Hi everyone. I've been exploring Kimi for the past few weeks, mostly for brainstorming and summarizing documents, and I'm really impressed. I work a lot with marketing automation setups in HubSpot, which often involves tweaking code snippets, custom modules, and API integrations.
I wanted to push it further and see if Kimi could help me understand our own internal code repositories. We have a private GitHub repo with our HubSpot CMS custom modules, some scripted workflows, and analytics tracking setups. The idea of having a knowledgeable assistant to ask questions about that specific codebase is incredibly appealing.
I've seen mentions that Kimi can connect to GitHub, but I'm a bit fuzzy on the practical steps and, honestly, the real-world usefulness for this kind of technical marketing code. Has anyone successfully set this up with a private repo?
I'm particularly curious about a few things:
- What the authentication process looks like and how permissions are handled.
- Whether it effectively understands the relationships between files in a project, like a module's JavaScript, its HubSpot module meta JSON, and associated CSS.
- If you've used it for similar marketing tech stacks, what kind of questions have yielded the most helpful answers?
I'm hoping to learn from your experiences before I dive in and potentially configure this for my team. Any pitfalls or "wish I knew" moments would be great to hear.
I've done this with a few client repos for HubSpot module troubleshooting. The OAuth flow via GitHub is straightforward; you grant Kimi access to specific repos, and it handles the rest. Permissions are scoped to what you authorize, similar to any GitHub app.
The real test is whether it understands HubSpot's specific structure. In my experience, it connects files decently. I uploaded a custom module directory, and it could link the module.html to its `meta.json` and explain the field dependencies. It struggled with the context of serverless functions calling third-party APIs unless I explicitly uploaded the function config file.
For marketing tech code, it's useful for quick questions like "Where is the form submission payload being transformed?" or "What's the logic behind this contact property assignment?" It won't replace digging through the codebase yourself, but it saves time on initial orientation. I wouldn't trust it for migration planning without verifying every line it references, though.
Show me the query.