Enrich documents with a local AI model
Use a local AI model running on your computer through Ollama to describe your documents, so Plexii finds and grounds answers better, with no credits used.
If you run AI models on your own computer with Ollama, PlexiDesk can use one to describe your documents: a short summary of each, what kind of document it is, the people, organisations and dates it mentions, a handful of keywords and its language. Plexii then finds the right documents more easily, and grounds its answers in richer context.
The work happens on your computer, with your local model. It uses no PlexiDesk credits and no API key.
Before you start
- Local AI enrichment is on every plan.
- Install Ollama on this computer and keep it running. Ollama is a free app for running AI models locally. Get it from its own website.
- Download at least one writing model in Ollama, such as a Llama, Qwen, Mistral, Gemma or Phi model. PlexiDesk uses it to describe your documents.
- Optionally, download an embedding model, one whose name includes embed. PlexiDesk uses it to rank your documents by meaning when you search or ask Plexii.
- PlexiDesk looks for Ollama at its standard address on this computer,
localhost:11434. You do not need to set anything up in PlexiDesk.
Find the settings
- Click the gear at the right-hand end of the header to open Settings.
- Click the AI tab.
- Under AI · source, find Local AI enrichment.
Check that PlexiDesk can see your model
A card under the heading shows what PlexiDesk found:
- Local AI is running. Writing model and embeddings, each followed by the model's name, or none if there is no model of that kind.
- No local model detected at localhost:11434. Start Ollama — PlexiDesk cannot reach Ollama, or Ollama has no models yet.
If you have just started Ollama or downloaded a model, click Refresh status to check again.
What the assistant can see
The box headed What the assistant can see shows how much of your work Plexii can draw on.
| Line | What it counts |
|---|---|
| Searchable | How many of your documents, files and conversations are indexed so Plexii can search them. Documents with no text are counted separately as empty. |
| Described | How many documents have been described by your local model. |
| Ranked by meaning | How many documents can be found by meaning, not only by the exact words. |
| Read for memory | How many documents have been read for facts Plexii can remember. |
A count that is not complete yet shows in bold.
Enrich your documents
You do not have to do anything: while a local model is running, PlexiDesk works through anything outstanding by itself, a few documents at a time, shortly after you open it and then every few minutes. The box says so: Anything outstanding is picked up automatically in the background.
To describe everything now:
- Click Enrich my documents. The button says Enriching… while it works.
- When it finishes, a line says how many documents it described, for example Enriched 12 documents, 30 already done. The AI now has richer context to answer from.
Documents already described are left as they are. Each document is described from its first 6,000 characters, one document at a time, so a large workspace can take a while. You can keep working meanwhile.
Enrich my documents is greyed out unless a writing model is running.
What changes for you
- Better answers. Plexii can match your question to the right documents by their summary, people, dates and keywords. See What Plexii can read and search.
- Search by meaning on this computer. With an embedding model, PlexiDesk ranks documents by meaning with your local model. See Add your own API keys.
- Nothing leaves your computer for this. Your documents go only to the model running on this computer. See Privacy settings and what leaves your computer.
Tips
- Leave Ollama running while you work. The background pass then keeps new and changed documents described, with nothing for you to click.
- A smaller model is faster. If enrichment is slow, try a smaller writing model in Ollama.
- Check the counts after a big import. Click Refresh status to see the new documents being picked up.
Common questions
It says no local model detected, but Ollama is open. Ollama needs at least one model downloaded. Download one, then click Refresh status.
Enrich my documents is greyed out. PlexiDesk found only an embedding model, or no model. Download a writing model in Ollama and click Refresh status.
It says some documents were skipped. The local model could not describe them, often because it took too long or returned something unexpected. They are tried again in a later pass.
Does this use my PlexiDesk credits? No. Local enrichment runs only on your computer.
Can I stop enrichment once it has started? Let it finish. It runs one document at a time, and you can keep working.
Related
Was this helpful?
Still need help? Contact support at [email protected].