Tools are grouped below by what they let an agent do. The picker also searches a much larger catalogue of third-party integrations, so if the service you need isn't named here, search for it before building an API call by hand.
Connections that are already built, so you authorise rather than configure.
Gleap — Gleap's own tools, for agents that work with your workspace.
Slack — channels and messaging.
GitHub — repositories, issues and pull requests.
Jira and Linear — issue tracking and project management.
HubSpot and Attio — CRM data and contacts.
Stripe — payment information.
API call — one HTTP request to an endpoint you specify. The workhorse: precise, and you decide exactly what the agent can reach.
Dynamic API — hands the agent a whole API to explore rather than a single endpoint. Powerful when an API is large, and correspondingly less predictable.
Custom MCP — connect your own MCP server and expose whatever it offers.
Frontend tool — run an action inside your app through the Gleap SDK, in the user's own session.
Reply to customer — post a message to the customer on the connected ticket. Without this, an agent's output stays internal.
Ticket status — move tickets between your status lanes.
Collect data — save what the agent gathered into ticket and contact properties, so it's still there afterwards.
Knowledge search — search your knowledge base, optionally limited to specific tags. This is how an agent gets the grounding Kai has by default.
Transfer — hand the conversation to a human or to another AI agent and step out of it.
Request human input — pause and ask a teammate an open question: research something, approve something, decide something. The agent waits and continues with the answer.
Ask for a decision — the structured version: a card where a teammate picks between named options. Use this rather than an open question when the choices are known.
Delegate — give a task to another agent and get the result back, staying in charge. This is how specialised agents are composed instead of built into one large one.
Workflow — trigger one of your automation workflows.
Kai Code — investigate and fix issues in connected repositories.
Code investigation — a read-only look at a ticket against your repositories, run in the background.
Propose code fix — produces a proposal a human approves, which can then start a coding session. The approval step is the point of it.
Scheduling — lets the agent schedule its own future run, for deadlines, reminders, and follow-ups. (KB lists "On a schedule" as a trigger, not a tool — check whether this belongs here at all.)
Web search — search the web. Remember that anything it finds is outside your control, so keep it away from agents that answer factual questions about your product.
