SI Fabrik

Version 1.0 · Free for macOS 26+

Turn your backlog into pull requests.

SI Fabrik picks up tasks from your tracker, runs Claude Code or Codex on each one in its own sandbox, and hands back a branch and a pull request for you to review. Run it on your Mac, or on a server for your whole team.

Coming soon Read the docs →

Signed by Think Define Create LLC and notarized by Apple. No account, no telemetry.

Works with Claude CodeCodexGitHubClickUpDockerCloudflare Containers

How it works

You write the task. An agent does the work. You review the pull request.

  1. 1

    Tasks come from your tracker

    Use the built-in board, ClickUp or GitHub Issues. SI Fabrik watches for tasks that are ready and takes them in priority order, up to the number of agents you allow.

  2. 2

    Each task gets its own workspace

    A fresh clone on its own branch. Setup hooks install dependencies. Each project's WORKFLOW.md sets the prompt, runner, model and limits, and lives in the repo like any other code.

  3. 3

    An agent works it in a sandbox

    Claude Code or Codex runs turn after turn in Docker or a Cloudflare Container, until the task reaches its hand-off state. You watch every tool call live, or take over in a terminal.

  4. 4

    You get a pull request

    SI Fabrik saves the branch, pushes it and opens the pull request. The task moves to Human Review. Want changes? Comment and move it back to Todo; the next run reads your comments.

What you get

One place to run a fleet of coding agents.

Mission Control

Every running agent, the retry queue, token throughput, spend and recent runs at a glance, plus a system check for every tool the agents use.

Live transcripts

Stream each session's reasoning, tool calls and results, with turn, token and cost counters. Stop or retry any run, or pause a whole project.

Boards that agents work

A Kanban board per project. Drag a task to Todo and an agent picks it up; drag it back to stop.

Guardrails on spend

Daily budgets, per-task cost caps and turn limits. Failing tasks are parked and failing projects paused, so nothing burns money overnight.

Projects that work together

An agent can read another project's code, run its service next to the agent, and file follow-up tasks in projects that accept them.

Browser testing

Agents can drive Chromium inside their sandbox. You watch the page live and take control when you need to.

Security

Agents run commands without asking. So they never get the keys.

Coding agents are most useful when they can run anything. SI Fabrik makes that safe by controlling what they can reach, not by asking you to approve every command.

Every agent in a sandbox

Only the task's workspace is mounted. Read-only root filesystem, no Linux capabilities, pid and memory limits. Projects on your Mac run sandboxed unless you explicitly allow otherwise.

No credentials inside

The sandbox holds a placeholder. Model requests go through a relay that adds your real Claude or OpenAI credential outside the sandbox, so a compromised agent has no key to steal.

Allowlisted, read-only network

Agents reach only the hosts you allow. GitHub and package registries are read-only from the sandbox: clone and install, never push or publish. SI Fabrik pushes the branch itself.

A separate network per task

One task's agent, terminal and services can't see another task's. Encrypted connections must go to the host that was allowed, so an allowed host can't be used to reach another.

Accounts and an audit trail

On a server: viewer, operator and admin roles, authenticator codes for admins, a hash-chained audit log, and TLS with certificate pinning in the Mac app.

A signed, hardened app

Built with the hardened runtime, signed with Developer ID and notarized by Apple. Secrets live in your Keychain. SI Fabrik sends no telemetry.

How sandboxes work →   Server security model →

Local or remote

Start on your Mac. Move to a server when your team joins.

On your Mac

The engine runs inside the app. Your repositories, workspaces and Keychain stay on this Mac. Agents run in Docker Desktop or in Cloudflare Containers.

  • No account or server needed
  • Your own Claude or Codex subscription or API key
  • Menu bar status, notifications and deep links
Download the app

On a server for your team

harnessd runs the same engine on a Linux VM. Everyone connects from the Mac app, and secrets stay on the server.

  • Roles, authenticator codes and an audit log
  • Shared projects, boards and budgets
  • Clients never hold model or tracker keys
Get the server

Questions

Frequently asked

Is SI Fabrik really free?

Yes. The Mac app is free to download and use. The team server is free too; contact us and we'll help you set it up. You pay your model provider (Anthropic or OpenAI) directly for what your agents use.

Which models and subscriptions does it use?

Your own. Claude Code works with a Claude subscription or an Anthropic API key, Codex with your OpenAI account or key. You can also route model traffic through your own gateway, such as LiteLLM.

Does my code leave my machine?

Only to the places you configure: your model provider, your git host and your tracker. Cloudflare sandboxes run in your own Cloudflare account. SI Fabrik itself sends no telemetry and has no account to sign in to.

What do I need to run it?

A Mac with macOS 26 or later, Git and Claude Code. Docker Desktop is recommended so agents run sandboxed; Codex and the GitHub CLI are optional. See the install guide.

Can agents push to my repository or publish packages?

Not from the sandbox. GitHub and package registries are read-only for agents. SI Fabrik itself pushes the task's branch and opens the pull request, so every change reaches you as a reviewable pull request.

How do I get updates?

New versions are published here. Download the latest from the download page and replace the app in Applications; your projects and settings stay where they are.

Put your backlog to work.

Free for macOS 26 and later. Set up your first project in minutes.

Coming soon Your first project →