Documentation

Admin guide

Connect your systems once, add your members with the right roles, and set the budgets. Everything is in the control panel.

The first steps

The control panel home shows a setup checklist until it is done:

  1. Connect Jira or Linear.
  2. Connect Bitbucket, GitHub or GitLab; Confluence, Magento, Akeneo and Figma are optional.
  3. Add members with their roles.
  4. Switch on a first routine.
  5. Share the Get started page with your team.

Systems

On Systems, connect each system once with an address and an access token. Each connection has an allow-list, such as Jira projects, repositories, Figma files or Confluence spaces; tools reach nothing outside it. Test a connection after saving: a failed test says what to fix.

  • Tokens are encrypted before they are stored and are never shown again; you can only replace them.
  • Use read-only tokens wherever a system offers them. A write token is only needed for writes you want to allow, such as creating tickets.
  • Branch pushes for autonomous development need a Bitbucket, GitHub or GitLab token that may write, and Allow branch pushes switched on.
  • Files that hold secrets, such as .env files and keys, are never read, whatever the allow-list says.
  • For the scraper watch: Server logs connects over SSH with a key made for Gunnie.ai, which the server only lets run the log analyzer, and checks the server’s host key; Cloudflare needs a token that can only read the zone’s analytics. To carry out approved blocks, connect Cloudflare blocks with a separate token that may edit the account’s filter lists (Account, Account Filter Lists, Edit), and on every zone the shop answers on, add a custom rule yourself, at the top of the custom rules, that blocks the addresses on Gunnie.ai’s list. Test checks each zone for it. Gunnie.ai never creates or changes rules. To block a company’s users, connect Magento account blocks with the shop’s address and an integration token. Without these connections, approving a block only records it.

GitLab as the code host. Connect GitLab with a group access token (role Reporter, scopes read_api and read_repository), the group as in its address, and its projects. Merge requests are what the tools call pull requests. Pushes by autonomous development also need write_repository and role Developer, and Allow branch pushes switched on. With more than one code host connected, Bitbucket is used first, then GitHub.

Linear instead of Jira. Connect Linear with a personal API key of a service member who is in the teams Gunnie.ai may see, and pick those teams. The same tools and routines work: a Linear team is what the tools call a project, its active cycle the sprint, a Linear project a release, and the estimate the story points. Linear has no custom fields, so acceptance criteria are read from the description, pre-refinement cannot write developer notes, and comments are visible to the whole workspace. With both connected, Jira is used.

Members and roles

On Members, add people by email address and give each a role. An invited member logs in with that address and is linked straight away. Change a role at any time; it applies to their next call. Deactivate someone and their next call is refused; their calls stay in the call log. There is always at least one active admin.

What you can ask lists the tools of every role.

Your organisation’s rules

On Skills, add rules that apply only to your organisation, for example what your team checks in a code review. They sit on top of Gunnie.ai’s own base, and the standards in your repositories (AGENTS.md, .ai/) take precedence over them. Every change becomes a new version, and a review records which versions it used.

Budgets, usage and value

  • Reviews shows the monthly budget, what each review cost, and AI usage per month as a CSV.
  • Each routine has its own budget; from 80% it shows a warning and at 100% it stops until next month.
  • Value shows what Gunnie.ai did per routine, the cost, and an estimate of time saved. Set your own minutes per item there; the estimate always shows its assumption.

Call log and big screen

The call log keeps who called which tool, when, the outcome, the duration and the assistant, and never the question, the arguments or the answer. The big screen shows calls, routine runs and approvals live, for a screen in the office.