Skip to content

How it works

A software factory you can run, read, and govern.

Spec-driven development decides what to build. Spec-Driven Agents govern how it gets built and prove the outcome: the work moves through defined stages, the right approver signs off on what matters, and the result arrives with evidence.

The loop

From human intent to verified outcome.

New work, fixes, and maintenance all follow the same loop. Each phase produces something a person can review, and work only moves on when the checks pass and the right approver approves. When something drifts in production, the loop starts again.

Human intent
  1. 01Plan
  2. 02Design
  3. 03Build
  4. 04Test
  5. 05Deploy
  6. 06Maintain
Gates: named checks and approvals
Verified outcome, with evidence

Phase by phase

What happens at each step.

01

Plan

Capture the intent and approve it.
02

Design

Write the spec, design, and plan, each reviewed and approved.
03

Build

Write tests that fail first, then make the change that passes them.
04

Test

Verify the change against the full spec.
05

Deploy

Deliver the change and record it.
06

Maintain

Watch quality in production. A drop below the agreed range starts a new run.

What makes it different

Six ideas behind the platform.

A spec you follow, not just one you write

The spec is the rules of the run. Each stage has to meet it before the work moves on, instead of the spec being read once at the start.

People stay in charge

Gates need passing checks and an approval from the right approver: a person or a different agent, never the agent that did the work. People can pause, deny, or roll back at any time.

Bring your own agents

Work runs on the agent SDKs, models, and AI coding tools your team already uses, so you can change any of them without starting over.

Knowledge that compounds

Decisions, lessons, and your own domain knowledge are kept with where they came from, so each run starts from what the last one learned.

Guardrails on every step

What goes into a model, what comes out, and what an agent is allowed to do are checked, with a path to a person when it matters.

Evidence, not plausible output

Every run keeps its artifacts, checks, approvals, and results, so you can see what happened and why.

Built to be replaced

Open contracts, no lock-in.

Each part of the platform does one job and connects to the rest through open, versioned contracts. Swap a model, an agent SDK, a work tracker, or a knowledge store without rebuilding everything around it.

Govern AI agents without the overhead.

We're working with a small group of early teams. Tell us what you're building.

Get early access