to look. Security issues go to SECURITY.md, never a public issue.
Questions and ideas go in Discussions, bugs and proposals in [Issues](https://github.com/5hirish/duct/issu
duct is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by 5hirish. Open-source AI marketing agent: a growth analyst with every login. Reads Google Ads, GA4, Search Console, Stripe & more together, tells you why a number moved, and fixes it when you approve. It has 9 GitHub stars.
duct's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/5hirish/duct" and add it to your Claude Code skills directory (see the Installation section above).
duct is primarily written in HTML. It is open-source under 5hirish on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh duct against similar tools.
No comments yet. Be the first to share your thoughts!