Local AI assistant

Lyra

One AI assistant on the computer, shared by Nebula, Nova and Horizon. Questions are answered here, and go to a cloud model only if a key is added.

Download

Local and privacy-focused

Memories, keys and chat history stay on the computer, and no account is needed. Cloud models are optional, and Lyra never sends them its memories.

The computer answers; what tries to leave turns back at its edge
>_lyra$lyraLyra is Focused. Fast, Qwen 3.5 4B.>Summarize what I wrote today.Three notes on the Sprout launch, a list of twentyplants and two open questions for Monday.>Qwen 3.5 4B, on this computerThis computerCloud modelOptional, with a keySent outside this computer0 B
The computer answers; what tries to leave turns back at its edge
>_lyra

$ lyra

Lyra is Focused. Fast, Qwen 3.5 4B.

> Summarize what I wrote today.

Three notes on the Sprout launch, a list of twenty plants and two open questions for Monday.

Qwen 3.5 4B, on this computer

Sent outside this computer

Memories, keys and history stay here.

0 B

The backbone of Cosmic

Nebula, Nova and Horizon connect on their own and share the same models, downloaded once. Requests wait their turn, and Status shows who is asking.

Nebula, Nova and Horizon ask Lyra in turn; Status says who is asking
LyraChatStatusModelsMemoryAppsKeysSettingsStatusFocusedFocused since 10:12Qwen 3.5 4B loaded. On AC power, it sleeps after 60 minutes without arequest.Sleep nowRequestsNowModelsFast and NormalQwen 3.5 4B, installedThinkingQwen 3.5 9B, 2.6 of 6.6 GB downloadedAnswering Nova in Fast.NebulaAnsweredNovaBeing answeredHorizonAsks in Fast
Nebula, Nova and Horizon ask Lyra in turn; Status says who is asking
NebulaAnswering
NovaWaiting its turn
HorizonAnswered

Focused since 10:12

Qwen 3.5 4B loaded, shared by every app.

Requests

Answering Nebula in Normal.

1 waiting, from Nova.

Your personal assistant

Fast, Normal or Thinking for each question, and a memory of what it is asked to remember. Every memory can be read, edited or removed.

One chat in three modes: a quick answer, a memory saved, a question thought through
LyraChatStatusModelsMemoryAppsKeysSettingsChatFocusedNew chatHistoryFastNormalThinkingQwen 3.5 4BWhich plants should the Sprout library start with?LyraStart with the forgiving ones: basil, mint, tomato, lettuce andpothos.Remember that Maya designs the Sprout screens.LyraSaved to memory, in the Sprout garden app Space.Memory: Sprout garden appMessage LyraFast0.2 sto the first words, with themodel loaded.NormalAbout 3 sfor a fuller answer, on thesame model.ThinkingUp to 90 son a larger model; past that,it answers without thinking.
One chat in three modes: a quick answer, a memory saved, a question thought through

Remember that Maya designs the Sprout screens.

Lyra

Saved to memory, in the Sprout garden app Space.

Memory: Sprout garden appNormal

Fast

0.2 s

to the first words, with the model loaded.

Normal

About 3 s

for a fuller answer, on the same model.

Thinking

Up to 90 s

on a larger model; past that, it answers without thinking.

Times measured on a laptop with an RTX 4070. Thinking needs 16 GB of RAM.

Get Lyra

Windows 10 and 11, 64-bit. It installs for the current user.

>_After installing

$ lyra setup

LightQwen 3.5 4B and the search model, about 3.7 GB

Completeadds Qwen 3.5 9B, about 10.3 GB, with 16 GB of RAM

Ollama and the models download only after a yes.

Free disk needed: the models plus 4 GB.

The Beta installer is not signed yet, so Windows may warn before running it.