Skip to content
LloomaLloomaLlooma

Early preview, not for production

Every model you use, in an app you host

Point it at Ollama on the machine under your desk, or at a provider you already pay for. One place for the conversations, the personas and the keys, on a server that belongs to you.

Install itSee what it doesSource on GitHub

Forked fromHollama by fmaclen

The home screen of the phone interface, with a persona ready to talk

Two minutes

Eight lines and a port

Section titled “Eight lines and a port”

No database to provision and no build step. Bring your own models, whether that is Ollama on the same machine or an API key you already have.

Read the installation guide

docker-compose.yml
services:
  llooma:
    image: ghcr.io/cedhuf/llooma:latest
    restart: unless-stopped
    ports:
      - '4173:4173'
    environment:
      - VITE_ALLOWED_HOSTS=localhost
A conversation on a desktop, over a wallpaper, with every surface translucent
Your own picture behind the app, and the translucency of every surface following it.

Library

Characters, procedures and notes, kept

A persona carries its own prompt, model, avatar and voice, and answers a single turn when you mention it with @. A playbook is a procedure written once and switched on where it is needed. Knowledge is the text you were retyping into every conversation.

All three install from a store over the network, so adding one is a pull request rather than a release.

How personas work

The library, with three personas

Conversations

Long ones keep fitting

When a conversation outgrows the context window, /compact replaces what was said with a structured summary. Nothing is deleted and one click puts it all back.

The reply is written on the server, so reloading the page, walking away or letting the phone sleep no longer costs you the answer.

How compaction works

A conversation, with the composer at the foot

On a phone

A second interface, not a smaller one

One column, a tab bar where a thumb already is, and the microphone set apart because talking is what it is for. Installable as a PWA, and offered to phones only.

Speaking is a loop rather than a button: it stops listening once the room goes quiet, answers out loud, and listens again. It works with the phone face down on a table.

How the voice mode works

Talking to a persona on the phone interface

Appearance

Six themes, each with a light and a dark ramp

Section titled “Six themes, each with a light and a dark ramp”

Pick a palette without committing to a mode, or follow the system without committing to a palette. Try them here, on a conversation that is going nowhere.

Theme style
Light or dark
LloomaNew chatSelf-hosting everythingReading a sourdough starterChoosing a backup strategy

How do I check a backup actually restores?

Assistant

Restore it somewhere else, on a schedule, and look at the result. A backup nobody has read is a hypothesis.

Write literally anythingRun

Built on

Section titled “Built on”

Llooma is a small amount of glue around a lot of other people’s work.

Counted by a self-hosted Umami, which knows this page was opened and not by whom.