# Gemini 2.5 Computer Use: follow the demo, local setup and API resources Choose one bounded browser task and inspect its actions before adapting the preview to a larger workflow. [HTML](https://lukejbyrne.com/library/guides/-xmcblrtwh8/) Content language: en Author: Luke Byrne Video published: 2025-10-11 Guide published: 2026-11-02 [Watch the video](https://www.youtube.com/watch?v=-XmCBLRtWH8) Video language: English ## Begin with the public demonstration The October 2025 video links Google’s computer-use announcement and three practical routes: a Browserbase demo, a local preview repository and API documentation. The announcement is https://blog.google/technology/google-deepmind/gemini-computer-use-model/ and the demo is https://gemini.browserbase.com/. The Browserbase section starts at 1:05 and the 2048 example at 2:13. Use a small visible task to understand what the agent attempts and whether the resulting page state matches the goal. The description’s free-access label is historical context, not a verified entitlement for every route today. 1. Open the demonstrated route and choose a small task. 2. Observe the browser actions and the final state. 3. Separate attempted actions from a result that actually satisfies the request. ## Follow the local route with its documented setup Deployment choices appear at 3:38, local setup at 4:25 and API-key configuration at 5:26. The retained repository is https://github.com/google/computer-use-preview and the key destination is https://aistudio.google.com/. Use the repository’s actual requirements and configuration rather than assuming that a local browser means an offline model. Local tool execution starts at 7:44 and the local demonstration at 9:47. Keep the selected model, browser environment and task visible while checking the result. The supplied description contains resource links, not a complete installation command sequence or a reproduced code example. 1. Follow the repository and recorded setup for the local preview. 2. Configure the required API access without exposing the key in shared output. 3. Run a small task and inspect the actual local browser result. ## Use the API and pricing references for the intended workload API details begin at 11:55, with the original reference https://ai.google.dev/gemini-api/docs/computer-use. The retained pricing link is https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-computer-use-preview-10-2025. Keep the fragment intact so the source’s model-specific price context remains identifiable. Review the relevant availability, usage and cost information before scaling a task. The final section at 13:11 covers limitations. Judge the preview through its observed actions and failures, not only the fact that the browser moved or that an API request returned. Running local tools, using a hosted model and paying for API usage are different parts of this workflow. ## The takeaway Choose the route deliberately, inspect a small browser task and keep local execution, model access and API cost distinct. ## Workflow resources - [ailuke.short.gy](https://ailuke.short.gy/-XmCBLRtWH81) - [ailuke.short.gy](https://ailuke.short.gy/-XmCBLRtWH82) - [blog.google](https://blog.google/technology/google-deepmind/gemini-computer-use-model/) - [gemini.browserbase.com](https://gemini.browserbase.com/) - [github.com](https://github.com/google/computer-use-preview) - [ai.google.dev](https://ai.google.dev/gemini-api/docs/computer-use) - [aistudio.google.com](https://aistudio.google.com/) - [ai.google.dev](https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-computer-use-preview-10-2025) - [datacamp.pxf.io](https://datacamp.pxf.io/yqr3jb) - [newsletter.lukejbyrne.com](https://newsletter.lukejbyrne.com) - [linkedin.com](https://linkedin.com/in/lukejbyrne) - [datacamp.pxf.io](https://datacamp.pxf.io/gO6WRB) - [datacamp.pxf.io](https://datacamp.pxf.io/XmnX4a) - [datacamp.pxf.io](https://datacamp.pxf.io/Bny25L) - [datacamp.pxf.io](https://datacamp.pxf.io/aOAWNZ) - [datacamp.pxf.io](https://datacamp.pxf.io/AP3Eg1) - [datacamp.pxf.io](https://datacamp.pxf.io/GKVZbk) - [datacamp.pxf.io](https://datacamp.pxf.io/Z6KyV1) ## Chapters - [0:00 Introduction](https://www.youtube.com/watch?v=-XmCBLRtWH8&t=0s) - [1:05 Browserbase Demo](https://www.youtube.com/watch?v=-XmCBLRtWH8&t=65s) - [2:13 Playing 2048 with AI](https://www.youtube.com/watch?v=-XmCBLRtWH8&t=133s) - [3:38 Deployment options](https://www.youtube.com/watch?v=-XmCBLRtWH8&t=218s) - [4:25 Local setup guide](https://www.youtube.com/watch?v=-XmCBLRtWH8&t=265s) - [5:26 API key configuration](https://www.youtube.com/watch?v=-XmCBLRtWH8&t=326s) - [6:16 Pricing and cost](https://www.youtube.com/watch?v=-XmCBLRtWH8&t=376s) - [7:44 Local tool execution](https://www.youtube.com/watch?v=-XmCBLRtWH8&t=464s) - [9:47 Live local demo](https://www.youtube.com/watch?v=-XmCBLRtWH8&t=587s) - [11:55 API technical details](https://www.youtube.com/watch?v=-XmCBLRtWH8&t=715s) - [13:11 Limitations and conclusion](https://www.youtube.com/watch?v=-XmCBLRtWH8&t=791s) ## Sources and limitations Written companion prepared 2 November 2026 from this exact published video description and its chapter sequence. The full recording was not independently watched or transcribed for this article, and the products and reported results were not retested. Checklists are editorial adaptations of the documented workflow. Model availability, interfaces, prices and offers may have changed since publication; use the retained original resources for current requirements. [Original video](https://www.youtube.com/watch?v=-XmCBLRtWH8)