Get Started
From idea to a team shipping under H1VE.
Two layers. First the method — what to do, tool or no tool. Then the reference platform — how to operate it for real, through the panel, the CLI, or your AI.
H1VE is a method, not a tool. These steps work no matter what you build with — they describe how a team of humans and AI agents works under H1VE. Layer 2 then shows how to make it concrete with the reference platform.
- 1
Design the project on one sheet
Before any code, the founder and architect fill the H1VE Canvas — problem, vision, users, scope, team, tech foundation, first slices, gates, risks. One sitting, one sheet.
Why
The canvas is the seed. From it, the four foundation documents are born — vision, functional spec, technical spec, and roadmap.
- 2
Write the foundation documents
Expand the canvas into the four documents that define the project, with the AI as your drafting partner. This is the cake; every spec that follows is a slice.
- Vision & Scope — why it exists, for whom, the limits
- Functional Spec — what it does, from the user's view
- Technical Spec — how it's built: stack, data, decisions
- Roadmap — the scope sliced into phases
- 3
Set the team and the roles
Assign the five roles: founder, architect, dev, QA, data. On a small team one person may wear several hats — but the responsibilities never merge. A reviewer never validates their own work alone.
- 4
Slice the first spec from the roadmap
Take the first roadmap item and turn it into SPEC-001 — scope, files touched, acceptance criteria, risks. The AI helps draft it; a human approves it. No code without an approved spec.
- 5
Run it through the cycle
The dev builds with the AI, only within the spec. Then the work flows through the gates:
- CI — checks the mechanics — types, lint, tests
- QA and Data — validate independently — both must sign off
- The architect — performs the final merge — the only one who can
The loop
When the feature reaches main, it's live. Slice the next spec. Repeat. The foundation is done once; the cycle repeats per feature.
The method needs a place to live. H1VE is tool-neutral, but a reference implementation makes the cycle and the gates real. The reference platform — H1VE Flow — gives you three ways to work, choosing by where you are, not by preference. They all act on the same project.
The reference platform · three interfaces
The platform — for management & visibility
The web panel is the home of the method. The swim lane shows every feature across the nine stages in real time. Founders create and assign features, approve specs, and watch the gates. QA and Data sign off here. The architect reviews and merges.
This is where you see the whole system — blockers, the architect's queue, technical-health metrics, the inbox of what each role owes.
Use it when
You're managing the team, approving specs, signing off, or want the bird's-eye view of where everything stands.
The CLI — for devs, in the terminal
The nf command-line tool lets a developer operate H1VE without leaving the terminal. It detects the current git branch and finds the matching feature — its spec, stage, and active blockers — automatically.
# see your assigned work and current featurenf status # start a feature — creates the branch, links itnf start # log a blocker without breaking flownf blocker "waiting on API keys" # submit the AI declaration and move to PRnf doneUse it when
You're a dev heads-down in the terminal and want to move work along without context-switching to the panel.
The MCP-Server — for your AI agent
The MCP-Server connects your AI agent directly to the H1VE flow. Through it, the agent sees the current feature — spec, stage, blockers — and can act: move a stage, log a blocker, submit the AI declaration, all from inside your conversation with it. No copy-pasting context.
# inside your AI agent, with the MCP-Server connected:You: what's the current feature and its spec?Agent: reads the branch → returns SPEC-014, stage dev, no active blockers. You: I'm done — submit the declaration.Agent: writes the AI declaration, moves to PR.Use it when
You're working alongside your AI agent and want it to operate the flow with you — the tightest loop between building and tracking.
One project, three doors
Panel, CLI, and MCP all act on the same data through the same rules. A dev can start a feature in the CLI, the AI can move it through MCP, and the founder sees it live in the panel — instantly. Pick the door closest to where you already are.
Ready to build the H1VE way?
Start with the canvas and the playbook. Then run your first slice through the cycle.