Choose an AI workflow according to the data and the task. A product name or plan label does not approve a use case. Review the exact contract and configuration with your organization.
We checked the linked vendor sources on October 14, 2026. This post reports no comparative product test.
ChatGPT Enterprise, Microsoft Copilot, LM Studio, Ollama, Otter, Cursor and Continue
1. ChatGPT Enterprise
OpenAI excludes ChatGPT Enterprise data from model training by default, unless the customer opts in. See its enterprise privacy terms. Review retention and connected apps separately.
2. Microsoft Copilot for organizational use
Microsoft states that prompts and responses remain within the Microsoft 365 service boundary unless the customer directs otherwise. Web search can send generated queries to Bing under separate terms. See the Microsoft web-search policy.
3. LM Studio
LM Studio documents offline use after model downloads. See its offline guide. Review the selected model, tools, extensions, and storage before sensitive use.
4. Ollama
Ollama supports cloud features as well as local models. Set OLLAMA_NO_CLOUD=1 to disable cloud features. Restart Ollama after the change. See the Ollama FAQ.
5. Otter Enterprise
The Otter retention guide carries an Enterprise label and requires an administrator to set a custom retention policy. The administrator arranges the policy with the account manager.
6. Cursor Privacy Mode
Cursor states that Privacy Mode excludes customer data from its model training. Its policy allows abuse-investigation retention and temporary encrypted file caches. Requests still pass through Cursor infrastructure. Non-ZDR models have separate controls. See the Cursor data-use policy.
7. Continue with a local model
The Continue offline guide requires a local model and disabled anonymous telemetry. Review all model roles, external tools, and extensions. Apply the Ollama cloud-disable control when you use Ollama.
PocketLLM is pre-launch. Private, on-device AI, starting on iPhone, with more platforms planned. No account required, no in-app tracking, and chat processing stays on your device. Join the waitlist.
ChatGPT, Copilot, LM Studio, Ollama, Otter, Cursor and Continue: data controls
| Option | Check before use |
|---|---|
| 1. ChatGPT Enterprise | Review the contract, retention, and apps. |
| 2. Microsoft Copilot for organizational use | Review web-search settings. |
| 3. LM Studio | Check downloads and optional tools. |
| 4. Ollama | Check the local model and server access. |
| 5. Otter Enterprise | Confirm the policy with the account manager. |
| 6. Cursor Privacy Mode | Review exceptions and the selected model. |
| 7. Continue with a local model | Check telemetry and all request destinations. |
Local AI and hosted services for work
Use an approved local configuration when the task requires local inference. Use an approved hosted service when the data rules permit it. Check connectors, shared folders, exports, and retention for either workflow.
Local inference keeps prompts on the device during inference when the selected model runs there. For a local-only workflow, select no cloud model. Disable chat sync and cloud backup. Check the app telemetry policy. Disable optional network features. Review extensions, exports, and local storage access.
AI for lawyers: what ABA Formal Opinion 512 says
This section gives product information, not legal advice. Check your professional rules and jurisdiction requirements before AI use. No entry provides a privilege-safety classification.
ABA Formal Opinion 512 discusses competence, confidentiality, client communication, and supervision. The opinion also covers candor toward the tribunal and fees. Its confidentiality analysis is fact-driven and depends on the client, the matter, the task and the GAI tool. The opinion requires informed client consent before a lawyer inputs information relating to the representation into a self-learning GAI tool. It states that boilerplate engagement-letter language does not provide sufficient informed consent.
Local execution does not replace that assessment.
Before use on a matter
- Check the applicable professional rules.
- Assess the client, matter, task, tool, and safeguards.
- Review the contract, storage, retention, and access controls.
- Determine whether the rules require client consent.
- Verify legal citations and each output before use.
Private LLM and Apple Writing Tools for lawyers: data checks
These two iPhone options are not among the seven entries above. For ChatGPT Enterprise and Microsoft Copilot, see entries 1 and 2.
| Option | Vendor source | Check before use |
|---|---|---|
| Private LLM | The Private LLM developer site documents offline use on iPhone after a model download. | Review device controls and the proposed matter before use. Review local access and backups. |
| Apple Intelligence Writing Tools | Apple documents text edits and summaries in Writing Tools. Apple Intelligence can also use Private Cloud Compute. | Check the exact feature offline on a supported device. Review the feature and request destination. |
AI for journalists: checks before source material enters an app
Source protection needs a review of the whole workflow. A local model does not establish complete source protection.
Before source material enters an app
- Check newsroom rules for the assignment.
- Remove names and details that can identify a source.
- Check access to the device, files, backups, and exports.
- Approve the exact tool and configuration before use.
Private LLM, Apple Writing Tools, Duck.ai, Claude, MLC Chat and LLMFarm: source checks
These six options are not among the seven entries above.
| Option | Vendor source | Check before use |
|---|---|---|
| Private LLM | The Private LLM developer site documents offline use on iPhone after a model download. | Review device access before use with source material. Check local files and backups. |
| Apple Intelligence Writing Tools | Apple documents text edits and summaries in Writing Tools. Apple Intelligence can also use Private Cloud Compute. | Check the exact feature offline on a supported device. Check each feature and request destination. |
| Duck.ai | The Duck.ai privacy policy describes requests to external model providers. Its default policy excludes model training on chats. Optional feedback and sync have separate data rules. | Check the newsroom data policy. Use only material that your newsroom permits in that service. |
| Claude | Claude accepts documents and images, according to its file upload guide. | Review each file for source identifiers before upload. Inspect the document and its metadata. |
| MLC Chat source build | The MLC iOS guide describes an Xcode source build with model libraries and model files. The guide also states that the App Store offers the MLC Chat app at no cost. | Treat the source build as a developer project. Confirm support for the exact model. |
| LLMFarm source project | The LLMFarm repository states that the app is temporarily unavailable through TestFlight and the App Store. | Treat the source project as a separate developer option. Check distribution status before use. |
Check AI summaries and source quotes
Use public material for the initial app check. Compare summaries with the original text. Check every quote against the source. Obtain an editor or security review before a sensitive task.
Before a work rollout
- Identify the data that the task needs.
- Remove unnecessary sensitive content.
- Confirm the permitted app, plan, and configuration.
- Test the task with non-sensitive material.
- Verify the output before use.
Frequently asked questions
Who can set an Otter custom retention policy?
The Otter retention guide carries an Enterprise label and requires an administrator. The administrator arranges the policy with the account manager.
How does Ollama disable cloud features?
Ollama documents OLLAMA_NO_CLOUD=1 as a control to disable cloud features. See the Ollama FAQ.
Does Cursor Privacy Mode remove all retention exceptions?
Cursor permits retention for abuse investigations and temporary encrypted file caches. Requests still pass through Cursor infrastructure. See the Cursor data-use policy.
When does ABA Opinion 512 require informed client consent?
The opinion requires informed client consent before a lawyer inputs information relating to the representation into a self-learning GAI tool. See ABA Formal Opinion 512.
Does boilerplate engagement-letter language provide sufficient consent?
The opinion states that boilerplate engagement-letter language does not provide sufficient informed consent. See ABA Formal Opinion 512.
Does Duck.ai use external model providers?
The Duck.ai privacy policy describes requests to external model providers. Optional feedback and sync have separate data rules.
Does MLC Chat require a source build?
The MLC iOS guide describes an Xcode source build and states that the App Store offers MLC Chat at no cost.
PocketLLM is pre-launch, with a waitlist for the iPhone launch. More platforms are planned. See the PocketLLM model catalog for its separate model list.