Boox is now Ullavox: a cleaner UI, with context you control
Boox is now Ullavox. The open-source AI workspace continues under a new name, with the same repository history and release numbering. The current release is v1.7.0.
The rename is also a good moment to show the recent work: a more focused conversation UI, optional response preferences, projects and explicitly saved notes. These features let you carry useful context between chats while choosing what the model receives.
A fictional workspace, using local Gemma 4 12B. Music: Gimme that Groove! by Michael Ramir C., via Mixkit.
Start with the conversation
The interface gives more room to the answer and keeps the composer close at hand. Context controls open when you need them, navigation works on smaller screens, and research activity can collapse while you read.
The model picker shows the Ollama connection supplying the available models. A model reached through Ollama may run locally or use a cloud service, so the model you choose still matters.
Set your preferences once
Optional preferences describe how you want answers written: response length, explanation style, assumed familiarity and working instructions. You can skip setup, return later and change the defaults.
For example, you might prefer short answers with practical examples, then ask for a detailed explanation in a particular message. Each chat has controls for including or excluding your saved preferences and project context.
Give related chats a project
A single idea can spread across several conversations. One chat explores scope, another examines estimates, and a third works through implementation.
Projects group those conversations and provide shared instructions. In the walkthrough, the fictional “Weekend studio” project is planning a small reading app. Its instructions favour practical suggestions and a focused first version.
That gives the model a useful starting point without requiring the same introduction in every chat.
Save the decisions worth carrying forward
The reading app's first version needs three features: a reading list, quick notes and a weekly reading goal. Social features can wait.
Save that decision as a project note and pin it. A fresh chat in the same project can then answer “What did we decide to include?” using that note. Context used shows which preferences, project instructions and saved notes were supplied for the response.
Saving is explicit. Ullavox does not automatically turn every conversation into a memory or semantically search your entire chat history. You can review, edit, select, pin or delete notes. For an ambiguous question, selecting the relevant note makes the input clearer.
Saved notes remain after their source chat is deleted. Manage them separately when a decision changes.
Move from a file to a plan
Ordinary chat sits alongside Research mode for text-file analysis and optional web research. The film uses a small CSV with estimates of five, seven and nine hours. The model identifies the three features and totals them to 21 hours.
Files remain available for follow-up questions until cleared. You can enable web sources when the task needs them and inspect the activity and source links.
Attachments support up to four text, Markdown, CSV, JSON or code files, with 24,000 characters combined. PDF and Office documents, file editing and code execution are not supported. Model calculations and citations still need checking; web-tool calls and conversation context are bounded.
Choose where the model runs
Ullavox runs with Docker and Ollama. Conversations, preferences and project notes persist in a local SQLite database. The release installer downloads the default granite4:3b model on first start; the film demonstrates another local model, Gemma 4 12B.
If you choose a cloud model, the included input and context go to that provider. Web research sends queries and requested URLs to external services. Turning off web sources does not make a cloud model local.
Try Ullavox v1.7.0
For a fresh installation, install Docker and Docker Compose, then run:
curl -fsSL https://github.com/manojisnow/ullavox/releases/download/v1.7.0/docker-compose.release.yml \
-o docker-compose.ullavox.yml
docker compose -f docker-compose.ullavox.yml up -dOpen http://localhost:8080. Allow time and disk space for the first model download.
Already using Boox? Read the migration guide before upgrading. Configuration and storage names changed. Back up the complete data directory and explicitly map your existing database filename and volumes so the new installation uses your existing data.
Ullavox is intended for a trusted, single-user installation. Its APIs have no login protection; published ports bind to localhost by default.
Explore the source, get v1.7.0, or read the technical release notes.
If you try it, I would like feedback on whether the interface makes the next action clear, and whether projects and saved notes reduce the context you need to repeat.