Start here
Get started
Adapt is an open source company agent workspace: your team chats with agents that use the models, sandboxes, knowledge, and tools your company connects.
This guide takes a workspace owner from first sign-in to a first useful chat. Only a model is required; everything else can wait until a job needs it.
Choose where Adapt runs
- Adapt Cloud: a managed Adapt workspace, available by request during early access. Request one at adapt.com/early-access.
- Self-hosted: Adapt is MIT-licensed, and its source is published on January 12, 2027. Run it on your own infrastructure with PostgreSQL and the providers you choose. See Self-host for requirements and deployment.
Both run the same application, so the steps below are the same.
Create your account
Open your Adapt address and create an account with your email and a password. On a new install, the first account creates the workspace and becomes its Owner. Sign-up then closes: everyone else joins through an invitation link from an admin.
Name your workspace
Setup opens as a conversation with your company agent. With a company email address, Adapt looks up your company's site and fills in the workspace name and picture; change them if they're wrong. With a personal address, you name the workspace yourself.
The workspace starts with one agent, named after it. That agent is what your team talks to first; you can rename it, change it, or add more agents later. See Agents.
Next, choose how to continue:
- Let's continue walks through each step here.
- Copy instructions hands setup to a coding agent you already use. It copies a prompt with a setup key for Adapt's setup MCP server. The key reaches only the setup tools, and stops working after one hour or when setup is finished. The coding agent asks you for every API key; each step appears in Adapt as it lands.
Connect a model
A model is the only required step: agents can't answer without one. Choose a provider (Anthropic, OpenAI, OpenRouter, another hosted provider, or any OpenAI-compatible endpoint, including a local model) and paste its API key.
Adapt lists the provider's models with that key. When the provider has a recommended model, Adapt allows it and makes it the default for new agents. See Models for every provider and how to allow more models.
Pick goals and tools
Tell the agent what to take off your plate. Pick as many goals as you like or write your own; it uses them to suggest tools.
Then connect the tools your team uses, such as GitHub, Linear, or Google Workspace. What you connect here is shared: everyone in the workspace can use it through the agent. Choose I'm good for now when you're done, or skip the step. See Integrations.
Add the optional pieces
Setup offers three more steps. Each one can be skipped:
- Slack: lets your team mention the agent in Slack and get answers there. See Slack.
- Sandbox: somewhere to run code and make files, such as charts, spreadsheets, and scripts. See Sandboxes.
- Memory: a knowledge provider, so the agent keeps what it learns from chat to chat. See Knowledge.
Anything you skip is offered again at the bottom of Home, with a button to connect it.
Start your first chat
When setup finishes, Home opens a new chat with your agent. Ask it something real, like a question about a tool you connected or a task you'd normally do by hand.
While it works, the chat shows each tool call as it runs and whether it finished or failed. A call that failed partway, such as one that timed out, may still have taken effect, so Adapt never retries it automatically. Closing the browser doesn't stop a run; Stop does. By default, a run stops after five minutes. See Chats and tool activity.
Your web chats are private to you; admins and owners can't read them. Runs of an automation are the exception: anyone who can view the automation can open them.
Invite your team
In Settings → Members, choose Invite, enter their email, and send them the invitation link Adapt creates. Invitations default to the Member role: members can chat with agents and create their own, but can't change workspace settings. See Members and roles.
Next steps
- Write company instructions and skills every agent follows.
- Add documents your agents can search in Knowledge.
- Run an agent on a schedule with Automations.
- Learn the terms used across these docs in Core concepts.