Overview
SetForth turns the work you describe into finished, reviewable changes. You set forth a vision — a project, an initiative, the tasks that make it up — and agents carry it out, opening pull requests you can review and merge on your own terms.
This page explains, at a high level, the two ways your agents can run: in the SetForth cloud or on your own machines with a self-hosted runner. You can use either, or mix both across projects.
How work flows
SetForth currently works only with repositories hosted on GitHub. Your project connects to a GitHub repository, and agents open branches and pull requests there. GitHub is the only supported repository host today — code on GitLab, Bitbucket, or a self-managed Git server is not yet supported.
However your agents run, the shape of the work is the same:
- You describe what you want — a project connected to a GitHub repository, broken into initiatives, epics, and tasks.
- Agents pick up tasks, make changes on a branch, and open a pull request.
- You review the pull request like any other and merge when it is ready. To have an agent act on feedback, submit it as a pull request review — see Pull Request Reviews.
Everything an agent does lands as an ordinary branch and pull request in your repository, so it fits the review and release process you already have.
Cloud agents
Cloud agents are the simplest way to start. They run in managed, isolated environments that SetForth provisions for you — there is nothing to install and nothing to keep running.
- Auto-started. When a task is ready, an environment is prepared and the agent begins automatically. You do not start or babysit anything.
- Isolated per run. Each run gets its own clean, sandboxed workspace, so runs do not interfere with one another.
- Zero maintenance. SetForth handles capacity and upkeep; you just review the results.
Cloud agents are a good fit when you want to move quickly and are comfortable having SetForth host the environment your code is checked out into.
Self-hosted runners
A self-hosted runner lets agents run on hardware you control — your own laptop, a server, or a machine inside your network. You install a small runner program, register it with SetForth, and point it at the projects you want it to serve.
Self-hosted runners are the right choice when you need:
- Your environment, your tools. Builds and tests run with the exact toolchain, dependencies, and OS you already use.
- Network access to private resources. Reach internal package registries, databases, or services that are not exposed to the public internet.
- Data residency and control. Your source code is checked out and processed on machines you own.
You stay in control of the runner: you decide which machines it runs on, when it is online, and which projects it serves. Unlike cloud agents, a self-hosted runner needs a little host setup: Node.js, Docker, and Anthropic credentials (either an ANTHROPIC_API_KEY or a Claude subscription signed in on the host). Setup is a short, guided flow, and a dedicated guide walks through requirements, installation, registration, and configuration.
Cloud vs. self-hosted at a glance
| Cloud agents | Self-hosted runners | |
|---|---|---|
| Where it runs | SetForth-managed environment | Machines you own |
| Setup | None — start right away | Install a runner, provide Anthropic credentials, and register it |
| Maintenance | Handled for you | You keep the runner online |
| Private network access | No | Yes |
| Best for | Getting started fast | Custom toolchains, private resources, data control |
Choosing between them
Start with cloud agents if you want the fastest path and your project builds and tests fine in a standard managed environment. Choose self-hosted runners when your work depends on private infrastructure, a specialized toolchain, or keeping code on your own hardware. The choice is per project, and you can change your mind later without redoing your work.