You write the note, attach whatever you want, and AiNotes chops it up, converts it into vectors using gemini-embedding-001, and gets it ready for queries. No need to press any "index" button.
The difference between a notebook and a notebook you can ask questions to lies in what happens right after saving.
AiNotes is the AI notebook from AIsToHire, and indexing is the piece that allows your notes to be searched by meaning rather than just by word match. The moment you save a note, its text and each of its attachments are queued as independent elements in your project's RAG index. There is no manual step: the note starts in a "processing" state and moves to "ready" when the index is complete.
From there, the pipeline works. Text is first extracted from the files (and a short summary of each is also saved); the resulting text is split into chunks of 400 words with a 50-word overlap, so that a phrase falling at the edge of a chunk still appears fully in the next one. Each chunk is converted into a 768-dimensional vector using Google's gemini-embedding-001 model and stored in the project's chunk tables, along with its order index and token count.
These vectors are what the chat uses. When you ask a question, your question is converted into a vector using the same model and compared for cosine similarity against the indexed chunks: the 5 most similar ones above a similarity threshold of 0.3 are retrieved, and these are what the assistant reads to respond, citing which notes each piece of data came from. In other words, it does not search for the word you wrote: it searches for what you meant to say.
The entire process is observable and retrievable. Each element of the index —the note text and each attachment separately— has its own state (pending, processing, ready, or error), the app refreshes that state every 5 seconds while something is in progress, and if an element fails, you can retry just that one. If a file has no extractable text —an unreadable scan, an image without letters, an empty PDF— it is marked as an error with the reason, never as ready: AiNotes prefers to tell you that the attachment is not searchable rather than let you believe it is. (Chunking values, dimensions, and threshold: current platform configuration as of November 8, 2026.)
No buttons
indexing starts when saving. There is no "process", "synchronize", or "rebuild index". You create or edit the note, and the work queues automatically.
Search by meaning, not by exact word
when comparing vectors, a note talking about "notice period" appears even if you ask, "How much notice do I need to give?".
Overlapping fragments
the text is cut into pieces of 400 words that overlap by 50, so an idea split in half doesn't get lost between two fragments.
768-dimensional vectors with gemini-embedding-001
the same model indexes your notes and your questions, which is what makes the comparison meaningful.
Attachments are included in the index, not just the text
text is extracted from each PDF or image and a summary is stored, making its content as searchable as what you wrote by hand.
One status per item, visible
the text and each attachment have their own status —pending, processing, ready, or error— and the app refreshes it every 5 seconds. You know exactly which part is ready.
Surgical retry
if an item fails, you retry that one and only that one. You don't reindex the entire note or have to re-upload the files.
Incremental reindexing when editing
if you change the text, the text is reindexed; if you remove an attachment, its part of the index is deleted; if you add one, the new one is indexed. What you haven't touched remains as it was.

AiNotes is the AI notebook from AIsToHire that indexes every note and attachment with Gemini embeddings so you can ask about its content. It’s available at https://aistohire.com/ainotes-ai-notebook-105