Set up projects, tasks, and workflows by describing what you need.
Kovca's agent works across your whole workspace. Open it from any console page with Ask Kovca in the sidebar or Ctrl+J (⌘J on a Mac). It opens beside the page you're on and keeps the conversation as you move around, so “add a review column here” works on the project you're looking at. Ask Kovca on a project, task, or the Workflows page starts a message about it. Its full page (Conversations with Kovca in the command palette, or the panel's expand button) lists your earlier conversations. In a workflow's Studio, it works on that workflow: you watch its edits on the canvas as it makes them, it knows the step you selected and your last test, and Undo takes its changes back. Fix with AI on a failed test or run asks it to fix the problem.
For a bigger request, such as “set up client onboarding with a welcome email when a task reaches Kickoff”, it first proposes a plan: the projects with their columns, the starting tasks, the workflows, and how they connect. Nothing is created until you choose Approve and build. Then it builds the plan step by step and checks each step off. Or tell it what to change.
It creates and changes things freely, and every change shows in history as Kovca agent (for you). It asks you to approve these, and does them only when you do:
It can't connect app accounts, change members, API keys, or billing, or turn workflows on or off. It tells you where to do those.
The agent can do only what you could do yourself (see roles). Members and up can use it: members work on projects and tasks within their project roles, and admins and the owner also build, publish, and activate workflows. Viewers can't use it. Changes it makes to tasks start task-triggered workflows exactly as if you made them.
kovca agent is the same agent (see the CLI), and your conversations are shared between the console and the terminal. There, what it can do is your CLI key's access limited by your role.
The agent is free within the assistant's daily limits. Past them, it uses your workspace's agent budget, like the Studio assistant (see Billing). AI steps it runs in tests are billed like any test.