An internet outage does not have to stop work that depends on notes already on your computer. You can keep drafting and editing if the AI model and source files are local.
OGAD (Off Grid AI Desktop) runs local text models after setup. Prepare the app and a model before you need them, keep your working material available, and use source-grounded prompts when the connection disappears. You can finish a useful draft and leave online checks for later.
| Download OGAD | 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.
This works best for writing based on known information: a project update, a proposal outline, an article from your own notes, or an edit of an existing draft. It does not make live research available without a network.
What has to be ready before the outage?
Install the app, download a suitable local text model, and test it while disconnected. Save real local copies of the documents you use. A cloud shortcut may still need a connection before it can open.
The free core workflow is available on supported Mac and Windows computers. Release 0.0.54-beta.108 also offers Linux beta packages.
If the outage has already started and the model files are absent, local inference cannot download them without connectivity. Prepare the baseline when you have a connection rather than assuming the app alone contains every model.
Use a model that fits the available memory and a short task for the first test. Keep the setup simple enough that you know which resources the task depends on.
Which writing jobs can you continue?
Choose jobs where the evidence is in your saved material. Keep a separate list of facts, links, and decisions that need checking once the connection returns.
Suppose you are writing a client update from yesterday’s meeting notes. You can organise completed work, open questions, and next steps without new web information. You cannot confirm a supplier’s current availability unless that answer is already in the records.
| Writing task | Offline source to prepare |
|---|---|
| Project update | Checked notes and current status |
| Article outline | Your source notes and saved references |
| Proposal draft | Client brief and approved scope facts |
| Copy edit | The current draft and voice guidance |
| Meeting agenda | Earlier notes and open decisions |
The useful distinction is between drafting from known facts and obtaining new facts. Local AI supports the first when the required material is ready.
How do you start the local writing session?
Select the downloaded local model and supply the relevant source. Use a fresh chat for a focused piece of work rather than carrying a long, mixed conversation into the outage.
- Open Models > Text and select your downloaded model.
- Start a new chat.
- Paste short notes or use + > Attach files for a saved document.
- Inspect the extracted text when using a file.
- State the audience, purpose, and facts the draft must preserve.
The file-processing implementation reads local source material for the model. Use readable TXT, Markdown, DOCX, or text PDFs; scanned documents may need checked text first.
For the client update:
Draft a short project update from these notes. Separate completed work, work in progress, and questions needing a decision. Use only the supplied facts. Do not invent a deadline, approval, or current external information.
How do you keep the draft useful without live research?
Mark missing information instead of asking the model to fill it from memory. An answer can sound current while relying on general training data that does not establish today’s facts.
Use clear markers in the working draft, such as VERIFY LINK, CONFIRM DATE, or CLIENT DECISION NEEDED. Ask the model to preserve them through revisions.
Improve the structure and wording of this draft. Keep every verification marker. Do not replace missing facts with plausible values. Put any new question you notice in a separate checklist.
This allows you to make progress on the argument and language without losing track of the checks that remain.
How should you edit with a local model?
Give a specific editing goal: shorten repetition, clarify a paragraph, improve the order, or make the action clearer. Ask the model to preserve facts and uncertainty.
For example:
Edit this update for clarity. Keep the distinction between completed testing and planned testing. Use short paragraphs. Do not add enthusiasm, guarantees, or new information.
Compare the revision with the original. A model can accidentally strengthen a claim while making it smoother. Watch for changes from “may” to “will,” “reviewed” to “approved,” or “in progress” to “complete.”
Keep your own copy of the working draft in a local editor. The chat helps you revise, but your saved document is the place to assemble and review the final version.
What if a long draft becomes difficult to handle?
Work in sections and keep a short statement of purpose and style. The model’s context must fit the text, instructions, conversation history, and answer. Longer input can reduce the quality of a focused edit or exceed the usable context.
Start a fresh chat with the section and the instructions it needs. After editing the parts, read the whole document yourself for consistency and repeated points.
If the laptop struggles, try a smaller model, shorter source, or fewer concurrent applications. A connection outage and a memory limit are different problems; solving one does not solve the other.
How do you prepare for the connection returning?
Finish with a handoff note for yourself. List the draft location, what you completed, and the checks still needed. That makes it easier to resume without rereading the whole conversation.
When online again:
- Verify current facts and links.
- Confirm any outstanding client decisions.
- Check the final draft against the sources.
- Review the destination and version before sharing.
This workflow does not automatically send the draft when connectivity returns. Publishing and messages remain separate actions through the tools you choose.
What should you test in advance?
| Check | A useful test |
|---|---|
| Model readiness | Start a new response with Wi-Fi and Ethernet disconnected |
| Source availability | Open the actual saved file |
| Extraction | Attach a document and inspect its text |
| Writing quality | Edit a paragraph with known facts |
| Saving | Store the reviewed output in a local document |
If you use remote models or connected tools in other work, make sure the outage workflow selects the local model. A private server address still needs a reachable network.
Keep one dependable local writing route
Download OGAD, prepare a modest text model, and test one draft from saved notes without internet. Keep the source and output locally available.
The useful result is continuity. When the connection fails, you can keep working on the parts that depend on your own material and return to online checks later.