The answer may be in the facilitator guide, the participant handbook, or last month’s course update. You need to find the right instruction without opening every document in turn.
OGAD (Off Grid AI Desktop) can search a project of training documents and use relevant passages to answer a question. Select a local text model and keep the collection on your computer. After installation and indexing setup, you can use this free core workflow without sending the documents to a cloud AI service.
| 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.
For a small training provider, this is useful when a tutor needs to check an exception or prepare an answer between sessions. It can also help a course administrator locate the approved wording before replying to a learner.
Which training documents belong together?
Group material that supports the same course, audience, and version. A mixed collection of current and retired courses can produce plausible but irrelevant answers. Keep the intended learner and course edition visible in the project name and files.
Suppose your team runs a facilitator course. The handbook explains preparation, the trainer guide describes the workshop exercises, and a later update changes how learners submit an assignment.
A question about submission should use the current update. A question about running an exercise may need the trainer guide. Clear source labels help you judge the result.
| Source | Useful label |
|---|---|
| Participant handbook | Course name, edition, intended learner |
| Trainer guide | Trainer-only context, edition, approval status |
| Assignment instructions | Task name and current version |
| Course update | Effective date and what it replaces |
| Internal notes | Draft or discussion status |
If the project contains trainer material, remember that it is your local working collection. Adding files does not create role-based learner permissions. Decide separately what you will share with learners.
What do you need to start?
Install OGAD, choose a downloaded local text model, and complete local indexing setup while connected. The free core Projects workflow is available on supported Mac and Windows computers. Release 0.0.54-beta.108 also provides Linux beta packages.
Use readable TXT, Markdown, DOCX, or text PDF files. Scanned handbooks need a text layer or a checked text version. Imported tables and diagrams may lose context, so inspect the original when an answer depends on layout or an illustration.
Start with three or four documents that answer questions you already know. This lets you check the collection before depending on it during a live session.
Keep model choices local for the processing described here. A remote text model changes where the answer is generated and requires its configured connection.
How do you build the first collection?
Create a project for one course and add the current approved documents. Wait for indexing to finish. Start a chat from inside the project so the question has the intended knowledge scope.
- Select the downloaded model in Models > Text.
- Open Projects > New project, enter the course name, and press Enter.
- Open Knowledge & settings > Knowledge base > Add files.
- Add the documents and wait for indexing. Keep the intended files enabled for retrieval.
- Open Chats > New chat inside that project.
The project interface provides the file list and retrieval switches. If you use Pro and want a selected-document collection, leave Include captured memory off in project settings and save the choice.
You can add a short system prompt:
Answer course questions from the current approved documents. Name the source for instructions. Distinguish trainer guidance from learner instructions. If a source is missing or the documents disagree, state the gap instead of guessing.
These instructions guide the response; they do not guarantee correctness.
How should you phrase a learner’s question?
Include the task and relevant condition. A specific question is easier to match to the right passage than a broad request to explain the whole course.
For example:
What should a learner submit for the workshop-planning assignment if they cannot attend the live session? Use the current course documents and name the source. Do not infer an exception from general attendance notes.
Open the returned source and check the condition. If the handbook explains normal submission but says nothing about absence, the answer should remain incomplete until you confirm the exception.
Ask a follow-up when needed:
Which part of that answer is stated in the source, and which part is an interpretation? Quote the wording that supports the submission requirement.
Check the quoted text yourself. A citation gives you a route to review the evidence, not a guarantee that the model interpreted it correctly.
What if two documents give different instructions?
Compare their edition, intended audience, and approval status. A trainer’s preparation note may describe something the learner never has to submit. A newer draft may not replace an older approved handbook.
Ask:
Show the conflicting instructions from both sources. Include their dates and labels where present. Do not choose the authoritative version unless the documents explicitly establish which one replaces the other.
Then resolve the conflict through your course process. Update the approved instruction and disable outdated files for retrieval when appropriate. Past project conversations can still contain earlier discussion, so keep the current source explicit in later questions.
For the assignment example, save a short approved clarification that names the affected course edition and condition. That is more useful than leaving the answer only in an informal chat.
Can you trust it to find every relevant document?
Project search retrieves selected passages within a limited context. It can help you locate an answer, but it is not an exhaustive audit of every page. The project chat path uses bounded retrieval before generating the response.
For a course update that affects several documents, use your own checklist of files to review. Ask about one topic at a time, then inspect the relevant originals.
Keep a short list of test questions with answers you have already verified. Recheck them after changing the collection. Useful tests include an ordinary instruction, an exception, and a question the documents cannot answer.
What should you check when an answer is weak?
| Symptom | Next check |
|---|---|
| The answer uses the wrong course | Check the active project and source filenames |
| A diagram is misunderstood | Inspect the original and add a checked text explanation |
| A retired instruction appears | Check document versions and enabled sources |
| The model invents an exception | Ask for the exact source or mark the question unresolved |
| The answer is too broad | Name the learner task and condition more precisely |
Use the answer as a draft for your reply. Review it before giving a learner an instruction, especially when a deadline or assessment requirement is involved.
Answer one real course question
Download OGAD and add a small set of current training documents. Ask a question you expect during the next session, inspect the source, and save a checked answer.
The useful result is a faster route back to the approved material. It gives your team somewhere to start when the same practical questions appear again.