Two versions of a proposal can look almost identical until a deadline or condition changes. OGAD (Off Grid AI Desktop) helps you compare their text on your own computer. Add the documents to a project, ask about a specific section, and check a side-by-side explanation without sending the files to a cloud AI service.
| 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.
Use this for understanding changes in meaning. It is not a guaranteed line-by-line diff, and it does not replace a specialist review of legal or other high-stakes documents.
What should you prepare?
Use two readable PDFs, DOCX files, or text files with clear version names. For example, proposal-original.docx and proposal-revised.docx. Install OGAD and select a downloaded local text model. Project document search is free on supported Mac and Windows computers.
Finish initial model setup and file indexing before you test offline. Scanned PDFs need a searchable text layer; tables and layout can lose structure when converted to text.
How do you compare the first section?
Create a dedicated project for the two versions. Import both files, then ask a narrow question that names both sources. Compare the answer with the same section in each original.
- Open Projects > New project, name it “Proposal comparison,” and press Enter.
- Open Knowledge & settings > Knowledge base > Add files and select both versions.
- Wait for indexing and confirm both retrieval switches are enabled.
- Open the project’s Chats > New chat.
- Ask about one named topic, such as delivery dates.
Try:
Compare the delivery dates in proposal-original.docx and proposal-revised.docx. Use columns for original wording, revised wording, and practical change. Name the supporting file for each entry. Say if you do not have the relevant passage from both files.
You should get a comparison you can inspect. Check dates, units, exceptions, and conditions in the originals. A clear table can still contain a model mistake.
Check a small example before comparing a full proposal
You can test the method with two short sample passages in a fresh local chat. Label them clearly:
Original: Deliver the draft on 12 October. The client must provide
feedback within three working days. One revision is included.
Revised: Deliver the draft on 15 October. The client must provide
feedback within three working days. Two revisions are included.
Ask for a table of changed dates, changed quantities, and unchanged conditions. These are invented practice passages, so you can inspect every result without exposing a real proposal.
A useful answer identifies the later draft date and the change from one revision to two. It should also retain the three-working-day feedback condition. It should not invent a price change, say that the client approved the revision, or turn working days into calendar days.
Now try a second request: “Explain these changes in two sentences for a project manager. Do not infer the reason.” Check that the shorter explanation keeps the same facts. This is where local AI adds value beyond a visual diff: it can help explain the practical meaning of selected changes once you have verified them.
For your real files, keep the same discipline. First establish the exact passages from both versions. Then request the comparison. Only after checking the changes should you ask for a summary to share with someone else.
What if only one document is represented?
Project retrieval selects relevant passages within a context limit. It does not guarantee an equal amount of text from each file. If the answer lacks one side, ask for that document’s relevant passage first.
For a short section, copy the exact text from both documents into a new local chat. Label the blocks Original and Revised, then ask for the comparison. Keep the combined text within the model’s usable context. This gives the model the actual sections you want to compare instead of depending only on retrieval.
The released text extractors read document text. They do not preserve every visual layout detail or provide a dedicated tracked-changes engine.
How do you review a longer change set?
Make a checklist from the document headings and compare one section at a time. Keep checked findings in your own notes. Ask the model to combine only those notes when you want a short change summary.
Include unchanged conditions that qualify a change. For example, a later delivery date might still depend on approval by a separate deadline. Asking only “What changed?” can lose that relationship.
For exact completeness, use a document comparison tool alongside the AI explanation. The local model is useful for explaining selected differences; it is not proof that no other changes exist.
What can make the comparison misleading?
| Problem | Action |
|---|---|
| Similar filenames confuse the versions | Rename them clearly before import. |
| The response gives a change with no supporting text | Ask for both passages and verify them. |
| A table’s columns are mixed up | Compare the original table visually or supply checked plain text. |
| One version dominates the answer | Paste the two relevant sections explicitly. |
| Old discussion affects the reply | Start a fresh chat in the dedicated project. |
Can the files stay off cloud AI services?
Select a local text model and prepare the search resources before disconnecting. In Pro, disable Include captured memory for this project if you only want the comparison material considered. Configured sync can still copy project files to paired devices; use an unpaired installation if they must stay on this computer. Paired workspace sync includes all projects; there is no per-project sync switch.
Install OGAD and compare one section from two versions. Check every stated change against both originals before you rely on it.