Custom AI agents for your workflows.

Custom AI agents are agents you build in Gleap with your own instructions, tools and triggers. Each agent gets a job, the tools it needs (MCP integrations, HTTP APIs, knowledge search, decisions, transfers, customer replies), a model and a cost cap, and runs in your tickets or on its own schedule. They live next to Kai, Kai Resolve and Kai Code on the same platform.

Instructions in plain text with @mentions of its tools, a knowledge scope from your help center and articles, tools, triggers and a model. Tools include Stripe, Slack, Linear, HubSpot, Jira, GitHub and Attio via MCP, a custom MCP server, API calls, a dynamic API with an OpenAPI spec, web search, workflows, SDK tools that run in your app, ticket status changes, scheduling, collecting data into properties, code investigation, code proposals and Kai Code.

No. Instructions, tools, triggers, model and access are configured in the dashboard. Connecting a custom MCP server or API needs its URL and credentials. SDK tools that run inside your app need a handler registered with Gleap.registerAgentTool, which is engineering work.

Yes. The agent page in Gleap is the builder: name and avatar, instructions, model, tools, triggers, quick actions and access on one side, a test chat on the other. Start from one of 14 templates such as Internal Support Agent, Feature Request Tracker, SLA Watchdog or Incident Commander, or from a blank agent.

Four trigger types: on a schedule, on an SDK event such as a page view or custom event, when a connected tool fires an event, or an API call from your own systems with your service account token. Teammates can also address a published agent from any ticket with an @mention, and a workflow can pass a conversation to it.

Gleap, Slack, Attio, Linear, HubSpot, Jira, GitHub and Stripe are ready to connect via MCP, and over 1,000 more integrations are searchable in the tool browser. You can also connect your own MCP server or an HTTP API. After connecting, you see every exposed action and turn off the ones the agent should not use. See integrations.

Yes. Ask for a decision shows a teammate a card with named options and waits for the answer. Request human input opens a free-form task for anything that needs a person, such as billing or account changes. Transfer hands the conversation to a human or another agent and assigns the ticket.

Only through the Reply to customer tool. Without it, everything the agent writes stays in the internal conversation your team sees. You decide per agent whether customer replies are part of its job.

Auto routing (Medium or Max) or one of 20 named models from Anthropic, OpenAI, Google and xAI, plus open models such as Qwen, GLM, Kimi and DeepSeek via OpenRouter. Model selection needs Pro or Enterprise; each agent has a maximum cost per run. The model list changes over time. See pricing.

Kai is the default support agent and Kai Resolve the managed second level; both are agents on the same platform. A custom agent can delegate a task to another published agent, start a read-only code investigation and propose a code fix that a teammate approves before Kai Code runs it.

Visibility is set per agent: everyone, only the creator or specific people. Settings are managed by admins and managers, the creator or specific people. Admins always have access. Tool credentials are stored per tool and authorized with OAuth or headers.

Yes. Publish the agent and test it in chat while its triggers stay disabled. Review the answers and tool calls, then enable the trigger. Unpublish pauses it at any time.

Custom agents are part of the Gleap platform. AI usage is billed separately by tokens and model, and each agent's max cost per run caps what a single run can spend. See current plans and AI billing.