openmarketing

Open-source AI marketing operating system for technical founders — research, positioning, launch, acquisition, measurement, and iteration. Built on Pi.

Packages

Package details

extensionskillprompt

Install openmarketing from npm and Pi will load the resources declared by the package manifest.

$ pi install npm:openmarketing
Package
openmarketing
Version
0.1.1
Published
Oct 31, 2026
Downloads
569/mo · 569/wk
Author
khalidbaba
License
MIT
Types
extension, skill, prompt
Size
79.3 MB
Dependencies
6 dependencies · 0 peers
Pi manifest JSON
{
  "skills": [
    "./skills"
  ],
  "prompts": [
    "./prompts"
  ],
  "extensions": [
    "./extensions"
  ]
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README


What it does

Building the product is the part you're good at. Marketing is ten jobs at once: research the market, understand your customers, pick a strategy, write the ads, make the video, track what works.

OpenMarketing is a team of AI agents that does that work with you, in your terminal or in a desktop app. You describe your product; it walks you through research, positioning, personas, ad creatives, short-form video and Meta Ads reporting. It suggests, you decide: the brand, the style and the plan are yours to approve, and nothing is published, uploaded or spent without you.

$ openmarketing start "an API observability product for small platform teams"
→ Shared context, readiness check, current bottleneck, one justified next action

$ openmarketing research api-observability
→ Competitors, alternatives, demand signals, category movement, distribution evidence

$ openmarketing strategy api-observability
→ Positioning, up to three focused bets, explicit non-bets, measurement

$ openmarketing video "15-second launch teaser"
→ Brand check, reference picks, style choice, storyboard approval, rendered and QA'd video

Every workflow reads and updates one shared context, so you pick up where you left off instead of starting over. Important claims and metrics must trace to customer evidence, a source URL, a saved artifact, or a raw platform response.


Install

npm install -g openmarketing
openmarketing setup

This adds openmarketing and its short alias om. Requires Node.js 20.19 or newer (up to 24.x). openmarketing setup connects a model: sign in with a subscription (Claude, ChatGPT, Copilot, …), paste an API key, or point it at a local model (LM Studio, Ollama, vLLM, LiteLLM).

Standalone bundle with its own Node.js runtime, from GitHub Releases:

curl -fsSL https://openmarketing.dev/install | bash                 # macOS / Linux
irm https://openmarketing.dev/install.ps1 | iex                     # Windows (PowerShell)

Rerun the installer to upgrade it. openmarketing update refreshes the installed Pi packages, not the bundle itself. To uninstall, remove the launcher and bundle, and ~/.openmarketing if you also want to delete settings and sessions.

Coming from MarketingAgents or AdsAgents? Your settings and sessions move to ~/.openmarketing automatically on first launch.


Desktop app

The desktop app runs the same agents over a folder you pick: each tool call gets its own card, decisions show up as cards you answer in the chat, and the side panel shows your references, brand page and latest video. See apps/desktop to run it.


Workflows

Ask in plain words, or use slash commands as shortcuts. /start is the front door.

Command What it does
/start <product> Set up context, check six readiness gates, choose one next action
/context <product-or-slug> Keep the shared product, customer, offer, funnel, proof and constraint record
/customer-research <slug> Analyze voice-of-customer evidence, with where each quote came from
/research <topic-or-slug> Research competitors, alternatives, demand, category changes and distribution
/diagnose <slug> Find the binding acquisition, activation, retention, referral or revenue constraint
/next <slug> Pick the single highest-leverage action from current evidence and capacity
/strategy <slug> Choose focused bets, non-bets, measurement and approval gates
/persona <slug> Build evidence-grounded user, buyer, champion and evaluator profiles
/psychology <slug> Map jobs-to-be-done, friction, awareness and ethical persuasion to messaging
/plan <slug> Turn the strategy into a sequenced 90-day plan
/creative <slug> Generate ad creative with Higgsfield, once the basics are in place
/video <brief> Make a short-form video, guided step by step, with an automated QA gate
/track <slug> Pull and verify Meta Ads results
/weekly-review <slug> Turn the week's evidence into decisions and the next action
/loop <slug> <purpose> Design a safe recurring process with state, self-checks and a kill switch
/report <slug> Write a decision-focused marketing report (Markdown, optional PDF)
/summarize <source> Summarize a URL, file or PDF without loading it all into context
/jobs, /log See running background work; save a session log

Agents

Nine bundled subagents take on work that splits cleanly or needs an independent check.

  • market-researcher: market, alternatives, audience, demand and channel evidence, with source URLs
  • customer-researcher: first-party and proxy voice-of-customer synthesis, with quote provenance
  • strategist: builds the strategy and plan, then critiques them adversarially
  • psych-analyst: behavioral-science levers mapped to ethical messaging
  • persona-builder: evidence-grounded personas and business understanding
  • creative-director: Higgsfield creative and Meta uploads; never launches ads
  • video-editor: storyboard, render, finish and QA for short-form video; never publishes
  • tracker: verifies Meta Ads results and refinement proposals against raw data
  • reporter: decision-focused reporting; never fills in missing data

Tools and integrations

  • Higgsfield CLI: image and video generation for ad creative
  • Meta Ads CLI: campaigns, ad sets, ads, insights, creative upload, budgets
  • Video studio: Remotion and ffmpeg rendering with a QA gate
  • Cross-model tools: ask another connected model for a second opinion, or generate images with GPT Image
  • Web search: provider-agnostic, set with openmarketing search set
  • Preview: browser and PDF export of reports

Higgsfield and Meta are Pi extensions that call their CLIs. Override the binary paths with HIGGSFIELD_BIN and META_BIN.


How it works

OpenMarketing runs on Pi. Workflows are prompt templates in prompts/, capabilities are Pi skills in skills/, subagents are defined in .openmarketing/agents/, and the tools are Pi extensions in extensions/. On startup these are synced into ~/.openmarketing/agent/.

The flow is context → customer and market evidence → diagnosis → strategy → plan → approved execution → measurement → weekly decision → next action. Drafting and analysis can run on their own; publishing, outreach, uploads, live-account changes and spend need your explicit approval.


Contributing

git clone https://github.com/babakhalid/openmarketing.git
cd openmarketing
nvm use || nvm install
npm install
npm run dev -- start "<your product>"

CONTRIBUTING.md has the project map, the checks to run before a PR, and the release process. AGENTS.md is the contract for agents working in this repo.

Release notes · MIT License


Acknowledgements

OpenMarketing is built and maintained by Youmake.

OpenMarketing was forked from Feynman (MIT). The Pi runtime and skills architecture, the install pipeline and many runtime patches come from upstream.

The context-first skills, workflow handoffs, full-funnel framing and recurring-loop design were inspired by Corey Haines' Marketing Skills (MIT), then adapted for a local, artifact-driven multi-agent runtime with stricter provenance and human-approval boundaries.