OPEN SOURCE · MCP-NATIVE · NO API KEYS

Your AI
product team.

AutoPM turns abstract ideas into shipping roadmaps. 18 specialist PM robots run market sizing, competitive intel, personas, financial models and execution specs — inside Claude, Cursor or ChatGPT, saving every artifact to your own disk.

18 ROBOTS 2 GATED PHASES 0 API KEYS MIT LICENSED
autopm — pipeline simulator
// awaiting idea — press RUN to watch the robot pipeline execute
SIMULATED RUN · real runs execute in your MCP client with live research & saved artifacts
The problem

“I can get AI to write a PRD.
I need help knowing whether it is the right PRD.”

— every PM using generic AI chat, 2026

Chat assistants produce polished documents on demand — and skip the discipline that makes documents worth trusting. No market gate before the spec. No evidence trail behind the verdict. No way to tell fresh thinking from stale. AutoPM is a product judgment system, not a text generator: a gated, two-phase method that forces the hard questions in order, scores its own confidence, and leaves an auditable file trail for every decision.

The method

Discovery first.
Execution second. Gated.

The most common PM failure is rushing into solution design without a validated strategy. AutoPM makes that impossible — Phase 2 stays locked until every Phase 1 analysis is fresh.

Phase 1 · Strategic Discovery

Why & Who

Eight robots interrogate the idea: demand, competitors, personas, unit economics, features, roadmap, priorities.

🎤 interview🔭 scout🔎 detective👥 people 💰 money📝 feature🗺️ plan⭐ priority
Phase Gate

All 7 analyses fresh — or blocked

Every robot output carries a staleness window and a machine-readable verdict block. Preliminary evidence caps the verdict at CONDITIONAL GO. No fresh strategy, no Phase 2.

Phase 2 · Execution Definition

What & How

Ten robots translate validated strategy into engineering-ready constraints: stories, scope, journeys, feasibility, KPIs, privacy, GTM, risks, DACI.

📖 stories🎯 scope🛤️ journeys⚙️ tech🎨 design 📈 kpis🛡️ privacy🚀 gtm⚠️ risks🤝 daci
Output pdd-<product>-v1.0.0.md Notion-ready Product Definition Doc strategy-presentation.html pitch-ready HTML deck
The team

18 specialists.
One opinionated pipeline.

Each robot is a prompt-engineered expert with a mandate, guardrails and a required output schema — not a generic assistant with a nickname.

Phase 1 — Strategic Discovery

8 robots · why & who
🎤
Interview

Locks the product idea with a dynamic PM interview.

🔭
Scout

Sizes demand — TAM/SAM/SOM, growth signals, validation.

🔎
Detective

Maps competitors, gaps, moat and positioning.

👥
People

Builds personas — pains, motivations, buying triggers.

💰
Money

Models unit economics and 3-scenario forecasts.

📝
Feature

Splits must-have from nice-to-have, with WHY and WHEN.

🗺️
Plan

Lays a phased 18-month roadmap with dependencies.

Priority

RICE-scores the backlog with principled reasoning.

Phase 2 — Execution Definition

10 robots · what & how
📖
User Stories

Epics and stories with acceptance criteria.

🎯
Scope Spec

Hard scope lines and out-of-scope boundaries.

🛤️
Journeys

End-to-end journey narratives per persona.

⚙️
Tech Feasibility

Architecture, constraints, vendors, infra deps.

🎨
Design Feasibility

UX guardrails, wireflow, accessibility.

📈
KPIs

Metrics, instrumentation, target baselines.

🛡️
Data Privacy

GDPR/CCPA, security and certification impact.

🚀
GTM Readiness

Rollout waves, channels, preview→GA gates.

⚠️
Risks Registry

Structured risk matrix across 5 categories.

🤝
DACI

Driver, approver, contributors, informed.

Radical honesty

We ran AutoPM
on AutoPM.

Most landing pages show you five-star quotes. We show you the score our own product gave our own hypothesis — vanity-free, straight from the Scout robot's saved output.

Hypothesis Supportrange 66–78
Evidence Maturityrange 52–66
⛩ VERDICT: PROMISING — VALIDATION-GATED
{
  "robot": "scout",
  "verdict": "MODERATE",
  "hypothesisSupportScore": { "low": 66, "base": 72, "high": 78 },
  "evidenceMaturityScore":  { "low": 52, "base": 61, "high": 66 },
  "confidenceStatus": "preliminary",
  "evidenceTier": "researched-cited"
}

A tool built to stop PMs from believing their own hype shouldn't ship with hype. Every AutoPM robot ends its analysis with a machine-readable verdict block like this one — scored, ranged, and capped when evidence is thin. This is the actual block from AutoPM's self-analysis.

The output

Don't imagine the deck.
Open one.

Every AutoPM run ends in a pitch-ready strategy presentation and a Notion-ready PDD. Here's a full Phase 1 deck, generated in AutoPM's real output format for a fictional product.

Every run exports

strategy-presentation.html — themed pitch deck
pdd-<product>-v1.0.0.md — Notion-ready PDD
*-output.md — every robot's full analysis, on your disk

Note the verdict

The demo ends in CONDITIONAL GO, not a rubber-stamp GO — thin evidence caps the verdict and blocks Phase 2. That gate is the product working as designed.

Built for

Three PMs. One method.

The solo builder
“The thinking is already there. I need people to see it without asking them to browse a directory tree.”
You get: a full product-strategy stack from one interview — every artifact a real file on your disk, ready to share.
The product leader
“I don't need another place to store docs. I need to know whether the thinking behind the docs is any good.”
You get: a repeatable judgment system — freshness windows, evidence tiers and verdict scores your whole team can be coached against.
The new PM
“I can get AI to write a PRD. I need help knowing whether it is the right PRD.”
You get: a guided path through the questions senior PMs ask — market, competition, economics — before anyone writes a spec.
Why not just…

Documents are easy.
Judgment is the product.

Generic AI chat
Brilliant drafts, zero lifecycle

Great first-pass text. But no gates, no freshness, no artifact lineage — the thinking evaporates with the conversation.

Notion / Confluence
Stores artifacts, can't judge them

Pages exist, but nothing tells you whether the thinking behind them is current, complete, or trusted.

PM suites
Manage roadmaps, skip the method

Enterprise-grade artifact management — that assumes the strategic thinking already happened somewhere.

AutoPM
A gated product-judgment system

Forces the hard questions in order, scores its own confidence, and leaves an auditable file trail — locally, for free.

Get started

Two minutes to your
first robot run.

No signup. No API keys — the model already inside your MCP client does the reasoning. Node 20+ is the only requirement.

Clone and install

AutoPM is MIT-licensed and runs entirely on your machine.

git clone https://github.com/fluidumber/AutoPM.git
cd AutoPM
npm install

Point your MCP client at it

Works with any MCP-compatible client. Pick yours:

// claude_desktop_config.json
{
  "mcpServers": {
    "autopm": {
      "command": "node",
      "args": ["/absolute/path/to/AutoPM/src/mcp-server.js"]
    }
  }
}

Start the interview

Restart your client and ask it to use the interview tool. The Interview robot locks in your product idea, then the pipeline takes over — every output saved as markdown you own.

REQUIRES NODE 20+ LICENSE MIT YOUR DATA STAYS LOCAL