Gleap has two kinds of AI agents, and they sit next to each other in the same list. Kai is there from the start and is labelled the default agent. Custom agents are ones you create.
The distinction that matters is not how clever each one is. It's what they're built around: Kai is built around your content; a custom agent is built around instructions and tools you give it.
One per project, always active, and it answers customers in conversations. Everything it says has to come from the knowledge you connected — help center articles, crawled pages, documents. You configure it centrally: name, avatar, tone, response length, language, model.
Because it's tied to your content, improving Kai usually means improving what it can read, not rewriting instructions.
You write the instructions, pick the model, and attach tools — calls into your own systems, MCP servers, actions that run in your product. An agent can look something up, change a record, or carry out a sequence of steps, which is the part Kai deliberately doesn't do.
They also start differently. A custom agent runs when something starts it: on a schedule, on an event in your app, on an event in a connected tool, or from an API call. Kai runs when a customer writes.
A new agent begins as a draft and does nothing until you set it live. You can also decide who on your team may use it — everyone, only you, or specific people you pick.
This isn't a choice you make once. A workflow can hand a live conversation from Kai to a specific custom agent — useful when a question turns out to need a lookup in your own system rather than an article. The agent has to be live for that to work; a draft agent is skipped.
Kai can also be given tools of its own, so the line is not absolute. The difference in practice is one of default posture: Kai reaches for your content first; a custom agent follows the procedure you wrote.
Ask what the job actually is.
Answering a question whose answer is documented, or could be — Kai. If it's answered wrongly, the fix is usually an article.
Reading or changing something in another system — an order, a subscription, a booking — a custom agent with a tool for it.
Work that isn't triggered by a customer at all, like a nightly summary or a reaction to an event — a custom agent, since Kai only responds to conversations.
A procedure with fixed steps and a specific tone — a custom agent, where the instructions are yours to write.
