Recommended Free Tools
Before connecting company files, mail, tickets, code, or databases to an AI service, identify the exact product and endpoint, then review five separate controls: model-improvement use, conversation history, memory, deletion, and connected-app access. Turning off training does not necessarily delete conversations, stop safety logs, or remove indexed copies. Retention periods and deletion exceptions vary by account type, subscription, feature, region, and API endpoint.
This guide reflects provider documentation available through October 13, 2026. Interfaces and terms change, so confirm each setting in the live administrator console and contract before enabling a connection.
As an Amazon Associate I earn from qualifying purchases.
Why retention review comes before a connector
An AI assistant can handle the same business content through several different paths: a consumer chat, a managed workspace, a vendor-hosted API, a cloud marketplace deployment, or a third-party app. Each path can create different records. A prompt may be copied into chat history, an uploaded file may be retained separately, a retrieval system may build an index, and security systems may keep logs even after a user deletes the conversation.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchThere is no universal “AI retention switch.” The relevant control can be at the organization, project, workspace, user, endpoint, model, connector, or feature level. A setting that applies to ordinary chat may not apply to an API call, a compliance feed, a coding assistant, or a synced application.
#1 Best Overall
- Certified to FIPS 197 - High-level information security standard approved by the U.S. Government
- Brute-Force Password Attack Protection - Data is automatically erased after 6 failed access attempts. The data and encryption key are securely destroyed and the crypto drive is reset
- Rugged Double-Layer Waterproof* Design - Protects the crypto drive against knocks, drops, break-in and submerging in water. The electronics are shielded by a hardended inner case. The rubberised silicone outer casing provides a final layer of protection
- Auto-lock - The crypto drive will automatically encrypt all data and lock when removed from a PC/Mac or when the screen saver or "computer lock" function is activated on the host PC/Mac
- Secure Entry - Data cannot be accessed without the correct high-strength alphanumeric 8-16 character password. A password hint option is available. The password hint cannot match the password
Use this order of operations:
- Identify the exact account, plan, endpoint, model, region, and enabled features.
- Map what the assistant can read or change and which identity it uses.
- Review training and feedback use separately from storage and deletion.
- Document every retention layer and its exceptions.
- Test deletion and access with non-sensitive data before production use.
- Assign an owner to recheck settings whenever the provider changes a feature or term.
Start with an exact deployment inventory
Write down the deployment before opening an integration. “We use an AI tool” is not specific enough for a retention decision.
| Inventory item | What to record | Why it changes the answer |
|---|---|---|
| Account boundary | Consumer account, business workspace, enterprise tenant, education or healthcare workspace, or API organization | Training defaults, administrator controls, contracts, and deletion schedules can differ. |
| Endpoint | Web chat, mobile app, embedded office feature, API endpoint, command-line tool, or marketplace-hosted model | Zero-retention and logging eligibility is usually endpoint-specific. |
| Plan and agreement | Subscription tier, enterprise addendum, data-processing terms, and any custom retention agreement | A plan may provide controls that are absent from another plan under the same brand. |
| Features | History, memory, file upload, retrieval, projects, feedback, agents, code execution, audit or compliance feeds | Each feature can create its own application state or log. |
| Connected sources | Drive, mail, ticketing, CRM, source control, databases, local folders, and external apps | Source permissions, indexes, subprocessors, and third-party terms enter the data path. |
| Region | Selected storage region and any cross-border processing disclosed by the provider | Regional storage does not automatically mean every processing step or integration stays there. |
Save screenshots or exported policy records with the date, administrator, tenant, and endpoint. A later reviewer must be able to tell whether a control applied to the production path or only to a test account.
Training, history, memory, and deletion are different controls
Model improvement asks whether provider personnel or automated pipelines may use content to improve models. History controls what a user can see or reopen. Memory stores facts or preferences for later personalization. Retention covers backend records such as abuse-monitoring logs, application state, indexes, and audit trails. Deletion describes which record is removed, when, and under what exceptions.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →OpenAI’s Help Center states: “When Improve the model for everyone is off, your new conversations won’t be used to train OpenAI models. They can still appear in your chat history.” Read the full control description at OpenAI’s ChatGPT data-controls documentation. The sentence illustrates why a training opt-out is not a deletion command.
Feedback can have a separate path. OpenAI says a conversation associated with feedback may be used for training, while its enterprise privacy page says business data is not used for training by default unless a customer explicitly opts in, including through feedback mechanisms. Check both the account category and the action a user takes when submitting a rating.
Map every retention layer before connecting data
Ask the provider to identify each class of record, not just a single headline retention period. Record the answer in a data-flow register.
| Data class | Questions to answer | Common exception to check |
|---|---|---|
| Prompts and responses | Are they stored, for how long, and can an administrator export or delete them? | Safety review, abuse monitoring, legal hold, or a feature that needs a transcript. |
| Uploaded files | Are files retained independently of the chat? Are derived chunks or embeddings created? | Indexes, project storage, backups, or a shared workspace copy. |
| Conversation history | What does “history off” change, and can users still access existing chats? | Short-term service storage or feedback processing after history is disabled. |
| Memory and personalization | Where are saved memories listed, and can users or admins clear them separately? | Account-level profile data surviving deletion of an individual chat. |
| Application state | Do agents, threads, vector stores, fine-tunes, or projects persist? | Endpoint features that are ineligible for zero-data-retention treatment. |
| Security and abuse logs | Who can access them, what triggers retention, and how long are they kept? | Safety classifiers, incident investigations, or legal obligations. |
| Audit and compliance records | Are administrator actions and exports retained on a different schedule? | Fixed compliance periods that cannot be shortened by a chat setting. |
| Connected-app copies | Does the AI provider sync or index content, and does the app retain request data? | Third-party terms, provider backups, or a separate subprocessor. |
Provider-specific settings and documented periods
ChatGPT managed workspaces and personal accounts
For personal ChatGPT accounts, the “Improve the model for everyone” control governs whether new conversations are used to train models; switching it off does not remove saved chats from history. Temporary chats do not appear in history or train models, but OpenAI says they may be retained for up to 30 days for safety. Treat that as a maximum stated for the feature, not as proof that every record disappears immediately.
OpenAI states that content from ChatGPT Business, Enterprise, Edu, and Healthcare workspaces is not used for training by default. Enterprise, Edu, and Healthcare administrators control retention, and deleted conversations in those workspaces are removed from systems within 30 days unless a legal requirement requires longer retention. Business has separate terms and retention conditions, so verify the workspace policy rather than inferring it from the product name. See OpenAI’s enterprise privacy statement.
Rank #2
- Certified to FIPS 197 - High-level information security standard approved by the U.S. Government
- Brute-Force Password Attack Protection - Data is automatically erased after 6 failed access attempts. The data and encryption key are securely destroyed and the crypto drive is reset
- Auto-lock - The crypto drive will automatically encrypt all data and lock when removed from a PC/Mac or when the screen saver or "computer lock" function is activated on the host PC/Mac
- Secure Entry - Data cannot be accessed without the correct high-strength alphanumeric 8-16 character password. A password hint option is available. The password hint cannot match the password
- SuperSpeed USB 3.0 - Transfer all your confidential files and folders faster than ever before. Works on both PC & Mac
Before enabling a connector, confirm whether it is a synced app or a non-synced app. OpenAI says synced-app indexes are maintained separately. For non-synced apps, information sent to the app is governed by that third party’s storage, processing, privacy, and residency terms. The administrator controls and app distinctions are described in OpenAI’s admin controls documentation.
Claude Enterprise and the Claude API
Anthropic’s standard API treatment says inputs and outputs are automatically deleted from its backend within 30 days, subject to exceptions for features with longer retention, agreement terms, policy enforcement, or law. Do not apply that 30-day statement to every Claude product or to every feature.
Claude Enterprise provides a custom chat and project retention control. An Owner or Primary Owner configures it under Organization settings > Data and Privacy. Anthropic documents a 30-day minimum. Project retention can take precedence over chat retention, and project activity can reset the period. Changes and deletions are tracked in audit logs. Details are in Anthropic’s Enterprise retention instructions.
Anthropic’s API zero-data-retention (ZDR) arrangement is enabled for an organization and does not automatically extend to other organizations under the same account. It covers only eligible features. If the service is delivered through Amazon Bedrock or Google Cloud’s Agent Platform, the cloud provider is the data processor, so review that platform’s retention documentation as well as Anthropic’s. Anthropic’s endpoint and feature notes are at the API and data-retention documentation.
Some Claude-adjacent records have much longer schedules. Anthropic documents six-year retention for the Claude Compliance API Activity Feed. It also says Cowork and Claude Code local session transcripts may be stored for six years by default unless an organization has a finite custom period. Those durations apply to the named features, not to all Claude conversations.
Gemini in Google Workspace
Separate Gemini embedded in a qualifying Google Workspace edition from the Gemini app used as an additional service by users without a qualifying edition. Google says Workspace submissions are not used to train models outside the domain without permission, while additional-service use can fall under separate Gemini Apps terms and activity settings. A company email address alone does not prove that a user is in the managed Workspace experience.
Workspace administrators can restrict Gemini’s access to Workspace data, and Google says existing permissions and content-owner restrictions govern what Gemini can retrieve. Review group membership and source permissions before granting broad access. See Google’s Gemini and Workspace access guidance.
A Google administrator help page documents Gemini app conversation-history choices of 3, 18, or 36 months, or history off, and reports 18 months as the default. The page is an older, 2025-era reference; verify the current default and available choices in the Admin console before publishing an internal policy. With history off, Google says chats can remain in user accounts for up to 72 hours for service provision and feedback processing. The referenced admin page is Google’s conversation-history setting documentation.
Rank #3
- Certified to FIPS 197 - U.S. Government Approved High Level Information Security Standard.
- Protection against brute force password attacks - Data is automatically erased after 6 unsuccessful access attempts. The data of the USB flash drive type c encryption with dual connectors is destroyed and the cryptographic drive is reset.
- Durable dual-layer waterproof design* — Protects the crypto reader from bumps, drops, run-in and immersion in water. The electronics are protected by a hardened internal case. Rubberized silicone outer case provides a final layer of protection.
- Auto-Lock —The cryptographic key automatically encrypts all data and locks when removed from a PC/Mac or when screen protection or "computer lock" is enabled.
- Secure Entry —Data on these flash drives cannot be accessed without the correct alphanumeric password of 8 to 16 characters. A password indication option is available for this flash drive. The hint cannot match the password.
Microsoft 365 Copilot with Anthropic models
Microsoft describes two different arrangements for certain Anthropic models in Microsoft 365 Copilot. In one, Anthropic operates as a Microsoft subprocessor under Microsoft terms and there is no additional Anthropic retention. In another, models with Data Retention are off by default, require tenant-administrator opt-in, and use separate terms. Microsoft says Anthropic retains most inputs and outputs for up to 30 days in that latter arrangement, with potentially longer retention for content flagged by safety classifiers.
Model availability and eligibility vary by organization. Check the model label and tenant setting shown at the time of use; do not assume that a Copilot control covers every model route. Consult Microsoft’s AI subprocessor documentation.
API and zero-data-retention checks
API buyers need an endpoint-by-endpoint review. Ask four questions for every request path:
- Is the endpoint eligible for Zero Data Retention or Modified Abuse Monitoring?
- Does the request create application state such as a thread, file, vector store, fine-tune, or agent?
- Which abuse-monitoring and safety logs remain, and for how long?
- Does a marketplace deployment place a cloud provider between the customer and model vendor?
OpenAI says approved organizations can configure Zero Data Retention or Modified Abuse Monitoring at organization and project levels, but its endpoint table lists exceptions. ZDR does not automatically erase application state, and some ZDR settings change API behavior. Review the eligibility table and limitations in OpenAI’s API data-controls guide before making a promise to developers.
Anthropic’s ZDR arrangement likewise covers only eligible features and the organization for which it is enabled. A cloud marketplace’s processor terms and logs must be evaluated separately.
Review connected apps as an access problem, not only a storage problem
Retention is only half of the risk. A connector can expose more content than intended if its scopes or user mapping are broad.
Check read and write scopes
- List every read scope, including files, mail, calendar, tickets, repositories, and directory data.
- List every write or action scope, such as sending mail, editing records, creating tickets, or changing permissions.
- Disable write actions until the read path has been tested and approved.
- Use a dedicated service identity or least-privilege group where the product supports it.
Confirm identity and permission inheritance
Determine whether retrieval runs as the individual user, a shared service account, or an administrator. Test with two accounts that have different source permissions. The assistant should not return a document that the requesting identity cannot open directly.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Find copies and indexes
Ask whether the provider synchronizes content, builds an index, stores embeddings, or caches snippets. Record how to pause synchronization, remove the index, and verify completion. Disconnecting an app may stop future retrieval without deleting already-created copies.
Rank #4
- FIPS 197 with XTS-AES 256-bit Encryption: Provides business-grade security with hardware-based encryption to protect your sensitive data
- Brute Force and BadUSB Attack Protection: Safeguards against unauthorized access attempts and malicious USB attacks with digitally-signed firmware
- Multi-Password Option with Complex/Passphrase modes: Offers flexible password configuration options to meet various security requirements and user preferences
- New Passphrase Mode: Enhanced security feature allowing users to create longer, more memorable password phrases for easier access without compromising protection
- Dual Read-Only (Write-Protect) Settings: Enables write protection functionality to prevent accidental data modification or deletion when needed
Review the third-party processor
OpenAI says information sent to non-synced apps follows the third party’s storage, processing, privacy, and residency terms. Obtain those terms and identify subprocessors before enabling the app. If the connector is delivered through a cloud marketplace, review the marketplace provider’s logs and deletion process as well.
Data residency is narrower than many teams assume
A selected storage region may cover only certain customer content. OpenAI warns that residency for supported stored content does not automatically place every processing step, system record, external integration, or third-party app in the selected region. Ask the provider to distinguish:
- Primary customer-content storage;
- Model inference and temporary processing;
- Security, abuse, and audit systems;
- Backups and disaster-recovery copies;
- Connector or marketplace processing; and
- Support access and administrator telemetry.
Write the geographic scope beside each retention period. “Stored in the EU” is not a complete answer if retrieval, safety review, or an external app occurs elsewhere.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
A step-by-step pre-connection review
- Classify the data. Mark personal data, confidential business information, regulated records, credentials, source code, and information subject to contractual or legal restrictions. Define what must never be sent.
- Record the exact route. Capture provider, product, plan, tenant, endpoint, model, connector, region, and enabled features. Include whether the route is first-party, API-based, or marketplace-hosted.
- Capture current controls. Screenshot or export training, feedback, history, memory, retention, deletion, ZDR, app-scope, and residency settings. Include the administrator and timestamp.
- Build the data-flow register. For each prompt, file, response, index, log, and audit record, name the system of record, processor, retention period, deletion trigger, and exception.
- Test permissions. Use a test document visible to one user but not another. Confirm that the assistant follows the underlying source permission and cannot perform unapproved write actions.
- Test deletion. Submit harmless test content, delete it through the documented path, then check history, files, project state, indexes, exports, and audit views. Record which copies remain and why.
- Test the off switch. Disable history, memory, training use, or the connector one at a time. Verify which behavior changes and which records remain available.
- Approve by risk tier. Start with public or synthetic data. Require security and legal review before allowing sensitive or regulated content.
- Assign ownership. Name an administrator responsible for quarterly checks and for reviewing provider notices, model changes, and new connector scopes.
Use a comparison matrix instead of declaring one provider “most private”
Different deployments can be suitable for different requirements. Compare the actual configuration on these axes:
| Axis | Evidence to collect | Decision question |
|---|---|---|
| Training and feedback | Default, opt-in or opt-out scope, treatment of ratings, and treatment of retrieved content | Can any user action send business content into improvement workflows? |
| Retention layers | Chat, API logs, files, indexes, application state, safety records, compliance feeds, and local transcripts | Which record survives after a user deletes a chat? |
| Deletion | Self-service and admin controls, stated time to deletion, export coverage, and exceptions | Can the organization prove that required copies were removed? |
| Connection scope | User identity, read/write scopes, group restrictions, sync behavior, and third-party terms | Can the assistant retrieve or change more than the user can? |
| Governance | Audit logs, policy granularity, residency, support access, and endpoint eligibility | Can an administrator detect and investigate misuse? |
| Product boundary | Consumer versus managed workspace, first-party versus third-party integration, and marketplace processor | Are users actually operating under the organization’s intended contract? |
Common failure modes and corrective actions
“Training is off, so nothing is stored.”
Why it fails: training, history, safety logs, and application state are separate systems. Correction: document each retention layer and its deletion schedule.
“History off means the old chats are gone.”
Why it fails: disabling history can affect future display while leaving existing records or short-term service copies. Correction: run a deletion test and obtain the provider’s stated backend timeline.
“Our business email guarantees the enterprise policy.”
Why it fails: a user may be in a consumer or additional-service experience. Correction: verify the tenant, managed domain, plan, and admin console shown for the session.
“Disconnecting the app removes the index.”
Why it fails: synchronized copies, embeddings, backups, or logs may have separate lifecycles. Correction: use the provider’s index-removal procedure and request confirmation.
Best Value
- FIPS 140-3 Level 3 (Pending) Certified Military-Grade Security
- OS/Device Independent
- XTS-AES Hardware Encryption
- Enforced Alphanumeric PIN
- Multi-PIN (Admin and User) Option
“Zero Data Retention covers every API feature.”
Why it fails: ZDR is endpoint- and feature-specific and may exclude application state. Correction: check the current eligibility table for every request type.
“Regional hosting keeps all processing in one country.”
Why it fails: safety systems, support, external apps, and marketplace processors may be outside the selected region. Correction: map geography per data class and processor.
Make the review repeatable
Store a one-page control record for each production integration. It should include the owner, business purpose, approved data classes, prohibited data, provider and plan, endpoint and model, connector scopes, identity model, retention periods, deletion procedure, residency statement, subprocessors, and links to the provider’s current documentation.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsRecheck the record at least quarterly and whenever any of these events occurs:
- A provider adds a model, agent, memory, file, or connector feature.
- The organization changes plan, region, contract, or cloud marketplace.
- An administrator changes history, retention, ZDR, or app scopes.
- A security incident, legal hold, deletion request, or regulator inquiry occurs.
- A vendor publishes revised terms or changes a default.
For regulated or high-sensitivity environments, an independent AI data-governance or privacy assessment can help map provider settings to contractual requirements. Treat that as an implementation service, not as proof that a product is legally compliant for a particular use case.
What to approve before production
Approve the connection only when the team can answer, in writing, what data enters the service, which identity authorizes access, every place a copy can exist, how long each copy remains, how deletion is initiated and verified, which exceptions apply, and who monitors policy changes. If any answer is “not stated,” restrict the integration to synthetic or public data until the provider supplies a usable answer.
Frequently Asked Questions
Does turning off model training delete my company’s previous chats?
No. Training use and storage are separate. For example, OpenAI says chats can remain in history when “Improve the model for everyone” is off. Use the product’s deletion workflow and check its stated backend timeline for existing records.
Free tools Windows power users keep installed
One-click scans. No signup required.
Is a 30-day retention statement the same for every feature?
No. OpenAI temporary chats, OpenAI enterprise deletion, Anthropic API inputs and outputs, and Microsoft’s Anthropic Data Retention arrangement each describe different products or paths. A 30-day period in one feature does not establish a universal provider policy.
What should I test when an AI connector can access our Drive or mail?
Test two users with different source permissions, inspect read and write scopes, identify whether the provider creates a synchronized index, and verify how to remove that index. The assistant should not return content that the requesting user cannot open directly.
Can zero-data-retention settings cover an AI agent’s files and threads?
Not automatically. ZDR arrangements are limited to eligible endpoints and features; agent threads, files, vector stores, or other application state may have separate storage rules. Check the current endpoint table and feature limitations for the exact API route.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
