Key Points
- 1.OpenAI introduced Computer History, a macOS desktop application feature that tracks user workflows across selected apps and websites to provide contextual assistance and task automation.
- 2.Access to Computer History requires a paid ChatGPT Pro, Business, or Enterprise subscription and is currently unavailable in Europe, Switzerland, and the United Kingdom.
- 3.Users can configure granular privacy settings by including or excluding specific applications and websites from activity tracking.
- 4.The recorded activity log enables ChatGPT to summarize daily actions, draft project plans, build reusable Codex skills from observed workflows, and suggest automation opportunities.
- 5.OpenAI also previewed Ultrafast mode for GPT-5.6 Sol, offering up to 14 times faster processing speeds, and released the ChatGPT desktop app for Linux.
Summary
OpenAI released a feature called Computer History within the ChatGPT desktop application. The tool observes user activity across applications and websites to maintain full context of ongoing work without recording raw screen video or capturing audio. Currently, the feature is limited to the macOS desktop app and requires a Pro, Business, or Enterprise subscription costing at least $100 per month. It is also geographically restricted, remaining unavailable in the United Kingdom, Switzerland, and the European Union.
Inside the application settings under the Integrations category, users can enable Computer History and define granular privacy boundaries. A setup interface provides options to include or exclude specific programs and domains. For the demonstration, applications like Messages and internal web dashboards were excluded, while tools such as Google Chrome, Finder, and TextEdit were permitted. System prompts remind users to pause tracking during video calls, personal messaging, or handling sensitive financial and medical data.
During a practical test, the presenter created a community onboarding redesign outline in Google Docs, converted the notes into a structured one-page project plan using ChatGPT, exported the file as a PDF, organized it in a dedicated Finder folder, and drafted an email in Gmail. When prompted with a simple query about what had been accomplished, ChatGPT returned an accurate breakdown of all recent document edits, planning tasks, and local file operations, referencing the specific items created.
When asked to convert the completed workflow into a reusable routine, ChatGPT built an operational skill named Rough Project Plan, automatically extracting deliverables, workstreams, risk assessments, and launch criteria directly from the recorded work. A follow-up prompt successfully extended the skill to integrate downstream task steps. Furthermore, asking ChatGPT to audit the day for automation opportunities produced a detailed table scoring each discrete task as high, partial, or low potential for automation, accompanied by proposed execution scripts.
The video compares this contextual tracking approach to previous attempts like Microsoft Recall and current agent platforms such as OpenClaw and Hermes Agent. The presenter also highlights the preview of Ultrafast mode for GPT-5.6 Sol, which operates up to 14 times faster via API for technical workloads, as well as the release of the ChatGPT desktop application on Linux.
Overview and Requirements of Computer History
OpenAI launched Computer History within the ChatGPT macOS desktop application. The feature functions by summarizing user activities across permitted apps and websites without capturing audio or recording the screen directly. To use the feature, users need the macOS desktop client and an active Pro, Business, or Enterprise subscription, which costs at least $100 per month for individual users. The tool is restricted geographically, remaining unavailable across Europe, Switzerland, and the United Kingdom.
Privacy Controls and Live Workflow Demonstration
The settings menu allows users to specify which applications to include or exclude, such as limiting capture to Google Chrome, TextEdit, and Finder while excluding personal messaging or sensitive internal dashboards. In a live test, the presenter created a project document in Google Docs, formatted a one-page plan using ChatGPT, saved the file locally as a PDF, and drafted an email in Gmail. Asking ChatGPT what occurred yielded an accurate chronological summary of the entire session.
Skill Generation, Automation Auditing, and Frontier Updates
ChatGPT converted the recorded workflow into a reusable Codex skill and successfully extended the skill when prompted. Asking the model to evaluate daily tasks generated an automation audit table detailing task feasibility ratings and automation plans. The video also covers related industry updates, including comparisons to SpaceX AI's Grok Bot, the rollout of the ChatGPT desktop application to Linux, and the preview of GPT-5.6 Sol Ultrafast mode running at up to 14 times speed.
Worth watching for
AI enthusiasts, workflow automation professionals, and macOS power users interested in contextual AI tools and desktop tracking features.
- chatgpt
- openai
- computer-history
- automation
- ai-agents
- productivity