1. Set up your workspace
Three things have to exist before an agent can answer a question: an organization, a project to hold the agent, and a model credential for the provider you want to use.
Sign in
Section titled “Sign in”Go to chat.app.urai.dev and select Sign In. Authentication runs through Urai’s identity provider, so you sign in with your work email and come back to the console with a session that lasts 30 days.
The first time you arrive, the onboarding form asks for your name, your role, your industry, your team size, and a name for your organization. The organization name is the one thing here that other people see, so use your company or team name rather than your own.
If a colleague already created the organization, ask them to invite you instead. Signing up separately creates a second organization that shares nothing with the first.
Create a project
Section titled “Create a project”Projects hold agents, threads, and notes. Open Projects and select New
Project, then give it a name that matches the surface you are building for,
such as Support or In-app help.
You can rename a project later and you can add members to it from the project’s Members tab, so do not spend long on this decision.
Add a model credential
Section titled “Add a model credential”Urai calls model providers with your organization’s own API keys. Without a credential, the model list in the agent editor is empty and nothing can run.
Open Credentials from the settings area. Only owners and admins can see this screen; if it redirects you back to the chat, ask an admin to do this step.
Select Add Credential, choose a provider, and paste the key:
| Provider | What to paste |
|---|---|
| Anthropic | API key |
| OpenAI | API key |
| Gemini | API key |
| DeepSeek | API key |
| OpenRouter | API key |
| Vertex | Service account JSON, GCP project id, and location |
Keys are stored per organization and are never shown again after you save them. To replace one, edit the credential and paste the new key; leaving the field empty keeps the existing one.
Add at least one provider that supports function calling, because the agent needs it to use tools in step 3. Anthropic, OpenAI, Gemini, and Vertex models all qualify.
Check before moving on
Section titled “Check before moving on”Open your project and select New Agent. If the model dropdown lists models, the credential is working and you are ready for step 2. An empty list means the credential is missing or was saved on a different organization.
Next: create the agent.