Answers for developers, team leads, and technical founders evaluating Cate.
Cate is a desktop application that orchestrates Claude AI coding agents into a coordinated development team. It connects to your issue tracker and git host, dispatches work from your board, and manages the full lifecycle: planning, coding, reviewing, and opening PRs. You retain final authority over what merges to main.
Cate orchestrates agents that write real code and open real pull requests. They implement features end-to-end — committing to isolated worktrees, running your guardrails, and submitting PRs for review. No code lands in main without a human merging it.
No. Agents work in isolated worktrees and open pull requests. Merging to main happens in GitHub following your existing safeguards.
Cate needs read/write access to your repository to create branches, commit code, and open pull requests. It also needs access to your project board to read and update issue statuses. The exact permission scopes depend on your issue tracker and git host.
Credentials are stored locally on the machine running Cate, using your keychain or OS-level encryption. They are not transmitted to Blue Ghost servers. Agents use credentials only to authenticate with your configured services (GitHub, Jira, etc.) on your behalf.
Cate is an Electron app, and it follows Electron's recommended best
practice for secret storage by using the
safeStorage
API. This delegates encryption to your operating system's native
secret store — macOS Keychain, Windows DPAPI, or libsecret on Linux.
The prompt appears once on first launch when Cate encrypts its first
credential. After you grant access, it works silently in the
background.
Agents run every guardrail defined in your project's
.cate/guardrails/
directory before opening a PR. Common gates include automated tests,
linting, HTML validation, and code formatting. If any required gate
fails, the agent fixes the issue before proceeding. You define the
standards once; agents enforce them every time.
After an agent submits a PR, a dedicated reviewer agent performs a structured code review—checking correctness, adherence to acceptance criteria, and guardrail results. If the review passes, the issue moves to "Human Review" so a human can do a final check and merge. If changes are needed, the issue returns to "To Do" for revision.
If an agent cannot complete a task, it posts a detailed handoff comment explaining what was tried and why it got stuck, then moves the issue back to "To Do" for triage. Bad code is caught at review: reviewer agents or humans request changes, the issue returns to "To Do", and the next agent picks it up with the feedback in hand.
No. Agent sessions start automatically when issues enter the corresponding part of the queue (e.g., a worker picks up a "To Do" issue, a reviewer picks up an "AI Review" issue). On-demand sessions — for planning, collaboration, or custom workflows — spin up when you need them and terminate when the session ends. You control the maximum number of auto-started worker and reviewer agents.
Cate supports Jira, Linear, and GitHub Projects. The tracker is configured during project setup — an agent walks you through the connection interactively.
About 15 minutes. You install the Cate desktop app, connect your issue tracker, and run the setup command — an agent interviews you about your project and configures everything for you.
Every time an agent plans, codes, or reviews, it writes structured
notes: what it tried, what worked, key decisions, and rationale.
These live in your issue tracker and in a shared research directory
(.cate/research/) committed to your repo. Future agents
— and your human teammates — read that context before
starting new work. Over time, agents make better decisions because
they build on what came before instead of starting cold.
No. Blue Ghost never sees your code — it stays between you and Anthropic. We don’t collect, store, or train on any of it. Whether Anthropic trains on your data depends on their terms and the plan you use — check Anthropic’s data policy for details.