Ask your library, get answers with sources
The Assistant searches your notes, video transcripts and files before it answers, and numbers each claim to the place it came from. Open it from any page of the app.
Free to start · used by 32K+ note-takers
Press ⌘J on a Mac, or CtrlJ on Windows, from anywhere in the app.
Questions people actually ask
Most questions are one of three things: explain it, find it, or make something from it. The answers come from what you saved, so they use your course's terms and examples.
Understand
Explain it again, from your own material.
Explain the Krebs cycle like I missed the lecture.
Your Lecture 5 note starts from pyruvate. In short: each turn strips two carbons off as CO₂ and loads the energy onto NADH and FADH₂…
How does Lecture 4 connect to Lecture 5?
Photosynthesis stores energy as glucose; respiration releases it. Lecture 5 even reuses the ATP synthase diagram from Lecture 4…
Find
Get back to where you saw it.
Which video explained the electron transport chain?
“Cellular respiration, part 2”, from 14:05. Your note on it says the chain is where most of the ATP is made…
Where did I write about RuBisCO?
In two notes: “The Calvin cycle”, under Carbon fixation, and “C4 and CAM plants”, under Photorespiration…
Make
Turn the answer into something to study.
Make 10 flashcards from my respiration notes.
Done. The deck Cellular respiration has 10 cards and is in your Flashcards.
Draw how glycolysis feeds into the Krebs cycle.
Done. The flowchart Glycolysis to Krebs cycle is in your Diagrams.
Where it looks, and where it doesn't
Every question starts with a search of your library. Attach notes, videos, files or saved pages to keep the answer to those, or ask inside an imported YouTube channel to keep it to that channel's videos.
Searched on every question
- Your notes Cited by note and heading
- Video transcripts Cited to the second
- Uploaded files PDFs and documents, by page
- Saved pages The full text of pages you saved Starter and Plus
- Your question
- Up to 24 closest passages
- An answer that cites them
Left out
- Notes in the trash
- Workspace notes nobody shared with you
- Anyone else's conversations
- The web, unless it runs a search and links what it found
It can act, not just answer
Ask for the thing you need next and it makes it, saved to your library as an ordinary item. Ask for several in one message and it does them in turn.
- “Write this up as a note” A new note in your library
- “10 flashcards from this note” A deck of question, fill-in-the-blank and ordering cards
- “Draw it as a diagram” A diagram you can keep editing on a canvas
- “Summarize my last three lectures” One summary note across all three
- “Start a note from my Lab report template” A note with that structure, titled and tagged
- “Find duplicate tags and merge them” “bio” and “biology” become one tag
- “Delete my empty drafts” It asks you first, every time
Pick the model for each conversation
22 models from 10 labs, and every conversation keeps its own choice.
- DeepSeek
- OpenAI
- Anthropic
- xAI
- Moonshot
- Z.ai
- Qwen
- Mistral
- MiniMax
Free includes 8, with DeepSeek Flash as the default. The figure beside each model is the estimate the picker shows for a short message.
| Model | Credits / message |
|---|---|
| On every plan | |
| GPT OSS OpenAI · 131K context Intelligence 1/5 Speed 5/5 | ≈9 |
| DeepSeek Flash Default DeepSeek · 1M context Intelligence 2/5 Speed 5/5 | ≈16 |
| GPT-6 Luna OpenAI · 1M context Intelligence 2/5 Speed 4/5 | ≈24 |
| Claude Haiku 5.5 Anthropic · 1M context Intelligence 2/5 Speed 4/5 | ≈24 |
| GLM Flash Z.ai · 1M context Intelligence 3/5 Speed 2/5 | ≈30 |
| Qwen Flash Qwen · 1M context Intelligence 2/5 Speed 2/5 | ≈30 |
| Gemini Flash Lite Google · 1M context Intelligence 1/5 Speed 5/5 | ≈64 |
| MiniMax M3 MiniMax · 1M context Intelligence 2/5 Speed 4/5 | ≈64 |
| Starter and Plus | |
| Mistral Large Mistral · 262K context Intelligence 1/5 Speed 3/5 | ≈97 |
| Kimi K2.5 Moonshot · 262K context Reasoning 2/5 Speed 2/5 | ≈140 |
| GLM-5 Z.ai · 205K context Reasoning 3/5 Speed 3/5 | ≈198 |
| Grok 4.3 xAI · 1M context Reasoning 2/5 Speed 4/5 | ≈217 |
| DeepSeek Pro DeepSeek · 1M context Reasoning 4/5 Speed 3/5 | ≈226 |
| Gemini Flash Google · 1M context Reasoning 4/5 Speed 4/5 | ≈350 |
| GPT-5.6 Terra OpenAI · 1M context Reasoning 4/5 Speed 4/5 | ≈507 |
| Plus | |
| Grok 4.7 xAI · 500K context Reasoning 5/5 Speed 3/5 | ≈387 |
| Qwen Max Qwen · 1M context Reasoning 5/5 Speed 1/5 | ≈387 |
| GPT-6.1 Sol OpenAI · 1M context Reasoning 5/5 Speed 2/5 | ≈467 |
| Claude Sonnet 5.5 Anthropic · 1M context Reasoning 5/5 Speed 3/5 | ≈467 |
| Gemini Pro Google · 1M context Reasoning 3/5 Speed 4/5 | ≈507 |
| Kimi K3 Moonshot · 1M context Reasoning 4/5 Speed 1/5 | ≈700 |
| Claude Opus 5.5 Anthropic · 1M context Reasoning 5/5 Speed 3/5 | ≈934 |
You pay for what the model reads and writes
There is no flat price per message. Each one is charged from the tokens the model actually used, at that model's rate, so a short question on a small model costs little and a long one on a flagship costs more.
One question on DeepSeek Flash, the default
It reads
- Your question
- Up to 24 passages from your library
- The conversation so far
It writes
The answer, and anything you asked it to make
An average measured across real questions. The picker says ≈16 for this model because it prices a short message on its own; the passages from your library and the conversation add the rest. Long answers and web searches cost more.
Monthly credits, in questions on the default model
- Free 1,000credits ≈16 questions
- Starter 30,000credits ≈500 questions
- Plus 120,000credits ≈1,900 questions
Free to use, every time
Searching your library, listing, tagging, sharing and starting notes from templates use no credits. Only the model's reading and writing is charged.
When credits run out
On Free, the Assistant stops until your credits reset. On Starter and Plus it keeps answering on DeepSeek Flash, up to 50 answers a day, and says so under each one.
Questions about AI chat
What people ask before they try it.
Yes. Before each answer it searches your library for the closest passages: your notes, video transcripts and uploaded files, plus saved pages on Starter and Plus. To keep an answer to particular items, attach them as sources. It only reads what you have access to.
Yes. Each message is charged from the tokens the model read and wrote. A typical question on the default model uses about 62 credits, so the Free plan's 1,000 monthly credits cover about 16 questions. Stronger models cost more per message.
It can edit the note you have open, and only with your approval: the edit shows as a diff in the note, and nothing changes until you accept it. Deleting a note or a deck always asks first too.
Free includes 8 models, with DeepSeek Flash as the default. Starter and Plus unlock more of the 22 in the catalog, including Claude, GPT and Gemini models. You choose the model for each conversation.
Up to 10 sources per message: notes, videos, files and saved pages from your library, or a new PDF, Word document, text, CSV, Markdown, HTML or image file up to 50 MB.
Yes. Notes the Assistant creates are ordinary notes, so you can download them as Markdown. The conversations themselves cannot be exported.
Your conversations are visible only to you.
Works even better with
AI chat is one part of LunaNotes. These pick up where it leaves off.
Explore what else LunaNotes does
Each one works on the same notes, so what you make in one feeds the rest.
Make it studyable
Bring it in
- Video workspace Your video library, with notes attached
- Timestamped notes Notes that jump back to the moment
- YouTube extension Notes and transcripts inside YouTube
- YouTube courses Work through a playlist in order
- Bookmarks Save any page, find it again
- Meeting notes An AI notetaker for Google Meet
- Podcasts Subscribe, listen, keep the notes
- Files PDFs, documents and images you upload