Using ReticulOS
AI Mode
AI Mode turns each collection into a knowledge base you can ask. Open the chat window, pick a collection — or all of them — and ask in plain language. Answers stream in with citations that link back to the exact cells they came from.
Opening the chat
Press Ctrl+Shift+Space anywhere, or click the AI button in the widget. The chat opens in its own window with a collection picker at the top — choose one collection to scope the conversation, or All collections to ask across everything you've saved.
Setup: bring your own key
AI Mode is BYOK — bring your own key. ReticulOS ships no bundled AI service and no account: you connect a provider you already use, and requests go directly from your machine to that provider.
- Open the chat window and click the ⚙ settings icon.
- Pick a chat provider — Google Gemini (the default), OpenAI, Anthropic, or OpenRouter — and paste an API key from that provider's console.
- Click Test. On success, the model dropdowns fill with the provider's live model list — no model ids to hand-type.
Chat and embeddings are configured separately, and can use different providers. Embeddings are what let the AI search your notes by meaning; the embedding provider can be Google, OpenAI, or OpenRouter.
Using Anthropic? Anthropic offers chat models but no embeddings API, so pair an Anthropic chat model with an embedding key from Google, OpenAI, or OpenRouter. Without an embedding key the AI can still chat, but it can't search your notes.
Keys are stored on your machine, encrypted with your operating system's secure storage — the same protection as the database key. They are never displayed again after you save them, never leave your device except to authenticate with the provider you chose, and never touch the app's UI layer again.
Indexing a collection
Before the AI can answer from a collection, that collection's cells need to be indexed — split into readable chunks, embedded through your embedding provider, and stored in a local vector index. When you open a chat on a collection that isn't fully indexed, a bar appears telling you how many notes are covered, with an Index now button. Indexing runs in the background with live progress; you can keep chatting while it works.
- New and edited cells are picked up automatically — saving a cell re-indexes just that cell.
- Index all items in settings queues every collection at once.
- Unchanged cells are skipped on re-index, so re-running it is cheap.
Cost warning: embedding uses your API key. Switching the embedding model or provider, or changing the embedding dimensions setting, invalidates the whole index and automatically re-embeds every item — on a large library that's a real API bill, so change those settings deliberately.
Asking
Ask like you'd ask a colleague. Behind the scenes ReticulOS retrieves the most relevant chunks from the collection — combining semantic search with keyword search — and hands only those to the model.
- Streaming answers render live, with the same rich formatting your cells use.
- Citations like
[#12]are clickable and jump to the source in the panel below the answer, so you can verify every claim against your own notes. - Sources panel shows exactly which cells were retrieved for the answer.
- Save as cell stores an answer back into the collection as a normal, editable cell.
The AI also covers your Schedules: day-book entries are indexed with their date and tags, so you can ask things like "what did I work on last Tuesday?" — and each day has a one-click AI summary.
Privacy
- Private by design: the vector index, chunks, and chat configuration all live inside your encrypted database. Nothing is stored on our side.
- What leaves your machine: when indexing, cell text is sent to your chosen embedding provider; when asking, your question plus the retrieved snippets are sent to your chosen chat provider. That's the whole surface.
- The vault is never indexed. Password vault entries are excluded from search and AI by design and never appear in the index.
Troubleshooting
| Symptom | Fix |
|---|---|
| "Invalid key" or auth errors | Re-paste the key and click Test — it checks both chat and embeddings and names which one failed. OpenRouter phrases an invalid key as a missing authentication header. |
| Answers ignore your notes | The collection probably isn't indexed — look for the index bar, or check that an embedding key is set (a chat key alone isn't enough for retrieval). |
| Some notes stay unindexed | Items that failed repeatedly are parked; the chat offers a retry that re-queues them. |
| Model dropdown is empty | The model list is fetched live from the provider, so it needs a working key and a network connection. |