THE ROAD SO FAR
The Road So Far
Mana's journey so far, one level at a time.
LV 1
First Words
- Started the project and got a local language model replying through llama.cpp.
- Wake-word listening: say her name and she starts paying attention.
- Speech is transcribed locally with whisper.cpp, so nothing leaves the machine.
- Her first voices, Kokoro and Chatterbox, with replies split into chunks so she starts talking sooner.
- A first avatar: a small picture of her in the corner of the screen that reacts while she talks.
- Her personality got its first round of tuning.
Abilities unlocked
- Voice
- Hearing
- Form
LV 2
A Body of Her Own
- A built-in Live2D avatar, in both the launcher and the desktop app.
- She waits for a real pause before replying, instead of cutting in mid-sentence.
- Multilingual speech, with the language detected automatically, and emoji and kaomoji read out as words.
- Self-hosted web access: search, wiki lookups, and reading a page you point her at.
- Local vision, so she can describe what’s on the screen.
- A first-run setup wizard, automatic updates, and Python bundled into the installer.
- Relicensed to Apache 2.0.
Abilities unlocked
- Form
- Voice
- Hearing
- Sight
LV 3
Spellbook & Memory
- Skills: she can propose new ones when she’s idle, or you can create them in a conversation, and nothing is saved until you approve it.
- Memory changes wait for your approval too, and she checks for contradictions before remembering something new.
- Search across past conversations, by keyword and by meaning.
- Deep Research got smarter about spotting gaps in what it found.
- She can choose her own Live2D expression, and her lips move with her voice.
Abilities unlocked
- Spells
- Memory
- Form
LV 4
The Guild Hall
- A plugin store with two tiers: plugins that add abilities, and add-ons that extend existing ones.
- A large refactor of how the pieces fit together, to make the rest of the work easier.
- Work began on a native Windows launcher to replace the Electron one, aiming to cut memory use from over 1 GB to about 500 MB.
- Folio started: a small HTML engine so Mana can show the pages she writes inside her own window.
- The busiest month so far, with over a thousand commits.
Abilities unlocked
- Gear
NOW
Current Quest
- Streaming speech: she starts talking while the rest of her reply is still being written.
- A Doctor panel and tray notifications that warn when something in her setup needs attention.
- A memory browser in Settings, to see and manage what she remembers.
- Diagrams in the pages she writes.
Abilities unlocked
- Voice
- Memory