Frostbyte agent
Chat with an in-app agent that reads your project and proposes areas, releases, tasks, and edits for you to approve.
The Frostbyte agent is a chat panel that lives in the corner of every project page. It reads your project (tasks, releases, areas, signals, and recent activity) and helps you set it up and keep it moving. It's the fastest way to go from an empty project to a real plan without filling in every field by hand.
Where it fits
The agent is for work that's easier to describe than to click through: drafting your product context, mapping out areas, planning a first release, or making sense of a backlog that's drifted. You stay in control. The agent reads freely, but it only changes your data when your message asks for a change, and destructive actions always wait for your approval.
Setting up a new project
Each onboarding task offers a one-tap handoff to the agent. Choosing one opens the panel with the question already framed:
- Enter product context: describe what the product does, who it serves, and the problem it solves, and the agent writes clear product context onto the Product page.
- Create your areas: describe the product and the agent proposes three to six stable areas, with the reasoning behind each boundary.
- Plan your first release: give it one shippable outcome and the agent drafts the release and the smallest set of tasks needed to reach it.
You can also just open the panel and ask. On the dashboard, releases, tasks, and Signals pages it offers quick starters suited to where you are.
How changes work
The agent never edits anything unless your latest message asks for a change ("create a task to…", "move that to the next release"). When you do ask, low-risk changes apply immediately and the agent tells you what it changed. The live board is your receipt.
Destructive actions are the exception. Deleting a task or subtask and dismissing a signal always show a proposal card first: approve it and the change lands, reject it and nothing happens. Proposals expire after a short window, so an old suggestion can't be applied by accident. If one lapses, ask the agent to propose it again.
Changes the agent makes are recorded in the activity feed, so you can always audit what it did.
Cost
The agent is included on every plan and runs on the shared monthly AI credit pool. Each message uses credits based on the model and the amount of text processed, settled to actual usage afterwards, so cost varies per message. Reading your project counts too, because it still invokes a model. The opening message in a guided handoff is display-only and costs nothing; you're only charged once you reply.
Availability
The agent is available on Free with the plan's smaller monthly credit pool. When the pool runs out, agent messages pause until the next reset, or until an upgrade adds a larger pool. Chat history is saved per user and project in your browser, so reopening the panel picks up where you left off.
What's next
- Define the brief the agent works from: Product.
- Hand ongoing coding work to Claude Code or Codex: Agent plugin.
- For the other AI actions and their credit costs: AI overview.