- Your first Project and file uploads
- The major AI families to try the workflow
- A small daily message allowance
GitHub projects with AI
Bring selected GitHub files into a Project, and let several AIs work from the real code.
AI coding help is only as good as what the model can see. Pasting one file at a time loses the context that makes a review useful: the data model, the conventions, the module the function calls. Where GitHub import is enabled, a MultipleChat Project lets you pick the files from a repository that matter and use them as context for ChatGPT, Claude, Gemini and the others. This page is what that is, what it is not, and how to get the most from it.
Availability depends on your plan; the documentation has the current details.
Honest scope
What GitHub import does, and what it does not
| It does | It does not |
|---|---|
| Let you select files from a repository you have connected and include them in a Project | Analyse every commit, branch and file of the whole repository automatically |
| Make those files available as context for reviews, explanations, refactors and debugging with any model | Push commits, open pull requests or run your code |
| Let you re-sync files after the code changes | Replace your editor, terminal or CI |
| Work alongside direct uploads and Drive import where enabled | Access repositories or files beyond what you select |
Selected files, deliberately
Choosing files is a feature, not a limit. A Project with the README, the schema, the module you are working in and its tests gives better answers than a dump of the whole repo, because the models see what matters and nothing that distracts.
What to import
A good context set for most tasks
| Task | Import | Ask |
|---|---|---|
| Understanding a codebase | README, entry points, the data model, one representative module | "Give me the map: data flow, the three files that matter, what will surprise a new developer." |
| Adding a feature | The module it goes in, the modules it calls, the tests, the schema | "Add [feature] to [file]. Only change what I asked. Then review it as a second engineer." |
| Reviewing a change | The changed files and their tests | "Review for bugs, security and edge cases, ordered by severity, with fixes." |
| Debugging | The file in the stack trace and what it calls | "Here is the trace. Cause and smallest fix; list other possible causes." |
| Refactoring | The module and every file that imports it | "Write tests that pin current behaviour, then refactor." |
| Documentation | The public modules and existing docs | "Write the README and API docs from the code, not from memory." |
The loop
How GitHub-backed Projects fit the coding workflow
Import the files. Add Project instructions with your conventions ("Postgres, integer cents for money, no secrets in code"). Ask one model for the change, a different model for the review, and a third when the first two disagree. Run the tests. Commit in your own environment. Re-sync the changed files so the next question sees the current code. The full loop, and the seven mistakes AI code makes most, are on coding with AI.
Free is the start. The latest models are the difference.
Code review is where the latest models pull furthest ahead.
Finding the retry bug, the missing account filter or the leaked secret in real code is a reasoning task over a lot of context. The newest models do it reliably; older ones approve plausible code. The paid plans also carry the upload capacity a real repository selection needs. For code going to production, that is the plan to review it on.
Newer models hold your whole Project in view instead of the last few messages, reason through trade-offs instead of picking the first plausible answer, and are wrong less often and with less confidence. For the questions on this page, that is the difference between advice that sounds right and advice you can act on.
Pro from $20 a month, cancel any time. The free plan stays free.
- The latest ChatGPT, Claude, Gemini, Grok and Perplexity models, all on the same Project
- Far more messages a day, so a working session does not stop halfway
- Modes where several models draft, challenge and verify each other's answers
- Larger uploads and more files per Project
- Independent verification passes for library versions and API facts
Start here
First messages for a GitHub-backed Project
Import the README, the schema and the module you are working in. Add your conventions to the instructions. Then start here.
MapFrom the imported files, give me the map: entry points, data flow, the three files that matter most, the thing that will surprise a new developer.
ConventionsRead the code and list the conventions it follows. Which ones are inconsistent? Write them as Project instructions.
ReviewReview [file] as a senior engineer who did not write it: bugs, security, edge cases, ordered by severity, each with the fix.
TestsWrite the tests [module] needs, including retries, empty input, other time zones and very large input.
DebugHere is the stack trace. Using the imported files, what is the cause and the smallest fix? List other possible causes.
DocsWrite the README for this repository from the imported code, with setup, architecture and the three things a contributor must know.
Honest answers
What people ask
Does MultipleChat read my whole repository?
No. You connect GitHub and select the files to include in a Project. Only those are used as context.
Is my code used to train AI models?
No. Imported code stays in your account, is encrypted, and is never used to train any model by MultipleChat or the AI providers. See security.
Can it commit or open pull requests?
No. It writes, reviews and explains; you commit in your own environment. That keeps your repository under your control.
Private repositories?
You connect your own GitHub account and choose from repositories it can access, private included.
Which plans include GitHub import?
It is available where enabled; see the documentation and plans.
Why not just paste code into a chatbot?
Because context is the whole game. A Project with the right files, your conventions and several models reviewing each other gives answers a single pasted file cannot.
Import the README, the schema, and the file you are stuck on.
Ask for the map, then the review. See what a second model finds. Free to start.
Also: coding with AI · build software with AI · build a SaaS with AI · Google Drive projects