For Builders

You ship with AI agents, Claude Code, Cursor, Copilot Workspace, or similar. You’re productive, but you’ve lost track of the bigger picture. Features land, but you can’t answer “why did I build that?” or “what should I build next?” You don’t want to slow down to wrestle your workflow into a traditional tool that thinks in sprints, tickets, and estimation ceremonies. You just need to know what you’re betting on, what’s next, and what proof looks like. Without leaving the command line. ProductBrain fills the gap right above doing great work: the thinking layer that keeps your bets visible without asking you to change how you build. It sits above your delivery tools, not instead of them. If you use Linear or GitHub Issues, ProductBrain is the strategy layer on top. If you use nothing, it’s the only layer you need.

​

What ProductBrain does for you

​

A control surface for agents

Your AI agent reads the planning tree to understand what to build and why. When it ships something, it writes back, marking jobs done, adding notes, flagging blockers. The tree is the shared state between you and your agents.

​

Structure without ceremony

No standups, no sprint planning, no story points. The tree captures your strategy (goals → needs → bets) and the Delivery Map shows what’s shipping when. Your agent consumes this through the API. It knows what iteration it’s working in and what proof of delivery looks like.

​

The picture, without the context switch

Living in the terminal doesn’t mean flying blind. Your agent reads and writes the tree through the API, but that same tree renders as a visual whenever you want it. Step back and the Planning Tree shows the shape of what you’re building and why. The Delivery Map lays it out across releases. Seeing which goals are well-served, or what’s bundled next, is a glance in the browser. Not a context switch into a tool you have to operate. You keep the framework and the visuals to stay oriented, without leaving the flow you build in.

​

Jobs are acceptance criteria

Each job describes an observable result your agent should produce, not an implementation step. “Price comparison shows for scanned barcode” tells the agent what to verify. The agent decides how to implement it.

​

Context survives compaction

LLM agents lose context. Sessions end, conversations compact, and the agent forgets what it was doing. The tree is the fix. It’s external memory that persists regardless of what happens to the agent’s context window. After a compaction, the agent re-orients from the tree, what’s done, what’s next, the current iteration, instead of being re-briefed from scratch. The context lives in the plan, not in the agent’s window, so it survives the reset: you keep the tree current, and the agent reads it before it acts.

​

One API, full access

Read the tree, search by meaning, create nodes, update status, manage iterations. The same API the app uses. Your agent gets a structured view of the entire project with one authenticated call.

​

How it fits your workflow

1. Set up your project. Map out goals, needs, and bets (in the UI or via API)
2. Get your API key. One click from Settings, copy the quickstart snippet
3. Give it to your agent. Paste the snippet into your Claude Code session or agent config
4. Agent reads the tree. It knows what to build, what iteration it’s in, and what proof to produce
5. Agent writes back, Marks jobs done, adds notes, flags issues
6. You review. Check the Delivery Map to see progress across all your bets

​

Three ways to connect an agent

• REST API. Give your agent an API key and the LLM guide. Works from any agent that can make HTTP calls. Start with the Builder Quickstart.
• MCP server. Claude Desktop, Claude Code and Cursor can drive the plan as MCP tools with npx -y @productbrain-com/mcp. Same API underneath, same coaching in every response. See MCP Server.
• Through chat, no install. If your assistant can only read web pages (a locked-down laptop, connectors turned off), open Agent relay in the app and start a chat session. The assistant reads the plan through a private link and replies with its changes as text, and you paste them in. See Use Your Agent Through Chat.

​

Link the plan to the tools you already use

ProductBrain sits beside your tracker and roadmap tool. It doesn’t replace them. Three fields help an agent keep the two in step:

• externalRefs on any node: pointers to the same thing elsewhere, such as a Jira key or a roadmap item. Several per tool are fine. See External references.
• serves on a bet: the other needs it also helps, beyond the one it sits under. See A bet that serves more than one need.
• demand on a need: who felt it, what’s behind them, and when. Evidence, never a ranking. See Demand.

What changed recently is in the changelog.

​

Get Started