What is Gunnie.ai, in plain words?
An AI delivery team that works inside the systems you already use. It reviews code, answers questions from your web shop, product data, tickets and code, and prepares, estimates and builds tickets. Your people check and approve the work. Gunnie.ai decides what the AI may reach and writes down every step.
What problem does it solve?
Developers lose a large part of their week to groundwork: reading tickets, collecting context, estimating, reviewing, writing release notes. AI can do much of that, but most companies hesitate to give it access to their shop, tickets and code, because nobody can see what it touches. Gunnie.ai does the groundwork and keeps the AI inside your rules, so your team spends its time on decisions.
Who is it for?
Companies with a development team, in house or hired, that work from tickets and pull requests. Developers get the reviews and the groundwork; product owners, support and sales can ask plain questions about products, orders and tickets without needing a developer.
How quickly can we start?
Asking questions and starting code reviews works as soon as your systems are connected. An admin connects each system once with an access token, the admin adds colleagues on Members, 5 to start with, and each adds one address to their AI assistant. An admin switches on the routines that run on their own, such as pre-refinement and estimates, under Routines, each within a monthly budget. During the beta we help you set them up and tune them to how your team works.
Where do we get help setting it up?
The control panel home shows admins a checklist that ticks itself off as each step is done: Claude connected, your tracker and code host, members and, where it applies, payment, with what each system’s token needs and its last check. In Claude, ask what is left to set up or how something works: it reads your setup and searches the Gunnie.ai docs, and links to the page. If that does not answer it, Claude passes your question to us, or email support@gunnie.ai.
Do we have to change our tools or the way we work?
No. Gunnie.ai works with your Jira or Linear, Confluence, Bitbucket, GitHub or GitLab, Magento, Akeneo and Figma as they are, and the routines follow your process, your definition of done and the standards in your repositories.
Can we add other systems through their own MCP server?
Yes. An admin adds up to 10 MCP servers over HTTPS. Every tool on them starts off: the admin turns tools on one by one, picks the roles that may use each one and whether it reads or writes. Writes wait for approval like any other change. When a server changes a tool, Gunnie.ai holds it until an admin approves the change. A server can use one secret for everyone, or each member signs in there with their own account, so its tools work with that person’s own permissions. Servers that run on someone’s own computer are not supported.
Which AI assistants does it work with?
Claude, ChatGPT and other assistants that support MCP, the open standard for connectors. Your team keeps the assistant it already uses.
Does it work with Slack?
Yes. Your admin adds the Gunnie.ai Slack app and picks a channel: the routines post their results there, so your team sees what the AI did without opening another tool. Approvals stay in Gunnie.ai; each post links to its result there. Nothing is written without an approval. The app is not in the Slack Marketplace yet.
Will it replace our developers?
No. It takes the groundwork off their plate. People still decide what gets built, approve every change and decide what ships.
Who checks Bob’s first tickets?
A Gunnie.ai engineer, for a new organisation's first 5: once before Bob starts and once before his change reaches you, within one working day. Until then you see “Being checked by Gunnie.ai” with the expected time. The engineer sees the ticket, the change, its test output and the report, nothing else of your systems.
Can Bob install our private Composer packages?
Yes. An admin saves a registry key for Magento Marketplace or Private Packagist on the Bob routine. It is stored encrypted and never shown again. The sandbox uses it only for the first line of your install, when that is one plain composer command, before any change of Bob’s arrives. On that line Composer, the plugins your root package allows and its scripts hold the key, and so does anything they start, so use a read-only key.
Can the AI change things in our systems?
Only what you allow. Each role gets its own tools, and allow-lists limit which projects, repositories and files are within reach. Most tools only read. Anything that changes something, like creating a ticket or a note a routine wants to add, waits until it is approved in Gunnie.ai, whatever AI assistant was used. Per tool, an admin can choose who approves, require two approvals or give a change longer to be decided; approval can never be switched off. The one exception: a code review routine an admin has set to do so sets its verdict as a build status by itself. Files that hold secrets, such as .env files and keys, are never read. The code review never posts on a pull request by itself.
Is our data safe?
Gunnie.ai reads your systems live and keeps no copy of your products, tickets or code; only a routine’s result, a code review’s findings, a change waiting for approval, the SKUs that catalog health found or its open tickets name, per run the token counts of each AI response with the ticket keys it named, per ticket autonomous development took on its key, outcome and cost, for a ticket Gunnie.ai checks who decided and their note, and the anonymised test dumps an admin uploads, encrypted, are kept. The call log, routine reports and decided approvals are deleted after 30 days, or the period your admin chose (7 to 365 days). Counts of what each run did are kept to show your history; they hold no ticket text, code or arguments. The call log records who did what and when, never the question or the answer; only a question asked of Gunnie.ai’s own help is kept, 30 days at most, to improve the docs. Names, email addresses, phone numbers and IP addresses from your tickets, code hosts and shop are masked for every role but admin on new connections, until an admin lets a role see them. Files that hold secrets are never read. Access tokens and sign-ins are stored encrypted, the server runs in a European data centre, and nothing is used to train AI models.
Can we log in with Microsoft or our own single sign-on?
Not yet: logging in with a Microsoft work account, or single sign-on through your own Microsoft Entra ID directory, comes with the live version of Gunnie.ai. Until then, people log in with a link sent to their work email.
Which AI runs the code review and the routines?
Claude, through Gunnie’s own account; on request, an organisation’s Claude runs in the EU. AI use is counted per organisation, within a monthly budget you set, and under Anthropic’s commercial terms your data is not used to train models.
We are an agency. Is Gunnie.ai for us?
Yes. It works for your team, not instead of it: your developers keep every client and the final say on every change, and the groundwork runs in the background. Each client gets its own organisation, and one overview shows them all side by side.
How do we report a bug or share an idea?
Gunnie.ai is in beta, so we want to hear it. Every member sends a bug, an idea or a question under Feedback in the control panel, or asks Claude to pass it on; you can also email support@gunnie.ai. The Gunnie.ai team reads every message and answers by email.
What does it cost?
For now we agree the price with each company. Code reviews and routine runs show their cost one by one, within a monthly budget you set yourself, and every routine has a budget of its own.