On this page
How to: create your first agent
- 1Open Agent Center → Agents.
- 2In the Starter fleet section, click Add on a role. For your own agent, use Create agent.
- 3Open the new agent's details. Check Tools: AI and Complete tasks should be on.
- 4Under Budget (per agent), check that the Token limit is above 0.
- 5Assign the agent a task in a case, as you would a person. See what happens.
- 1Create agent: your own agent with the wizard
- 2Starter fleet: eight ready-made roles
- 3Add provisions a role
| For an agent to work, it needs … | Out of the box |
|---|---|
| an AI provider | Met: Bricksta's central access |
| the AI tool | On for every new agent |
| a budget above 0 | 500,000 tokens per month |
| a free seat under the agent cap | Ten active agents per organization |
With the Manage members / roles permission you can only view agents. The page then shows “View only” at the top.
Which starter-fleet role fits?
The starter fleet has eight prepared roles. All share the same tight permissions; only the persona differs. A colleague that already exists shows “Created” and is not made twice.
| Role | What for | Not suitable if … |
|---|---|---|
| Process lead | Completes a case's tasks promptly. A good lead for missions. | … it should lead missions without you granting Run mission. |
| Process curator | Works recurring tasks and spots patterns for new templates. | … you want to automate one specialist task. |
| Account researcher | Gathers customer context for handovers. | … the research needs sources outside Bricksta. |
| CRM sync agent | Works tasks coming from connected systems. | … no integration is set up. |
| Communications agent | Drafts and sends messages. The only one that brings Send. | … nothing should go out. |
| Agent builder | Prepares the setup of further agents and templates. | … it should configure things itself. No agent may. |
| Knowledge-retention twin | Preserves process know-how. Starting point for a digital twin. | … there is no real person behind it. |
| Platform concierge | Explains the application and helps people use it. | … it should do process work. |
No role gets Run mission, Work mission tasks or Send messages at creation. You grant those later, one by one.
Build your own agent
Create agent opens a wizard with a start screen and four steps. Only the last button creates anything. Cancel before that and nothing is created.
| Step | What you enter | Good to know |
|---|---|---|
| Start | Start blank or Start from a template | The template fills everything from a starter-fleet role. You can change it all afterwards. |
| 1 Identity | Display name (required), role label | Everyone sees the name on tasks and in the activity trail. It should name the remit. |
| 2 Persona | Task, tone and boundaries | Empty works too; the agent then works unspecifically. More in Personas and digital twins. |
| 3 Tools | AI, Complete tasks, Send | With no tool you get an agent that executes nothing: “No tools (passive)”. |
| 4 Budget and preview | Monthly token limit | At 0 the wizard warns you. The preview shows what is about to be created. |
Throughout, the wizard shows how many seats are free. Once the cap is reached, Create agent stays disabled.
Which tools do you grant?
Tools decide what an agent may do. You find them in the agent's details under Tools and apply them with Save tools. You grant each one individually.
| Tool | Allows | Without the tool |
|---|---|---|
| AI | Thinking and drafting with the AI model | The agent can do nothing at all. |
| Complete tasks | Completing assigned tasks | It only proposes; the case stays unchanged. |
| Send | Carrying out a send task from the template | It sends nothing. |
| Run mission | Breaking a mission goal into a plan (lead) | The mission does not start. The lead shows “not granted”. |
| Work mission tasks | Working through delegated mission tasks | Its tasks stay open and are listed separately in the report. |
| Send messages | Producing send proposals inside a mission | Switches Work mission tasks on automatically. |
Whether a message needs approval is decided by the send task, not the tool. Its Trigger is Confirm before sending (the default) or Send automatically. Send proposals in a mission always need approval.
An agent is never an administrator. No tool allows managing roles, members, templates or other agents. The persona governs how it works, never what it may do.
How much budget does an agent get?
The budget is a monthly token allowance per agent, preset to 500,000. You change it in the details under Budget (per agent) with Token limit and Save budget. If it says “No per-agent budget set yet”, the agent has none of its own.
Budget 0 is not a pause: the agent takes tasks on, every run is rejected and the task stalls as an incident. To put an agent on ice, use the Agent active switch at the top of its details.
The budget counts tokens, not euros. How it behaves in operation is in Budget, cost and runs.
What if all seats are taken?
Your organization may have at most ten agents active at once. The wizard shows “x of y active agents used — z free”. Once the cap is reached, the wizard and the starter fleet are blocked.
| Way out | What happens |
|---|---|
| Deactivate an agent | The seat is free immediately. Its history is preserved. |
| Have the cap raised | Contact your Bricksta representative. You cannot raise it yourself, not even with Manage organization. |
Use your own AI key
Out of the box, agents think through Bricksta's central access. You store your own key under Settings, not in the Agent Center. It applies to the whole organization and to almost every AI feature, including the AI step in a template.
Enter the provider, model and key there and check the connection with Test connection in its card's menu. Then mark it as the default. You do not need to touch the agents.
| Field | What for |
|---|---|
| Test connection | Checks exactly this connection. The card then shows when it was checked and whether it worked, or else the kind of error. |
| Own endpoint (optional) | OpenAI only: a self-hosted, OpenAI-compatible model such as vLLM, Ollama, LM Studio or LiteLLM. Empty means api.openai.com. |
| Model for structure and tasks (optional) | AI templates only: a stronger model builds workflows, phases and tasks, the model above does the rest. Not offered for Azure OpenAI. |
An own endpoint must be publicly reachable over https and must not carry a port. Bricksta rejects private networks and localhost. A model inside your company network needs a public https entrance in front of it.
Where your key does not apply
- In these places the step quietly runs through central access, with no notice in the interface:
- Missions, twin material and the persona suggestion: only keys from OpenAI and Anthropic.
- Image recognition of scanned PDFs: only keys from Anthropic and Google.
- If a step reads a file along, such as prefilling from a PDF: not with OpenAI and Azure OpenAI.
- If you set up the key because of rules on where data is processed, check these routes one by one.
Which channel does an agent send through?
The channel belongs to the send task in your template, not to the agent. With nothing specified, it is email.
| Channel | Reaches | Setup | Delivery |
|---|---|---|---|
| Everyone, subject and body in full | None | “Sent to …” becomes “Delivered to …” or an error. | |
| Teams | Internal members only, just the subject and a link to the case | Enablement under Integrations | Stays at “sent”. |
| Outlook | Everyone, subject and body in full | Agent: link to a person and enablement. Person: own Microsoft account under Settings → Connections | Stays at “sent”. |
Teams and Outlook from an agent run are barely proven. Send a harmless message to yourself first. Outlook has no Send automatically, so set the Trigger to Confirm before sending beforehand.
Enabling Outlook
- Under Integrations, on the Microsoft Outlook card, you enter the tenant ID and flip the switch.
- Before that, your Microsoft administration must allow the Bricksta app to send on behalf of your mailboxes.
- It confines that permission to the approved senders via an Exchange access policy. Otherwise the app could send from any mailbox.
- Your IT gets the technical steps from Bricksta.
Once your agent runs, continue with Running missions.