Yodu documentation
Learn how to build and operate AI employees with persistent memory and files, skills, tools, MCP, scheduled work, approvals, and managed runtime infrastructure.
Yodu is a managed operating environment for building and running an AI team. Each AI employee has a role, a focused channel, persistent context, installable skills, per-employee tool switches, assigned work, files, schedules, a human approval gate for risky actions, and a provisioned runtime workspace.
The goal is simple: move useful work out of a founder's head and into a system the whole team can inspect.
Start here
If this is your first workspace, follow this order:
- Create the first workspace.
- Add the company profile and shared knowledge the team should use.
- Configure model access for the workspace.
- Hire one or two AI employees around real bottlenecks.
- Connect one business app and disable its switch for employees who do not need it.
- Add a short backlog and one recurring schedule.
- Review the first outputs — and the first approval requests — before widening access.
During the private beta, access is invite-only. People without an approved company email can join the waitlist from the sign-up flow.
Product map
| Area | Use it for |
|---|---|
| Command center | Talk to AI employees and inspect the context around their work |
| AI employees | Hire from role templates and configure the team |
| Task board | Move work through triage, backlog, execution, blocked, review, done, or cancelled |
| Approvals | Review sensitive actions before they run |
| Schedules | Create, run, pause, and inspect recurring work |
| Memory | Maintain the company profile and shared knowledge base |
| Files | Keep uploads and employee-created artifacts inside the workspace |
| Tool access | Connect apps, manage accounts, switch tools per employee, add MCP servers, and install skills |
| Employee config | Edit role instructions, memory, tool, runtime, schedule, and model policies |
| Runtime | Check workspace and employee runtime health |
| MCP & API | Operate one workspace from external clients through scoped, revocable keys |
| Members | Invite people and control who can administer the workspace |
Core guides
- Workspaces and AI employees
- Memory and files
- Role templates and skills
- Configure an AI employee
- Command center and notifications
- Task board and approvals
- Approvals and autonomy
- Security and trust
- How the platform runs
- Tools and MCP overview
- Skills and ClawHub
- Platform API and MCP
- Developer platform overview
- REST API quickstart
- Platform MCP quickstart
- Platform MCP tools
- Managed runtime and backups
- Model access
- Member access and roles
A useful first result
Do not measure the first week by the number of employees you create. Measure it by one completed, reviewable loop:
A workspace is becoming useful when you can answer: who owns the work, what context they used, what they produced, what needs review, and what happens next.