Creating a client
Open your MSP dashboard
Navigate to your MSP in the dashboard. You’ll see a list of all your existing clients.
Click New Client
Click the New Client button. Enter the client’s name (displayed in the dashboard) and a slug (a short, lowercase identifier used internally, e.g.
acme-corp). The slug must be unique across your clients and can only contain lowercase letters, numbers, and hyphens.The client detail view
Click any client’s name to open its detail page. The detail view is organized into tabs:Overview
Operating mode, agent status, and client management actions including delete.
Integrations
Connect and manage third-party tools the agent can use for this client. Tool policies for each integration live on the integration’s detail page.
Activity
A full audit log of every tool call the agent has made.
Prompt
Client-specific prompt instructions layered on top of your MSP prompt.
Diagnostics
Agent container health, logs, and resource usage for troubleshooting.
There is no top-level Policies or Sessions tab. Tool policies are configured per integration — open the integration from the Integrations tab. Conversation transcripts are surfaced in the Activity log alongside tool calls.
Client status
Two status fields appear on the Overview tab:| Field | What it means |
|---|---|
| Status | The agent’s current state: active, provisioning, pending, creating, error, or destroyed |
| Deploy status | The last config push: idle, deploying, or error |
creating → provisioning → active) while the agent is being set up. The dashboard polls every 5 seconds and updates automatically.
If Deploy status shows an error, the error message appears below the badge so you can see what went wrong.
“Provisioning” means the system is setting up the agent container on your MSP’s server. This typically takes under a minute. No action is required on your part.
Operating mode
Each client can run in one of three modes, configurable from the Overview tab:- Plan
- Execute
- Inherit
The agent reads the request, analyzes it, and writes its plan as an internal note on the ticket — then stops. It never executes any actions. Use this when you want to evaluate the agent’s judgment without it making changes.