Hi All,
Me and my team were coming up with an interesting use case to answer our customers queries using the codebase. Has anyone tried exposing cloud agents to end users directly ?
Our main concerns were
Currently spinning up the cloud agents for the first time takes a bit of time. It has to index the github codebase that we pass everytime when we create a new agent.
How do we stop malicious users from misutilizing our code.
Looking forward to hear from you all !
Hey, interesting use case. A couple of points:
Indexing:
Cloud Agents are designed for async work on a repo. Re-indexing on every start is by design, since the agent runs in an isolated environment. For large repos (like the one you mentioned before with 7,000+ files), this can take a few minutes.
Using snapshots can speed up startup a bit, since they save the environment state after install commands.
Security:
Cloud Agents aren’t meant to be exposed directly to end users. They were built as a developer tool, not a customer-facing service. There is an API to start agents programmatically, but it requires an API key with repo access.
If you want to build a product that answers questions about a codebase, you probably need your own RAG pipeline with access control and strict context limits, rather than Cloud Agents.
What exact use case are you looking at, support for an internal team or a public service for external users?
Using custom RAG with open source alternatives like cocoindex to parse codebase was giving us not so great results. Cursor seems to be doing this job very well !
Our use case is to respond back to a layman about the logic implemented in our codebase
Also is there any way to clear the agent context ? So that I dont have to spin up a new agent for a new question. I can just clear the previous question context and reuse the same agent for the next question