This experimental platform from Google lets you build and run AI experiments with direct ties to their ecosystem. You get a workspace for testing AI concepts with built-in connections to Drive, Docs, Sheets, and Slides. Think of it as a sandbox where you can prototype AI workflows without starting from scratch.
The setup creates a dedicated Opal folder in your Drive automatically. You can spin up experiments, cache context to save resources, and use agent mode to let AI handle multi-step tasks. There's a stream planner for sequencing actions and tools for generating webpages on the fly. The text editor includes remix features and a token counter if you need to track usage.
It lives on multiple Google domains. If you're on a google.com email, you'll likely use the corp.goog version. That domain restriction hints at who this really serves, internal Google employees first, wider Google users second. The system asks for consent before fetching or opening webpages, which adds friction but makes sense for security.
Now the limitations. Pretty extensive. Third-party modules aren't allowed. You can't publish experiments publicly if you're on a google.com domain. Several features are explicitly turned off — no MCP support, no NotebookLM integration, no output templates, no graph editor agent, no consistent UI mode. The list of disabled functionality is surprisingly long for a tool meant to encourage experimentation.
The agent mode can resume runs if something breaks, which is useful. The Drive picker works in lite mode only. You can share experiments with the underlying AI, though the specifics of how that works aren't detailed. There's a survey system baked in, probably for collecting feedback on experiments.
This makes sense for two groups. Google employees testing internal AI ideas will find the pre-built integrations valuable. Developers already deep in the Google workspace who want to prototype AI features without managing infrastructure might benefit too. But if you're outside that ecosystem or need third-party tools, the walls feel pretty high.
The disabled features suggest this is either still early or intentionally limited. It's not trying to be a full-scale AI development environment. More like a controlled setting for specific kinds of experiments. You won't get the flexibility of open platforms, but you also won't spend time wiring up Google API connections.
