How to Connect AI Agents to GitHub When You Don't Own the Repo
You can still use AI agents when an agency owns your GitHub repo. Leave the agency in charge of the repository, let it hold access, review each pull request, and do the merge, and route every approved change through the ticket queue it already works from. The agent proposes changes; the people who own the code decide what ships.
Many ecommerce teams hand their storefront code to an outside agency, and that agency is often the only party with access to the repository. When a team wants to try AI agents, two worries come up fast: we cannot connect something we do not own, and pushing work through GitHub would pull it out of the agency's priority queue. Both have workable answers, covered below.
TL;DR
- Changes reach your live store through a pull request that the repo owner merges, so the agency stays in control.
- Pick one operating model: the agency connects and merges, a named engineer is delegated, or you hand over findings only.
- Put each approved change in the agency's existing ticket queue so priorities do not change.
- Tracing an issue into code and opening a pull request needs repository access. The evidence behind it does not.
Why does an agency-owned repo get in the way of AI agents?
Noibu's guide to shipping Shopify theme changes with GitHub says the person making the connection needs to be able to create branches, open pull requests, and merge in the theme's repository. It also says the only route from a feature branch to the live site is a pull request merged into the main branch. If an agency holds those permissions, the agency is the party that decides what merges, which is a sound control and not a blocker in itself.
The friction is practical. Agencies usually restrict repository access to one person or one team after past problems with several parties editing the same code. They also work from a ticket queue, and a stream of outside pull requests can bypass the order they agreed with you.
What are your options when you do not own the repo?
Three operating models cover most agency setups. Choose the one your agency is comfortable with, and confirm the access details with them and with Noibu support before you begin.
| Model | Who holds repo access | Who reviews and merges | Best when |
|---|---|---|---|
| 1. Agency connects and merges | The agency | The agency | The agency wants full control and you bring the findings. |
| 2. A named engineer is delegated | One engineer or team the agency approves | That engineer, with the agency informed | You want a faster loop and one accountable owner. |
| 3. Findings-only handoff | No outside access | The agency builds from your ticket | The agency will not grant outside access at all. |
How do you keep the agency's priority queue intact?
Treat every approved change as a ticket in the tool your agency already uses, whether that is Jira, Trello, or another system. Create one ticket per change, attach the evidence, and let the agency slot it into the order you have agreed. This keeps the queue as the single source of priority and avoids a second channel of work competing with it. Useful evidence to attach includes the affected sessions, the error details, and the revenue at risk, which is the same material you would use to prioritize website issues by revenue.
How do you keep approval with the people who own the code?
Keep the approval step exactly where the code lives. Noibu's agents stage changes for review and nothing ships without approval. For code changes, the Bug Resolution agent opens a pull request for review, and the repository owner reviews and merges it. To roll back, revert the pull request in GitHub, which Noibu's Shopify theme guide names as the preferred option, and keep a duplicate of the live theme as a backup. For the wider question of letting agents touch a live store, see whether to let an AI agent push code changes and how human-in-the-loop approval works.
What can AI agents do without repository access?
The Bug Resolution agent finds the top issues that block checkout, traces each to its root cause in your codebase, and opens a pull request for review. Tracing into code and opening that pull request needs repository access, so model 3 above stops short of it. The findings behind an issue, including sessions and error details in the Noibu console, can still be shared with your agency as the basis for a ticket. Read more in how the Bug Resolution agent works. Agents work in the Noibu console and in the AI tools your team already uses.
What should you ask your agency before you start?
Bring three questions to the agency. First, who may create branches, open pull requests, and merge in the repository. Second, which ticket system and priority order the agency works from. Third, what review and rollback steps the agency requires before anything reaches the live store. Their answers tell you which of the three models fits. To see what is worth sending them first, run Noibu's free website audit.
Frequently asked questions
Can I use AI agents if my dev agency owns the GitHub repo?
Yes, if the agency stays in control of the repository. The agency can hold access, review each pull request, and merge, while the agent supplies the findings and the proposed change. Which access arrangement works is the agency's decision, so agree it before you begin.
Do I need admin access to the repository?
Noibu's guide to shipping Shopify theme changes says the person connecting needs to be able to create branches, open pull requests, and merge in the theme's repository. It does not describe admin access as a requirement. Confirm the exact permissions with Noibu support and your agency.
Can an AI agent push changes straight to my live store?
No. Changes reach the live store through a pull request that is merged into the main branch, and nothing ships without approval. The repository owner reviews the pull request before it merges.
How do I keep my agency's priority queue?
Create one ticket in the agency's existing system for each approved change, attach the evidence, and let the agency order the work. This keeps one queue as the source of priority instead of adding a second channel.
What if the agency will not allow any outside access?
Use the findings-only handoff. Share the sessions, error details, and revenue impact from the Noibu console as a ticket, and let the agency build the change in its own workflow.
Noibu is AI for ecommerce operations, the always-on optimization system that keeps online stores running at their best. A team of supervised AI agents finds what's broken, slow, or leaking revenue, does the work, and hands results back for human approval. Trusted by retailers like Free People, Guess, Liquid I.V., and Ariat.
