fab.engineering
Place your company's AI agents on the org chart, manage them, and audit them.
fab.engineering is a source-available platform that runs on your own server. Every agent has a name, a role, a department and a responsible human; agents sit in the company structure like any other employee. Policies, approval flows and an audit trail come built in.
Who is it for, and where to start
| Who you are | Start here |
|---|---|
| Company executive / founder: I want to know what it does and how to set it up | First Steps → AI Setup |
| Department employee: I will work with an agent | Chatting with an Agent, Task Requests and Inbox |
| Department manager: I manage my team's agents and approvals | Agents, Change Requests |
| IT / installer | Requirements, Installation, Settings |
| Developer: I will work with opencode on my own machine | Developer Workstation |
| Department employee who wants to work in a terminal | The fab terminal client |
| Building or reviewing the platform | Technical, Developer Notes |
Three ways to work
Same server, same policies, same audit trail; three doors depending on the job:
| Surface | For | Status |
|---|---|---|
| Web interface | Everyone: chat, flows, approvals, administration | Ready to use |
Developer workstation (3pa + opencode) | Software developers | Ready to use |
fab terminal client | Departments such as finance, accounting, sales | In development: there is no real workspace runtime yet, see fab |
Core concepts
| Concept | Meaning |
|---|---|
| Company | The top-level unit. One installation can manage several companies; their data is kept separate. |
| Department | A unit of the company. It can belong to a parent department and has goals and policies. |
| Personnel | A person on the org chart. The type is either human or agent. |
| Agent | An AI model (e.g. Claude, GPT, Gemini) bound to a personnel record with its skills and policies. Every agent has a responsible human. |
| Skill | A tool the agent can use: web search, chart generation, a database query, an external API, and so on. |
| Policy | A written rule that applies to a company, a department or an agent. |
| Platform user | A person who signs in with an email and password. Added by invitation; the role decides what they can see. |
Principles
- On your own server. Your data stays with you; AI providers are reached only with the keys you define.
- Model independent. Anthropic, OpenAI, Google, Mistral, Qwen, local models (Ollama, LM Studio) and any OpenAI-compatible address can be used.
- Human approval. Agent configuration changes go through two-stage approval; handing a task from one agent to another needs human approval.
- Traceability. Important actions are written to an audit log. In the developer workstation layer the log is additionally a tamper-evident hash chain.
- Source-available. Source code is on GitHub; the license is Commons Clause + MIT: free for your own organization, while selling it or offering it as a hosted service to third parties needs a commercial license (see
LICENSE).
Try it on your own server
Go to Installation, then First Steps.