How agents are started and reached

An agent's availability is one switch and a checkbox for each way in, deciding what can start the agent and who can reach it.

An agent's availability is the Available switch and its checkboxes, in the Availability card on the agent's Versions & Availability tab. Every way in reaches the same agent, runs under the same limits, and needs its own box ticked.

Why it matters#

You build an agent once and reach it from chat, your product, a schedule or another system without copying it. You open only the ways in you mean to, and switching Available off closes them all at once.

What starts a run#

What starts it	Checkbox it needs	See
An admin in the Studio, on the agent's Chat tab	None	Build an agent
A person in the Hub	The Hub	Build an agent
An AI client such as Claude or Claude Code, over MCP	The Hub for members, none for admins	Use your agents from AI clients
Your own product, in an embedded session for one of your users	Embedded in your own product	Put an agent on your page
Another agent that has this one as a resource	Other agents	Build an agent
A schedule	Schedules	Start it from your server or a schedule
Another product's webhook, through a trigger	Triggers	Start it from your server or a schedule
Your own software, calling the Mindset API with an org API key	Outside systems	Start it from your server or a schedule

A schedule can start the invoice exceptions agent every weekday at 07:00 UTC, and your finance system can call it the moment an invoice fails its match.

Only agents start runs. To schedule a function, schedule an agent that calls it.

The Available switch and its checkboxes#

An agent is available only when a version is active, the Available switch is on and at least one box is ticked. Switching it on with nothing ticked ticks every box. Switching it off refuses every way in at once, and the boxes are remembered.

A change takes effect when you press Save. Availability sits outside versions, so activating or rolling back a version leaves it as it was.

A new agent starts with every box unticked. Builders tick only what their own wiring needs.

A schedule, trigger, API call or new embedded session without its box is refused with agent_not_available_here, naming the box. A member, an open embedded chat or another agent is told the agent doesn't exist.

Who can reach what#

Who	What they reach
Org admins	Any agent in the org, whatever its availability: in the Studio and over the administrators' MCP server
Members	Agents with The Hub ticked, in the Hub and from their own AI client through the users' MCP server
Your product's users	Agents with Embedded in your own product ticked, through a session your server creates

Wherever it's reached, the agent carries its own resources, and a write operation still needs its approval.