Computer use (OpenAI Docs)
https://platform.openai.com/docs/guides/tools-computer-use
モデルページ https://platform.openai.com/docs/models/computer-use-preview
#Computer-Using_Agent のAPI提供
combines the vision capabilities of GPT-4o with advanced reasoning to simulate controlling computer interfaces and performing tasks.
サンプルアプリ https://github.com/openai/openai-cua-sample-app
Before integrating the tool, prepare an environment that can capture screenshots and execute the recommended actions.
playwright
Dockerイメージ (TODO)
https://cdn.openai.com/API/docs/images/cua_diagram.png
modelもtoolもcomputer useを指定するっぽい
Acknowledge safety checks
プロンプトインジェクションやモデルの誤りへの対策
a safety check is raised when the model returns the next computer_call, with the pending_safety_checks parameter.
メッセージ
We've detected instructions that may cause your application to perform malicious or unauthorized actions. Please acknowledge this warning if you'd like to proceed.
You need to pass the safety checks back as acknowledged_safety_checks in the next request in order to proceed.
同じメッセージを含めてる?