You need an answer from your company’s documents, but uploading them to a public AI service may not be an option. OGAD (Off Grid AI Desktop) can search files on your own computer and use a local model to answer. Add the documents you are allowed to use, ask a focused question, and check the source behind the reply.
| Download OGAD for Mac or Windows | Desktop releases |
What would you like to do with Off Grid AI?
Have a feature or use case you would like us to support? Tell us what you want to do and which device you use.
Write to support@offgridmobileai.co, join our Slack community, or talk to us on Reddit.
A useful first task is finding a process in an internal handbook. You keep the original file available and use the answer to find the relevant passage, rather than treating generated text as a new company policy.
What do you need before importing company files?
Use a supported Mac or Windows computer, a downloaded local text model, and readable files you are authorized to store and process there. Projects and document search are free core features.
Start with a small set of text PDFs, DOCX files, or checked text notes. Scanned PDFs need a searchable text layer first. Complete app/model downloads and an initial import while connected before testing offline.
Local processing does not replace your organization’s data rules. A project is not a separate employee account, access-control system, or certification that a use is approved.
How do you make a project for the documents?
Create a dedicated project and import only the material needed for your task. Keep related versions clearly named so the answer can point you back to the correct source.
- In Models > Text, select a downloaded local model.
- Open Projects > New project, enter a name such as “Internal handbook,” and press Enter.
- Open Knowledge & settings > Knowledge base > Add files.
- Import the approved files and wait for indexing.
- Keep the current files enabled for retrieval. Disable old versions you do not want used.
- Start Chats > New chat inside that project.
If you use Pro, turn off Include captured memory for this project and select Save when the task should use only the uploaded material and project discussion. Review device sync before importing: paired workspace sync includes all projects, with no per-project switch. Use an unpaired installation if the files must stay on one computer.
What should you ask first?
Ask a question whose answer you can confirm in the original. This checks the whole workflow before you use it for an unfamiliar process.
According to the uploaded handbook, what information must a new equipment request include? Name the source file and quote the short passage that supports the answer. Say when the documents do not specify something.
You should get a focused answer with a source you can inspect. Check the actual wording, revision date, and exceptions. A generated answer can still use an obsolete passage or miss a condition.
For a follow-up, separate the document’s rule from an interpretation:
Which parts of that answer are stated in the handbook, and which are your interpretation?
How do you keep answers tied to current policy?
Use clear filenames with revision dates. Disable superseded versions and begin a fresh chat after a material change. Earlier answers in the same conversation can remain context even when a file is no longer enabled.
The desktop project store scopes the document collection to a project. Search uses selected passages, so a broad question about every policy can miss relevant material. Ask one process question at a time.
There is no automatic approval by a policy owner. Verify any answer that will guide an operational decision against the current source and responsible person.
What if an answer is unsupported?
| Problem | Action |
|---|---|
| No source is named | Ask for the supporting passage and inspect the file. |
| The answer mixes departments | Separate their document collections into projects. |
| Two versions conflict | Identify the authorized current version instead of asking the model to guess. |
| A file cannot be read | Use an extractable text version and check indexing. |
| The reply depends on internet | Confirm that the active text model is local. |
Where does the document text go?
With local model selections, the computer extracts, indexes, and processes the document text locally. Initial setup can need downloads. Remote models, external tools, cloud-managed folders, and configured sync create separate data paths.
For a single-computer check, use local files and an unpaired setup, disconnect internet after model setup, and ask a new handbook question. Do not infer company-wide permissions from a successful offline result.
Get OGAD, start with one approved handbook, and verify one answer against its source before expanding the project.