OpenAI has introduced Computer History, a new feature for the ChatGPT desktop app on macOS that can turn a user’s activity across apps and websites into a searchable timeline. The feature records interaction events such as clicks, typing, keyboard shortcuts and app switches, then converts that information into summaries and local memory files that ChatGPT and Codex can use as context later.
The feature is designed to help users pick up where they left off, find recently used documents and conversations, summarize their work and identify repetitive workflows that could potentially be converted into reusable skills or automations. However, Computer History also introduces significant privacy considerations because activity data is temporarily processed on OpenAI’s servers and the resulting local memory files are stored as unencrypted Markdown files.
What Is OpenAI Computer History?
Computer History is an optional feature that gives ChatGPT a broader view of what a user has been doing on their Mac.
Instead of limiting ChatGPT’s context to conversations that happen inside the chatbot, the feature can use activity from selected applications and websites to reconstruct recent workflows.
OpenAI says the feature captures interaction events through macOS’s accessibility system.
These events can include:
- Clicks
- Typing
- Keyboard shortcuts
- App switches
- Activity from selected websites
- Activity from selected applications
The information is periodically converted into text summaries and stored as local memory files.
How Computer History Works
User works on Mac
↓
Clicks + typing + shortcuts + app switches
↓
Computer History records interaction events
↓
Events are processed into summaries
↓
Memory files are created
↓
Searchable timeline is generated
↓
ChatGPT and Codex can use the context
This allows the AI assistant to understand what the user was doing outside the ChatGPT application itself.
Computer History Replaces Chronicle
Computer History replaces an earlier research preview called Chronicle.
Chronicle used screenshots to capture a user’s computer activity.
Computer History takes a different approach by recording interaction events rather than continuously capturing the screen.
| Feature | Chronicle | Computer History |
|---|---|---|
| Approach | Screenshot-based | Interaction-event based |
| Clicks | Captured indirectly | Captured |
| Typing | Captured through screen | Captured |
| App switches | Not the primary mechanism | Captured |
| Screenshots | Yes | No |
| Screen recordings | Not the focus | No |
| Microphone | No | No |
| System audio | No | No |
| Searchable timeline | Limited preview | Yes |
| Memory summaries | Yes | Yes |
OpenAI says Computer History does not capture screenshots, screen recordings, microphone input or system audio. Private-mode browsing activity is also excluded. :contentReference[oaicite:2]{index=2}
The Feature Creates a Searchable Timeline
One of the main advantages of Computer History is the timeline it creates.
Users can browse activity organized by date and time, with entries describing what they were working on and which applications contributed information.
This means users do not necessarily need to remember the exact name of a document or application to recover an earlier task.
Example Workflow
User works on a proposal
↓
Switches between browser and document editor
↓
Researches information
↓
Edits document
↓
Takes a break
↓
Returns later
↓
Asks ChatGPT what they were working on
↓
Computer History searches recent activity
↓
ChatGPT reconstructs the workflow
This turns ChatGPT into a more persistent work assistant rather than a tool that only remembers individual conversations.
ChatGPT Can Help Users Pick Up Where They Left Off
OpenAI says Computer History can help users resume unfinished work without manually reconstructing what they were doing.
For example, someone could ask what they were working on before a break or where they saw a particular document.
The system can use the activity timeline to identify the relevant source and, where appropriate, access that source directly.
Finding Previous Work
User asks:
“What was I working on before lunch?”
↓
Computer History searches activity
↓
Relevant applications identified
↓
Recent documents and websites identified
↓
ChatGPT summarizes the previous workflow
This could be especially useful for people who frequently switch between multiple applications.
It Can Identify Repetitive Workflows
Another major feature is the ability to recognize repeated patterns.
If Computer History notices that a user repeatedly performs the same sequence of actions, the timeline can suggest turning that workflow into a reusable skill or automation.
Repetitive Workflow
Open spreadsheet
↓
Copy data
↓
Open browser
↓
Search information
↓
Update document
↓
Repeat every day
↓
Computer History recognizes pattern
↓
Suggests skill or automation
↓
User reviews suggestion
↓
Codex can build reusable workflow
This moves Computer History beyond simple activity logging and toward workflow automation.
Codex Can Turn Workflows Into Skills
OpenAI says users can ask Codex to create reusable templates from recorded workflows.
This could allow a repetitive process to become an automated workflow instead of something the user has to perform manually every time.
For example, a recurring reporting process could potentially be transformed into a reusable skill.
From Activity to Automation
Daily workflow
↓
Computer History records pattern
↓
Pattern recognized
↓
Skill suggestion
↓
User approves
↓
Codex creates reusable workflow
↓
Future task becomes easier
This is part of OpenAI’s broader push toward AI agents that can perform multi-step computer tasks.
Computer History Is Off by Default
Privacy is a major consideration, so Computer History is not automatically enabled.
Users must explicitly turn it on.
OpenAI’s documentation says the feature requires users to opt in, while Business and Enterprise customers also need their workspace administrator to enable access before individual users can activate it
Activation Model
Computer History
↓
OFF by default
↓
Business/Enterprise
↓
Administrator enables access
↓
Individual user opts in
↓
User chooses apps and websites
↓
Collection begins
This gives both organisations and individual users control over whether the feature is used.
Users Can Choose Which Apps and Websites Are Included
Computer History does not require users to provide access to every application.
OpenAI provides controls that allow users to specify which apps and websites can contribute information.
Users can either include specific sources or exclude particular applications and websites.
Permission Controls
All supported apps
↓
OR
Only selected apps
↓
All supported websites
↓
OR
Only selected websites
↓
Computer History collects from permitted sources
This allows users to exclude applications containing particularly sensitive information
Users Can Pause Computer History
The feature can also be paused without completely disabling it.
Users can control collection through Computer History settings or the macOS menu bar.
Pause Workflow
Computer History active
↓
User selects Pause
↓
New activity collection stops
↓
User resumes work
↓
User selects Resume
↓
Collection starts again
This can be useful when moving into a private conversation or opening sensitive material.
Private Browsing Is Not Included
OpenAI says activity from private-mode web browsing is never included in Computer History.
This provides an additional boundary around browsing activity.
However, users still need to configure application and website permissions carefully because activity from permitted sources can contribute to the timeline.
OpenAI Warns About Sensitive Information
OpenAI recommends that users pause Computer History or exclude applications containing sensitive information.
The company’s documentation specifically highlights health, financial and personal information as categories that users should consider protecting.
It also warns users to pause Computer History when communicating with other people unless those people have given appropriate consent.
Sensitive Data Risk
Sensitive application
↓
Computer History active
↓
Interaction events captured
↓
Information summarized
↓
Potentially included in memory
↓
Privacy concern
The feature therefore requires more active privacy management than ordinary ChatGPT conversation history.
The Data Is Temporarily Processed by OpenAI
Although Computer History stores activity events locally first, OpenAI says those temporary event files are processed on its servers to generate the memory summaries.
The company says it does not retain those temporary event files after processing unless required by law.
OpenAI also says the event files are not used for AI training. :contentReference[oaicite:7]{index=7}
Data Flow
Mac
↓
Temporary activity files
↓
OpenAI servers
↓
Memory generation
↓
Memory stored locally
↓
ChatGPT uses memory later
This architecture is important because the feature is not purely local even though the generated memory files remain on the user’s Mac.
Temporary Event Files Are Deleted After 48 Hours
According to OpenAI, temporary Computer History event files stored on the Mac are retained for up to 48 hours.
After processing, those event files are deleted.
However, the generated memory files remain on the Mac until the user deletes or clears them. :contentReference[oaicite:8]{index=8}
| Data Type | Storage | Retention |
|---|---|---|
| Raw interaction events | Mac | Up to 48 hours |
| Processing | OpenAI servers | Temporary |
| Generated memory files | Mac | Until deleted |
| ChatGPT conversations using memory | ChatGPT systems | Subject to account/data controls |
This distinction is important because deleting temporary event data does not automatically mean all generated memory information disappears.
Local Memory Files Are Not Encrypted
One of the most significant privacy concerns is how the generated memory files are stored.
OpenAI says Computer History creates plain-text Markdown memory files on the user’s Mac.
The files are not encrypted by Computer History.
That means another program operating under the same macOS user account could potentially access them. :contentReference[oaicite:9]{index=9}
Local Storage
Computer activity
↓
Memory generated
↓
Markdown file
↓
Stored locally
↓
Not encrypted
↓
Potential access by other programs under same user account
This is an important trade-off for users handling sensitive information.
Prompt Injection Is Another Risk
OpenAI’s documentation warns that Computer History increases the risk of prompt injection.
Because the system can process information from websites and applications, malicious instructions embedded in external content could potentially influence an AI system that later uses the recorded activity as context.
Potential Prompt Injection
User visits webpage
↓
Webpage contains malicious instructions
↓
Computer History records relevant interaction/context
↓
AI processes activity
↓
Malicious content may influence later AI behavior
↓
Potential security risk
OpenAI specifically flags this as a consideration for users. :contentReference[oaicite:10]{index=10}
Why Prompt Injection Matters More With Computer History
Traditional chatbots generally process information that users intentionally provide.
Computer History expands the amount of environmental information that can enter the AI’s context.
That creates a much larger attack surface.
Traditional AI
User
↓
Prompt
↓
AI
↓
Response
Computer History
User
+
Websites
+
Apps
+
Documents
+
Activity
↓
Computer History
↓
Memory
↓
AI context
↓
Response or action
The more sources an AI can observe, the more carefully those sources need to be treated as untrusted input.
Computer History Could Make ChatGPT More Useful
Despite the privacy risks, the feature has a clear productivity benefit.
People often forget where they saw information, which document they edited or what task they were working on.
Computer History can potentially reconstruct that context automatically.
Productivity Benefits
Find documents
+
Recover previous work
+
Summarize activity
+
Understand workflows
+
Create automations
↓
Less manual searching
↓
More persistent AI assistance
This could make ChatGPT significantly more useful as a daily work companion.
It Moves ChatGPT Toward an Agentic Assistant
The feature is also part of a broader industry shift from conversational AI toward agentic AI.
Traditional chatbot
↓
User asks question
↓
AI answers
Agentic assistant
↓
Observes context
↓
Understands objective
↓
Finds information
↓
Uses tools
↓
Completes workflow
Computer History provides the contextual layer needed for an agent to understand what the user has been doing before being asked to take action.
Computer History Could Improve Personalisation
The timeline can provide ChatGPT with more context about an individual’s ongoing work.
Instead of asking users to repeatedly explain the background to a project, the assistant could potentially retrieve relevant recent activity.
Personalised AI
User activity
↓
History
↓
Memory
↓
Context
↓
Personalised response
This could make AI assistance feel more continuous across different applications and sessions.
The Feature Could Help With Daily Work Summaries
One potential use case is automatically summarizing what someone worked on during a day.
Instead of manually tracking tasks, users could ask ChatGPT to reconstruct their activity.
Daily Summary
Morning
↓
Research
↓
Documents
↓
Meetings
↓
↓
Coding
↓
Afternoon
↓
Computer History
↓
Workday summary
↓
Completed tasks
+
Open tasks
+
Next steps
This could be useful for standups, project reporting and personal productivity.
It Could Help Find Files and Conversations
Another potential use is locating recently used files or conversations.
A user may remember what they were doing but not the exact filename.
Computer History could use contextual clues to identify the relevant source.
Search by Memory
User remembers:
“I was editing a proposal about AI infrastructure.”
↓
Computer History searches timeline
↓
Identifies relevant applications
↓
Finds document activity
↓
ChatGPT helps locate source
This could make traditional file and browser search less important for some workflows.
Computer History Is Currently a Mac Feature
The new functionality is being rolled out through the ChatGPT desktop application for macOS.
It is not currently a general feature across every ChatGPT platform.
The availability also varies by plan and region.
Current Availability
ChatGPT for Mac
↓
Computer History
↓
Supported plans
↓
Pro
+
Business
+
Enterprise
↓
Regional restrictions apply
OpenAI’s documentation says the feature is not yet available in the European Economic Area, Switzerland or the United Kingdom
Business and Enterprise Customers Have Additional Controls
For organisations, Computer History has an additional administrative layer.
Workspace administrators must explicitly enable access before employees can choose to activate the feature.
Even after an administrator enables it, individual users still need to opt in.
Enterprise Permission Model
Administrator
↓
Enables Computer History
↓
Employee
↓
Chooses to opt in
↓
Selects apps and websites
↓
Computer History activates
This creates a two-level permission structure.
The Feature Is Different From Normal Chat History
Computer History should not be confused with the existing ChatGPT conversation-history system.
Normal chat history primarily records conversations with ChatGPT.
Computer History expands the scope to activity that happens outside ChatGPT across selected apps and websites.
Chat History vs Computer History
Chat history
↓
ChatGPT conversations
Computer History
↓
Apps
+
Websites
+
Clicks
+
Typing
+
App switches
↓
Searchable activity timeline
This broader scope is what makes the new feature particularly powerful—and more privacy-sensitive.
Apple and OpenAI Take Different Privacy Approaches
Computer History also highlights a broader difference between Apple and OpenAI’s approaches to AI personalisation.
Apple has emphasised on-device processing and privacy-preserving cloud infrastructure for many of its AI features.
OpenAI’s Computer History instead temporarily processes activity data on its servers to create the memory summaries.
Two Approaches
Apple approach
↓
More emphasis on local processing
VS
OpenAI Computer History
↓
Local event collection
↓
Temporary server processing
↓
Local memory storage
The difference could become increasingly important as AI assistants gain access to more personal information.
The Feature Could Change How People Use AI
If Computer History works reliably, users may no longer need to think about what information to give an AI assistant.
The assistant could potentially infer context from recent work.
Current AI Workflow
User
↓
Explains context
↓
Provides files
↓
Provides links
↓
Explains previous work
↓
AI responds
Computer History Workflow
User
↓
Asks task
↓
Computer History supplies relevant context
↓
ChatGPT identifies previous work
↓
AI responds
The second approach could make AI assistance much more seamless.
But More Context Does Not Always Mean Better Privacy
The same feature that makes ChatGPT more useful also gives it access to more information about the user’s daily activities.
A timeline of clicks, typing and application switches can reveal work habits, interests, projects and interactions.
Even if raw events are not permanently stored on OpenAI’s servers, the resulting memories can contain meaningful information.
Privacy Trade-Off
More activity context
↓
More useful AI
BUT
↓
More personal information captured
↓
More privacy responsibility
The value proposition therefore depends heavily on whether users are comfortable making that trade-off.
Users Can Delete Individual History Items
OpenAI provides controls for reviewing and deleting Computer History items.
Users can delete individual timeline entries or clear larger periods of history.
The documentation says users can clear the last 10 minutes, last hour, last day or all history.
History Controls
View timeline
↓
Select item
↓
Delete item
OR
↓
Clear recent period
OR
↓
Clear all history
↓
Data and associated memories removed
OpenAI says clearing history deletes the relevant interaction events and memories created from them, and that the action cannot be undone. :contentReference[oaicite:12]{index=12}
The Feature Could Become More Important for Codex
Computer History also provides context for Codex.
A coding agent could potentially understand what a developer was working on before being asked to continue.
Coding Workflow
Developer works on project
↓
Opens editor
↓
Reads documentation
↓
Runs tests
↓
Switches applications
↓
Computer History records activity
↓
Developer asks Codex to continue
↓
Codex receives contextual history
↓
Coding workflow resumes
This could reduce the amount of manual explanation developers need to provide.
Repeated Work Could Become Automated
The combination of activity history and Codex creates a path toward automated workflow creation.
Instead of explicitly programming an automation, users could demonstrate a process naturally through their everyday computer activity.
Future Workflow
User performs task repeatedly
↓
Computer History detects pattern
↓
AI suggests automation
↓
User approves
↓
Codex creates workflow
↓
Automation handles future repetitions
This is a major step toward AI systems that learn workflows from observation.
The Model Could Reshape Productivity Software
If Computer History becomes widely adopted, traditional productivity tools could increasingly compete with AI assistants that understand activity across applications.
Rather than users opening individual applications to search for information, an AI layer could become the interface connecting them.
Traditional Productivity
+
Browser
+
Documents
+
Calendar
+
Files
↓
Separate applications
AI-Assisted Productivity
Apps
↓
Computer History
↓
AI context layer
↓
ChatGPT / Codex
↓
Unified assistance
This could become an important direction for future operating-system and productivity software.
What It Means for Users
For users, Computer History offers a powerful productivity feature but requires careful privacy management.
Potential benefits include:
- Finding recent work
- Recovering forgotten context
- Summarizing activity
- Finding documents
- Understanding workflows
- Creating automations
- Giving ChatGPT better context
- Helping Codex resume coding tasks
Potential risks include:
- Sensitive information entering memory
- Local memory files being unencrypted
- Prompt-injection attacks
- Privacy concerns involving other people
- Over-collection of unnecessary activity
- Greater dependence on AI for personal context
What It Means for Businesses
Businesses could potentially use Computer History to improve employee productivity and automate repetitive workflows.
However, companies will need clear policies governing what applications employees can include and what types of information should be excluded.
Privacy, employment law, security and data-governance requirements will also matter.
Enterprise AI Governance
Computer History
↓
App permissions
+
Website permissions
+
Employee consent
+
Security controls
↓
Controlled deployment
↓
Productivity gains
The administrative controls are therefore likely to become increasingly important as adoption grows.
What It Means for OpenAI
For OpenAI, Computer History strengthens ChatGPT’s position as a persistent AI assistant.
The company is moving from a model in which users provide context manually toward one in which ChatGPT can obtain relevant context from a user’s digital activity.
OpenAI’s Evolution
Chatbot
↓
Memory
↓
Computer use
↓
Computer History
↓
Workflow understanding
↓
Skills
↓
Automations
↓
Agentic assistant
This represents a broader shift in the role OpenAI wants ChatGPT to play.
What Investors Should Watch
Investors and technology companies should watch several developments as Computer History expands.
Key areas include:
- User adoption
- Expansion to Windows
- Regional rollout
- Enterprise adoption
- Privacy incidents
- Security research
- Prompt-injection safeguards
- Workflow automation adoption
- Codex integration
- Growth of AI-generated skills
OpenAI’s Computer Strategy
Computer History
↓
More context
↓
Better AI assistance
↓
More automation
↓
More user dependence
↓
Stronger ChatGPT ecosystem
The success of the feature will depend on whether users believe the productivity benefits justify the privacy trade-offs.
Key Facts at a Glance
| Feature | Details |
|---|---|
| Product | ChatGPT Computer History |
| Platform | macOS desktop app |
| Status | Rolling out |
| Default | Off |
| Activity captured | Clicks, typing, shortcuts and app switches |
| Screenshots | No |
| Screen recordings | No |
| Microphone | No |
| System audio | No |
| Private browsing | Excluded |
| Temporary event retention | Up to 48 hours |
| Memory format | Plain-text Markdown |
| Local memory encryption | Not encrypted |
| AI systems using context | ChatGPT and Codex |
| Admin approval | Required for Business/Enterprise workspace access |
| Current regional limitation | EEA, UK and Switzerland unavailable |
Infographic: How OpenAI Computer History Works
USER WORKS ON MAC
↓
CLICKS
+
TYPES
+
USES SHORTCUTS
+
SWITCHES APPS
↓
COMPUTER HISTORY
↓
ACTIVITY EVENTS
↓
TEMPORARY PROCESSING
↓
AI-GENERATED MEMORY
↓
LOCAL MARKDOWN FILE
↓
SEARCHABLE TIMELINE
↓
CHATGPT + CODEX
↓
BETTER CONTEXT
↓
WORKFLOW SUGGESTIONS
↓
SKILLS + AUTOMATIONS
The Bigger Picture
OpenAI’s Computer History represents a significant step in the evolution of AI assistants. Instead of relying only on conversations and manually provided files, ChatGPT can now use a structured record of selected computer activity to understand what users were doing across applications and websites. That could make the assistant substantially more useful for finding documents, resuming unfinished work, preparing summaries and eventually automating repetitive workflows.
However, the feature also demonstrates the difficult privacy trade-off that comes with more contextual AI. Computer History is opt-in and provides controls for selecting apps, websites, pausing collection and deleting history, while OpenAI says raw activity files are temporary and not used for training. At the same time, generated memory files remain locally as unencrypted Markdown, and OpenAI warns about prompt-injection risks and the need to exclude sensitive applications. The feature therefore represents both a major productivity opportunity and a new category of AI privacy risk.
Looking Ahead
Computer History could become one of the most important components of OpenAI’s strategy to turn ChatGPT into a persistent computer assistant. By understanding recent activity rather than simply waiting for users to provide context, ChatGPT and Codex can potentially help people recover unfinished work, locate documents, summarize their day and transform repeated workflows into reusable skills. If OpenAI expands the feature to more platforms and improves its ability to safely interpret activity, the assistant could become a layer connecting many of the applications people use every day.
The biggest challenge will be trust. Users will need confidence that the benefits of giving an AI broader access to their computer activity outweigh the privacy and security risks. OpenAI’s current controls—including opt-in activation, app and website permissions, pause controls and history deletion—provide important safeguards, but the unencrypted local memory files and prompt-injection risks show that the feature is not without vulnerabilities. The long-term success of Computer History will therefore depend not only on how accurately it remembers users’ workflows, but also on how effectively OpenAI protects that increasingly valuable personal context.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
