How it works
01
You describe the decision.
The agent asks two questions: what are you deciding, and what does it turn on? That's the intake gate. It doesn't start building until it understands.
02
The model takes shape.
Your AI assistant calls Flatland's tools — adds typed assumptions, writes formulas, sets assertion guardrails. The dependency graph builds in real time. You can watch it or let it run.
Install
Get an API key at flatlandfi.com/install — no card — then run the agent:
npx flatland-agent
Requirement: Node.js 20+. Bring your own Anthropic or OpenAI key for inference; your FLATLAND_API_KEY meters engine answers.
What you get
Typed model
Named, typed assumptions. Not cell references.
Yours
~/.flatland/models/
When to use the harness
Use the harness when
- You're starting from a decision, not a model structure
- You want the agent to ask before it builds
- You need the two-phase guardrail (intake → workspace)
Use direct MCP tools when
- You already have a model loaded
- You're in a scripted agent workflow
- You need to call specific tools without the harness wrapper