# Product Brain — Full Description

> Chain-first product management for AI teams. The single source of truth for agentic workflows.

## What is Product Brain?

Product Brain is where your team's decisions, standards, tensions, insights, and work packages live in one typed, versioned graph: the Chain. It is the single source of truth your agents read and write through MCP, CLI, and the surfaces you use daily, not a folder of docs that drifts from reality.

Most tools organize information. Product Brain makes thinking visible: what blocks you, what needs a decision, and what already shipped — with human entry IDs your team can say out loud in standup and in Slack.

## The problem it solves

AI agents are fast builders, but teams need governed context that carries across sessions. Decisions made yesterday can be hard to locate today. Teams spend too much time re-explaining the same domain boundaries, business rules, and architectural choices to the same tool in the same codebase.

The gap between what agents can do and what teams can govern is the core problem. Product Brain bridges it with one rule: durable product knowledge goes on the Chain. Agents read it before they start. They write to it as they work. The next session starts from governed context instead of reconstructed context.

## Who is it for?

Product Brain is for AI-native builders who spend 10+ hours a week building with agents and are tired of scattered context:

- **Solo founders** building with AI who want their agents to understand their product from day one of each session
- **Staff Engineers** who need decisions logged, tensions tracked, and standards enforced without committee meetings
- **CTOs of product-led teams (5–50 people)** who need governance that compounds value instead of creating overhead
- **Anyone re-explaining their product to their AI every morning**

## The Chain

The Chain is a typed knowledge graph at the heart of Product Brain. Every entry is:

- **Typed** — decisions, tensions, standards, business rules, architecture entries, work packages, insights, features, glossary terms, and more
- **Versioned** — every change is tracked with history
- **Connected** — typed relations like `resolves`, `constrains`, `depends_on`, `governed_by`, `part_of`
- **Identified** — human-readable IDs like WP-109, DEC-301, TEN-445 that teams use in conversation
- **Queryable** — AI agents search, read, and write via MCP and CLI

The Chain holds everything: why a decision was made, what tension it resolved, which standard governs the implementation, and which work package is driving the work. Context compounds instead of decaying.

## The Product Brain Method

A seven-phase workflow where each phase uses the right AI model tier:

### 1. Orient
Get grounded in the current state of the project: active work packages, open tensions, and workspace stage. Happens automatically when the agent loads context.

### 2. Triage
Classify incoming work. Bug, feature, tension, question — the agent checks the Chain for related entries and routes to the right phase. No more guessing where to start.

### 3. Shape
Design a work package: problem, elements, risks, boundaries, and done-when criteria. This is where most quality is determined. Structured facilitation and review lenses stress-test the pitch before build.

### 4. Phase
Decompose a shaped work package into implementation slices with conversation primers. Each slice stays within one domain. The build agent starts with the right context on call one.

### 5. Build
Implement one slice per conversation. For multi-slice work packages, an orchestrator spawns build and review agents; no agent self-certifies its own work.

### 6. Review
Six-round verification across three lenses: Staff+ Engineer, CTO, and Chain Expert. Every finding gets fixed. Nothing deferred as "minor."

### 7. Ship + Retro
Commit and deploy. Then retro: capture friction, confirm what works, and patch what does not. Every session improves the next session's orientation.

## Three surfaces, one truth

### The Chain
Not another wiki. A typed, versioned graph of decisions, tensions, and rules that agents can query natively via MCP. Every entry has a human-readable ID and typed relations to other entries.

### Living Skills
Stop writing static .cursorrules or CLAUDE.md files. Build dynamic workflows that compound as you learn how your product needs to be built. Skills, rules, and standards are all versioned Chain entries that evolve with your team's learning.

### CLI First
Works in any terminal, integrates with Cursor and Claude Code. Your AI reads the context before it writes a single line. Commands like `pb orient`, `pb capture`, `pb search`, and `pb get` give agents and humans the same interface to the Chain.

## The model pyramid

Not every task needs the most powerful model. The method matches model capability to task complexity:

| Tier | Use for |
|------|---------|
| Frontier | Shaping, security audits, architectural redesign |
| Strong | Review, phasing, complex implementation |
| Workhorse | Daily implementation, preflight, iteration |
| Budget | Commits, boilerplate, mechanical tasks |

## Governance without bureaucracy

Lightweight, consent-based governance inspired by Holacracy:
- Async-first — no default meetings
- Decisions are logged on the Chain with full context
- Tensions drive change — they're first-class entries, not complaints
- Standards and business rules are queryable by agents
- The Chain holds everything so nobody needs to ask "why did we decide that?"

## Pricing

Product Brain uses usage-based pricing — you pay for work the product does, not for headcount.

### Builders (Free)
- 1 workspace with unlimited Chain entries
- MCP + CLI access (metered to shared pool)
- At least $25/mo equivalent in included usage credits during beta
- Community support
- No credit card required, no trial period

### Teams (from $99/mo)
- Everything in Builders
- Unlimited workspaces
- Teams, roles, governance flows
- Higher usage pool — same price for one collaborator or fifty
- Overage rates and priority support

### Enterprise (custom)
- Everything in Teams
- SSO / SAML, custom domains
- Dedicated MCP option
- Finance exports, SLA
- Schedule a call: hello@productbrain.io

### What counts as usage
- **MCP / API** — Tool calls and read/write operations against your workspace
- **Captures & writes** — Commits to the Chain: captures, updates, relations
- **Intelligence** — LLM-backed operations (classification, quality, suggestions)

No surprise bills. Usage is always visible. Free tier throttles gently when limits are approached — no silent charges.

## Data and privacy

- Chain entries are stored in Convex (managed database)
- Product Brain does not train on your data
- Your workspace is isolated
- No data sharing with third parties

## Getting started

1. Run `npx @productbrain/mcp@beta setup` in your terminal (~2 minutes)
2. Full setup guide: https://github.com/synergyai-os/Product-OS/blob/main/docs/community/product-brain-setup-guide.md
3. Join the community: https://community.productbrain.io
4. Say hi in #getting-started

## Origin story

Built by Randy Hereman after a decade in product governance and Holacracy, watching decisions decay across every organization. When AI agents arrived, the same context problem got sharper: teams had to re-explain domain boundaries to each tool and each session. Product Brain exists to make governed product knowledge usable before agents act.

## Links

- Website: https://productbrain.io/community
- Community: https://community.productbrain.io
- Method: https://productbrain.io/method
- Pricing: https://productbrain.io/pricing
- FAQ: https://productbrain.io/docs/faq
- Origin Story: https://productbrain.io/origin-story
- Setup Guide: https://github.com/synergyai-os/Product-OS/blob/main/docs/community/product-brain-setup-guide.md
- GitHub: https://github.com/synergyai-os/Product-OS