First: The Stability Problem
I’ve been seeing the threads on this forum for a while now, and the repeating memory errors and internal failures are very inconvenient; they’re blocking my projects from going to production. That needs to be the top priority before anything else. I say this not to criticize, but because AI Studio has genuine potential and these issues are holding it back.
Now, my proposal: a local workspace model similar to Antigravity.
AI Studio is currently web-only, which works for small projects but breaks down for serious development. My proposal: turning Google AI Studio into a dedicated mobile app (Android/iOS) with a local workspace system.
How it would work:
- Each chat/project gets its own isolated folder in a directory on the device
- The AI can only access files within its assigned directory and the folder for the chat, e.g., Chat A cannot see Chat B’s files
- API keys are configured through a strict, secure settings interface, not injected directly into code
- API keys are then injected from the Env var menu once it’s launched in local development
- Chats are listed and managed inside the app like projects, not browser tabs
Tools That Need to Exist:
- Test File / Code Runner: Catch errors early in development, not at deployment
- Feature Testing: automated UI interaction testing (Antigravity has a desktop version; a mobile equivalent is missing)
- Security Review Tool: Before shipping anything to production, the AI audits the project for exposed keys, unsafe patterns, permission issues
- Compact / Context Management: structured control over what the model holds in memory per project, which would also directly address the memory-related errors people are experiencing
If Google sorts out the database tools, frontend tooling, 3D capabilities, and security review in one coherent package, with per-project system prompts and built-in skills, AI Studio could genuinely be the platform where things go from idea to production without switching tools.
- Note: Execution would require files to be sent to the server, and after the output was determined, it would return to the user device.
Happy to discuss any of this further. I’m building on AI Studio right now and running into a lot of these gaps.