What is Slab? Meet your AI agent workspace
What is Slab?
Slab is an open-source workspace for running AI agents as a team. Give each agent a role, put work in its queue, and connect the information and tools it needs. You can follow the progress, review decisions, and see what happened after a run finishes.
Think of a small company: people have responsibilities, work has an owner, and important knowledge has a home. Slab gives your agent team that same structure.
You can start with one agent and one useful job. Add more roles and connections as you need them.
Find your way around
| Part of Slab | What it helps you do |
|---|---|
| Agents | Decide who does the work and what they can use. |
| Work | Keep track of what needs doing, who owns it, and what happens next. |
| Docs | Give your team a shared home for company knowledge. |
| Sources | Bring in knowledge from your existing websites and repositories. |
| Runner | Carry out an agent's work through its configured AI runtime. |
| Runs | Follow each attempt and review its actions, approvals, and result. |
| Automations | Start recurring work on a schedule. |
| Blueprints | Set up a team from a reusable starting point. |
| Integrations | Connect agents to the tools your company uses. |
| Overview and Settings | See what needs attention and manage your workspace. |
Agents
Give someone the job.
An agent is an AI team member with a name, a role, instructions, and a set of permissions. You might create a researcher to gather information, an operations lead to organize work, or a customer communication agent to prepare replies.
Its role stays in place between conversations. When you send a message, assign work, or start an automation, Slab uses that role to begin a run. You choose the tools and knowledge available to it, so a researcher and a customer communication agent can have different access.
Agents can also hand work to other agents. That lets you split a larger job into clear responsibilities while keeping track of who owns each part.
Explore agent and runtime setup
Work
Make the next step visible.
Work is the shared place for projects and work items. Each item records a job to do, its owner, its status, and the conversation around it. You can follow progress, identify blockers, and see which results need review.
For example, “Prepare the customer onboarding guide” can be one work item. A research agent gathers the details, a writing agent prepares the draft, and you review the result. The assignment and its history give everyone a place to pick up where the previous person or agent left off.
Use Work when a job needs an owner and a clear outcome, especially when it spans more than one conversation.
Docs
Give the team something reliable to refer to.
Docs is your workspace's shared knowledge base. Keep product notes, operating procedures, customer guidance, and decisions in organized documents that people and authorized agents can read and update.
Documents have a revision history, so you can look back at earlier versions. You choose which collections an agent can access. A customer communication agent might need your product guide and support policies; a research agent might work with a different collection.
For an onboarding task, Docs could hold the approved product information and the finished guide. Work tracks the assignment; Docs keeps the knowledge it produces.
Docs inside your workspace holds your company's knowledge. The documentation you're reading here explains how to use Slab.
Sources
Bring your existing knowledge with you.
Sources connects information you already maintain elsewhere, such as a website, a WordPress site, or a GitHub repository. Slab synchronizes that content and makes it available to the agents you assign, with information about its origin and freshness.
Use Docs for knowledge you maintain in the workspace. Use Sources when the original content lives in another system and should continue to be maintained there.
Runner
Turn an assignment into a running agent.
Runner is the service that carries out an agent's turn. Slab sends it the instructions for the run and the allowed tools. Runner works with the configured AI runtime and sends progress and results back to the workspace.
A runtime is the software through which an agent uses an AI model and its tools. Choosing one determines how the agent executes its work. Runner provides the connection between that execution and the Slab interface you use to follow it.
You normally interact with your agents through the workspace. Runner works behind the scenes, while Work keeps your assignments and Docs keeps your documents.
Runs
See how the work happened.
A run is one attempt by an agent to carry out a request. It gives you a record of the messages, tool activity, status, and result, along with the usage and cost information reported by the runtime.
An agent can have many runs over time. A work item can also take several runs to complete: one to research, another to draft, and another to respond to feedback.
When an action requires approval under the configured policy, you can review it before execution continues. If a run fails or gets stuck, its record helps you understand where it stopped and decide what to do next.
Automations
Give recurring work a regular place in the day.
Automations start agent work on a schedule. You choose the agent, the instructions, and when the job should run. A weekly project summary or a morning review of open work can then begin without you typing the same request each time.
Each execution creates a run you can inspect. Connected email can also start workflows when you configure rules for incoming messages.
Start by running a job yourself and checking the result. Once the instructions and access are right, put it on a schedule.
Explore email-triggered workflows
Blueprints
Start with a team you can make your own.
A Blueprint packages a reusable setup, including agent roles, instructions, and supporting configuration. It gives you a starting point for a particular kind of work.
Review what the Blueprint includes, connect the services it needs, and adapt the roles to your company. You keep control over the agents and permissions it brings into the workspace.
Integrations
Let agents use the systems where your company already works.
Integrations give agents specific tools for connected services. You configure the connection and decide which agents can use it.
- Email lets an agent search or read messages, prepare drafts, and send through a connected account when its permissions allow it.
- Calendar lets an agent check availability and work with events, according to the provider and write policy you choose.
- Data connections bring information from services such as analytics tools into the agent's work.
- Custom tools connect read-only HTTP APIs or MCP servers. MCP is a standard way for an application to make tools available to an AI agent.
Connecting a service and granting an agent access are separate choices. You can give one agent permission to read a mailbox and another permission to draft replies.
Explore integrations · Email · Calendars
Overview and Settings
Know what needs you, then decide what to change.
Overview brings together the state of your team's work. Check what is active, what is blocked, and what is waiting for review or approval before deciding where to focus.
Settings is where you manage workspace connections, runtimes, and configuration. Agent permissions and spending controls help you define the boundaries within which your team operates.
Optional memory can help an agent carry preferences and corrections between conversations. Keep project status in Work and company reference material in Docs, so the team has a shared record to return to.
One job, from request to result
Imagine asking your team to prepare a customer onboarding guide. With the right agents, connections, and permissions configured, the flow can look like this:
- You give an agent the outcome: explain what the guide should cover and who it is for.
- Work holds the assignment: the job has an owner, progress, and a place for feedback.
- Docs and Sources provide context: the agent consults the product information it is allowed to access.
- Runner executes the work: the agent uses its runtime and available tools, delegating parts when appropriate.
- You review the result: Runs shows the execution record, and the work item keeps the review attached to the assignment.
- The team keeps the useful output: the approved guide can live in Docs, ready for the next person or agent who needs it.
That is the role of Slab: connect the people, agents, knowledge, and actions around a piece of work you can follow.
Ready to try it? Build your first AI agent workflow with a sample brief, a Work assignment, tool approvals, and a result you can check.
Choose how to run Slab
Self-hosted Slab is available today. The installer brings the services together on your server. You control the infrastructure and use the included tools to manage updates, backups, and recovery. The software is open source and free to self-host; you pay for your server and the model or third-party services you use.
Slab Hosted is accepting early-access requests. It is the planned managed option for teams that want Slab without running the infrastructure themselves. Join the waitlist using “Request Hosted access” at the top of this page.