AI assistant guide

What is it?

Fast Memo's AI assistant is Qwen3.5, an open-source language model that runs directly on your phone. You download it once; from then on it works on the device, even offline, and your notes never leave it.

• Private

  Your notes are processed on the phone. No account, no API key, no server.

• Works offline

  Once the model is downloaded, it works in airplane mode too.

• Free

  No subscription and no limits: your phone does the work.

Which model?

There are three Qwen3.5 models. The bigger the model, the better the results, and the slower and heavier it gets. The app only lets you pick the ones your phone has enough RAM for, and marks the one it recommends.

• Qwen3.5 0.8B

  Titles, categories, dictation

  Download

  ~530 MB

  RAM

  3 GB+

• Qwen3.5 2B

  Every action, everyday use

  Download

  ~1.3 GB

  RAM

  4 GB+

• Qwen3.5 4B

  Best quality, slower

  Download

  ~2.7 GB

  RAM

  6 GB+

How to set it up

1. 1

   Open the AI assistant settings

   In the Fast Memo app on Android, go to “Settings” → “Advanced” → “AI Assistant”.

2. 2

   Pick a model

   Tap one of the cards under “Model”. The models your phone can't run say “Not enough RAM on this device”; the one that fits best is marked “Recommended”.

3. 3

   Download it

   Tap “Download model”. The file comes from Hugging Face and weighs from ~530 MB to ~2.7 GB, so use Wi‑Fi if you can. When the download ends, the assistant turns on by itself (“Enabled”). You keep one model on the phone: downloading another one replaces it, and “Delete model” frees the space.

4. 4

   Use it in a note

   Open a note and tap the sparkle button in the bar at the bottom of the editor, then pick an action. You see the text “Before” and “After”, and nothing changes until you tap “Apply”; “Cancel” stops the AI and keeps your note as it was.

What it can do

The menu shows the actions that make sense for the note you're in. The ones that need a bigger model are greyed out with “Minimum 2B”.

• Generate title

  Writes a short title, 2 to 5 words, from what the note says, in the app's language.

  Every note type

• Suggest category

  Picks the best fit among your categories and moves the note there. It never invents new ones.

  Every note type

• Summarize

  Writes 2 or 3 sentences and puts them at the top, under “Summary”; the note stays as it is below. Long notes are summarized in parts.

  Text notesMinimum 2B

• Fix grammar

  Fixes spelling and grammar line by line, in the note's own language. Code, tables and images are left alone.

  Text notesMinimum 2B

• Make shorter

  Tightens the long paragraphs and leaves headings and list items as they are.

  Text notesMinimum 2B

• Translate

  Translates the note into one of 16 languages; the app's language comes first in the list.

  Text notesMinimum 2B

• Suggest items

  Adds three new items to a checklist, based on its title and on what's already in it.

  To-do listsMinimum 2B

Dictation, cleaned up

With “Voice recognition” on, dictate in a text note or a to-do list, then tap “Clean up with AI” before “Insert”.

• In a text note it adds punctuation and capitals, fixes misheard words, drops the fillers and starts a new paragraph when the topic changes.
• In a to-do list it turns what you said into one item per task.

It works with every model; with the 0.8B it only adds punctuation and capitals.

Tips

• Not sure which model to pick? Take the one the app recommends. The 4B writes better, but it's two to three times slower than the 2B.
• The first action after a pause takes a few seconds: the model loads when you open the AI menu and is released 3 minutes after you last used it, or when you leave the app.
• On very long notes some actions say “Note too long for this model”: a smaller model is faster, or work on a shorter part.
• “Fix grammar” and “Make shorter” rewrite one line at a time: the lines they change lose bold, links and other inline formatting.
• Used the Qwen 2.5 models before? Fast Memo no longer uses them: download a new model, or free the space with “Delete old models” on the same screen.