● How we work
The factory, managed for you and set up with your team.
We run the factory, and your tracker, repository and CI stay as they are. We start with one workflow and widen it only as fast as the numbers allow.
01 Platform
We run the factory
Agents plan, build, review and merge your work in the workflows you agree on. You see every change, its checks and its cost.
- Agents work in sandboxes you choose: Docker on your own machines, Kubernetes, AWS, Cloudflare, or cloud sandboxes from E2B and Daytona.
- Your tracker, repository provider and CI stay. Your required checks decide what merges.
- Claude Code, Codex, OpenCode or Pi writes the code, under the same checks and limits.
- People sign in through your identity provider, with access set per project.
02 Services
We set it up with you
The team that builds this product on itself works with yours. We bring the rules we use and fit them to how your team works.
- Methodology: definitions of ready and done, review rules and decision rules, written for your team.
- Consulting: how work moves through your team today, and which work to hand to agents first.
- Implementation: your tools connected, your workflows encoded, budgets and limits set.
- A weekly review of the process with you: the numbers, what a person had to decide, and which rule to change next.
● Three stages
Start with one workflow. Widen it as it earns trust.
The first workflow runs next to the way your team works today, with a person approving every merge. Each later step is taken when the numbers support it, and you decide when that is.
Map
- We
- Write down how work moves through your team today, and pick the first workflow with you: usually chores that cost time and carry little risk.
- You
- Time with the people who run delivery, and access to your tracker.
- At the end
- Your process as a workflow, with its checks and the points where a person decides.
Pilot
- We
- Run the workflow on one repository, connected to your tracker, repository provider and CI, with budgets and limits set. A person approves every merge.
- You
- The repository and its CI, a place for agents to run, and someone who answers decision requests.
- At the end
- Every change with its review, checks and cost, and the numbers: changes merged per week, cycle time, and how often a person stepped in.
Expand
- We
- Add workflows and repositories, and let a workflow merge on its own once its numbers show it is safe.
- You
- The person who owns the process, and the people who decide where it needs them.
- At the end
- More of the work runs without a person, workflow by workflow, with each change to the process marked on the charts.
● What we bring
The rules that make agent work dependable
We build our own product this way. Each part is a written rule and, where it can be, a check in the factory that enforces it.
Read the methodology →A definition of ready
What a task must say before an agent starts it. An incomplete brief is refused before any work begins.
A definition of done
The tests, review and docs every change needs, enforced by your checks and the merge gate.
Review rules
Reviewers block only on an agreed list. Everything else becomes follow-up work, not another round.
Decision rules
Which questions go to a person and which the factory settles, and how a workflow moves toward merging on its own.
Roles for agents
Planner, implementer and reviewer, each with its own instructions for its step.
A weekly review
The numbers, read together: where time went, what a person had to decide, and which rule to change next.
● What we need from you
Your tools, a place to run agents, and an owner
Your engineers keep the tools they know. The factory works in the tools your team already runs.
- A repository and its CI: GitHub, Bitbucket, GitLab or Azure DevOps.
- Your tracker: Linear, Jira, GitHub Issues, Azure DevOps Boards or Beads.
- A place for agents to run: Docker in your network, Kubernetes, AWS, or cloud sandboxes.
- A person who owns the process and decides how it changes.
- The people who answer when the process needs a person.
● Contact
Bring us one workflow
Tell us which work you would hand to agents first. We show you the factory shipping its own code, then sketch that workflow with you, step by step.
Book a demo● FAQ
Questions before you start
Which coding agents does the factory run?
Claude Code, Codex, OpenCode and Pi. The workflow, its checks and its limits stay the same whichever agent runs a step.
Do we have to change our tracker or our CI?
No. Work comes from your tracker, and the issue’s status follows the work. Changes go to your repository provider as pull requests, and your required checks decide what merges.
Where do the agents run?
In sandboxes you choose: Docker on your own machines, where the factory reaches a Docker host in your network, Kubernetes, AWS, Cloudflare, or cloud sandboxes from E2B and Daytona. Each task gets its own sandbox.
What do people still do?
They own the process: which work comes in, which merges a person approves, and how the rules change. When the factory stops and asks, they answer.
How is spend kept in check?
Every task has a budget and every project a ceiling. At the cap the next agent turn does not start, and a person chooses to raise the budget, replan or cancel.
What happens when an agent gets stuck?
New plans, review rounds and repair rounds are capped. When the work cannot converge or checks stay red, the factory stops and asks a person, with the evidence and a recommended answer. The answer continues the work where it stopped.
How do we change the process later?
In the web app, or by asking an agent. Every change is a new version, and a change to intake or routing is previewed on the last seven days of work before it goes live. Work in flight stays on the version it started with, and rolling back means switching to the previous version.
Can we start with one repository?
Yes, that is how a pilot starts: one workflow on one repository. More come in the next stage, when the numbers support it.