What it does

MarketingAgents turns a technical product description into a focused, evidence-backed marketing system:

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

$ marketingagents customer-research api-observability
→ Customer language, jobs, pains, triggers, objections, contradictions, source trail

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

$ marketingagents diagnose api-observability
→ Binding full-funnel constraint and the smallest useful test

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

$ marketingagents plan api-observability
→ Owners, dependencies, assets, approval gates, decision rules, 90-day sequence

$ marketingagents weekly-review api-observability
→ Verified learning, continue/change/stop/investigate decisions, one next action

The canonical context is reused across every workflow, so a returning founder resumes from prior evidence and decisions instead of restarting discovery. Paid advertising remains available through /creative and /track, but it is readiness-gated rather than treated as the default answer.

Built on Pi. Important claims and metrics must trace to customer evidence, a source URL, a saved artifact, or a raw platform response.

Installation

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/youmake-ai/marketingagents/main/scripts/install/install.sh | bash

Windows (PowerShell):

irm https://raw.githubusercontent.com/youmake-ai/marketingagents/main/scripts/install/install.ps1 | iex

The one-line installer fetches the latest tagged release and downloads a standalone native bundle with its own Node.js runtime.

The installer adds both marketingagents and its short alias ma. To upgrade, rerun the installer. marketingagents update only refreshes installed Pi packages inside MarketingAgents' environment; it does not replace the standalone runtime bundle itself.

Upgrading from AdsAgents? Install marketingagents and uninstall the former AdsAgents package. On first launch, MarketingAgents moves ~/.adsagents to ~/.marketingagents when the new directory is absent; if both exist, it preserves both and uses the new directory.

To uninstall, remove the launcher and runtime bundle, then optionally remove ~/.marketingagents if you also want to delete settings, sessions, and installed package state.

Local models are supported through the setup flow. Run marketingagents setup, choose your provider — LM Studio, LiteLLM, Ollama, vLLM, or OAuth/API-key for any of OpenAI, Anthropic, Google, OpenRouter, etc.

Workflows

Ask naturally or use slash commands as shortcuts. /start is the beginner-friendly front door; specialist commands remain available for direct use.

| Command | What it does | | --- | --- | | /start <product> | Establish context, assess six readiness gates, choose one next action | | /context <product-or-slug> | Maintain the shared product, customer, offer, funnel, proof, and constraint record | | /customer-research <slug> | Analyze first-party or public voice-of-customer evidence with provenance | | /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> | Select one highest-leverage action from current evidence and capacity | | /strategy <slug> | Choose focused full-funnel bets, non-bets, measurement, and approval gates | | /persona <slug> | Build evidence-grounded user, buyer, champion, and evaluator profiles | | /psychology <slug> | Map JTBD, friction, awareness, and ethical persuasion to messaging | | /plan <slug> | Turn strategy into a sequenced 90-day execution plan | | /creative <slug> | Generate readiness-gated ad creative via Higgsfield | | /track <slug> | Pull and verify Meta Ads evidence for the paid specialist branch | | /weekly-review <slug> | Turn full-funnel evidence into operating decisions and the next action | | /loop <slug> <purpose> | Design a bounded recurring process with state, self-checks, and a kill switch | | /report <slug> | Produce a decision-focused marketing report (Markdown + optional PDF) | | /summarize <source> | RLM-style summarization of URLs / files / PDFs |

Agents

Eight bundled subagents are dispatched when decomposition or an independent verification pass materially helps.

• 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 — full-funnel strategy/plan building plus adversarial critique
• psych-analyst — behavioral-science levers mapped to ethical messaging and creative directions
• persona-builder — evidence-grounded user, buyer, champion, and evaluator profiles
• creative-director — orchestrates Higgsfield + Meta tools; never launches ads
• tracker — verifies Meta insights and paid-campaign refinement proposals against raw evidence
• reporter — decision-focused full-funnel reporting; never fabricates missing data

Tools & Integrations

• Higgsfield CLI — image and video generation for ad creative (image, video, jobs, workspaces, balance)
• Meta Ads CLI — campaigns, ad sets, ads, insights, creative upload, budget updates
• Web search — provider-agnostic, configured via marketingagents search set
• Preview — browser and PDF export of generated reports

Both Higgsfield and Meta integrations are Pi extensions that shell out to their respective CLIs. Override binary paths with HIGGSFIELD_BIN and META_BIN env vars.

How it works

Built on Pi for the agent runtime. Capabilities are delivered as Pi skills — Markdown instruction files synced to ~/.marketingagents/agent/skills/ on startup.

The operating flow is context → customer/market evidence → diagnosis → strategy → plan → approved execution → measurement → weekly decision → next action. Recurring loops add cadence, an action condition, durable state, idempotency, self-checks, escalation rules, and a kill switch. Drafting and analysis can run autonomously; publishing, outreach, uploads, live-account changes, and spend require explicit user approval.

Run locally

git clone https://github.com/youmake-ai/marketingagents.git
cd marketingagents
nvm use || nvm install
npm install
npm run dev -- --help
npm run dev -- start "<your product>"

Run the interactive development build with npm run dev. Before publishing a change, run:

npm test
npm run typecheck
npm run build
ASTRO_TELEMETRY_DISABLED=1 npm --prefix website run build
node ./bin/marketingagents.js --help

Publish

marketingagents is a new unscoped npm package; npm does not rename the former package in place. For the first release, add a granular npm publish token as the repository secret NPM_TOKEN, then run the Publish and Release workflow (or push the new version to main). As a local fallback, publish while signed in to the npm account that will own the package:

npm login
npm whoami
npm pack --dry-run
npm publish

The GitHub workflow then handles later releases, native bundles, provenance, and release notes whenever package.json contains a version that is not yet fully released. After the first npm publish, configure npm trusted publishing for repository youmake-ai/marketingagents and workflow publish.yml; then remove the temporary NPM_TOKEN secret. The repository.url must exactly match the public GitHub repository for provenance.

See CONTRIBUTING.md for development conventions and RELEASES.md for the release entry.

Release Notes · MIT License

Acknowledgements

MarketingAgents was forked from Feynman (MIT). The Pi runtime + skills architecture, install pipeline, and many runtime patches are upstream contributions we depend on.

The context-first skills, explicit 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.