other / agent
Lumina
A desktop AI agent app written in Python with persistent memory, swappable personas, voice, coding tools and local or cloud model backends.
At a glance
- Purpose
- A desktop AI agent app written in Python with persistent memory, swappable personas, voice, coding tools and local or cloud model backends.
- Runs in
- Local agent with shell
- Operating systems
- macOS, Linux, Windows
- Needs
- Runtime: Python; Needs internet: No; Needs accounts or API keys beyond your agent: No
- Files and data
- Reads files: Yes; Writes files: Yes; Sends data to other services: Yes
- Tool type
- Agent
- Delivery
- From GitHub
- Version
- Latest on the main branch
- Access
- Code on GitHub
- Compatibility
- According to the README: Linux, macOS, Windows or a VM, with macOS and Windows still in the testing phase. Local backends: llama.cpp (primary, recommended), LM Studio, Ollama and vLLM. Self-hosted gateway: OmniRoute. Cloud backends: Anthropic, Google, OpenAI, Moonshot, Alibaba, OpenRouter, DeepSeek, Groq and any OpenAI-compatible endpoint through a Custom slot. Text-to-speech: Kokoro by default, plus Voicebox, Chatterbox, Supertonic and ElevenLabs. Speech input: Whisper. The Browser Companion works with Chrome. Remote channels: Telegram and Discord.
- External costs
- The README says local operation needs no account and lists cloud model backends and ElevenLabs text-to-speech. The README describes image generation with an estimate followed by owner confirmation before spending. The README gives no provider prices. Your coding agent or model provider may charge for usage.
- Permissions
- The README's setup installs Python dependencies with pip, installs Chromium with Playwright and copies config.example.py to config.py. The README describes tools for file reading and editing, Python and shell execution, Git, worktrees, test execution, web fetching and Playwright browser automation. The README also describes Browser Companion reading and owner-granted navigation in your logged-in Chrome profile. The README says an owner flag enables the owner toolset while operation-specific grants, routing checks and the Emergency Interlock still apply. The README says non-owner sessions receive tools through named profiles. The README describes Telegram access locked to your chat ID and a restricted Discord profile. The README labels subagents and background or scheduled tasks optional and off by default. The README describes a local-UI Emergency Interlock and confirmation before image-generation spending. The README points to docs/security.md for the full authority model.
- Usage terms
- Apache License 2.0, as stated in the repository. See its LICENSE file for the exact terms.
- Interface
- desktop
Code on GitHub
Bino5150/lumina
Apache-2.0A full featured, powerful, and efficient AI Agentic Harness/Desktop Agent app designed from the ground up with local inference on consumer hardware in mind. It evolves, grows, and gets smarter as you go. And it REMEMBERS...
Public GitHub data fetched · 0 days old
Skill folders in this repository (1)
Folders that contain a SKILL.md file, from GitHub's file list on .
What it does
Lumina is a Python desktop AI agent app with a PySide6 interface. The README describes local inference on consumer hardware and labels the project active development with public beta testing.The README lists persistent memory, JSON personas, speech input, text-to-speech, vision, image generation with a spending confirmation, and a Browser Companion. The README lists named tool profiles, Git, worktrees, structured test execution, coding checkpoints and diff review.
The README describes backend switching in Settings, with llama.cpp, LM Studio, Ollama and vLLM as local backends, OmniRoute as a self-hosted gateway, and cloud providers. The README also describes Telegram, Discord and a CLI mode.
Who it is for
People who are comfortable with a terminal and want a desktop AI agent with persistent memory and personas.Example tasks
Copy config.example.py to config.py and point it at your llama.cpp server and model.Start Lumina with the lumina command.
Switch the model backend from the Settings dropdown.
Create a persona as a JSON file in the personas folder.
Use the coding tools for Git, worktrees and test execution on a project.
Getting started
According to the README, you clone the repository, then run pip install -r requirements.txt and playwright install chromium. You copy config.example.py to config.py and point it at your llama.cpp server and preferred model, then start Lumina with lumina, which the README says is an alias for start_lumina.sh. The README points to docs/getting-started.md for requirements and first-run notes. The README has the exact commands.Limitations
The README labels Lumina active development with public beta testing and lists Linux, macOS, Windows or a VM, with macOS and Windows still in testing. The README says users need a GPU and terminal familiarity and describes development on a 4GB Nvidia Quadro T1000. The README's setup installs Python dependencies and Playwright Chromium. The README says cloned persona voices require Chatterbox Turbo or Voicebox.The README labels Oracle coming soon. The README says custom endpoints are supported through llama.cpp, LM Studio, Ollama, vLLM, OmniRoute and Custom; other listed backends use provider-owned endpoints. The README labels subagents and background tasks optional and off by default. The README lists email, wake-word listening, iMessage, a Browser Companion pairing UI, smart-home management, Klipper management and home-security monitoring as not yet merged into the public release. The repository can change at any time; check its README before you install.
Data practices
The README describes local inference, memory, voice synthesis and speech recognition, while also listing cloud model backends and ElevenLabs. The README does not detail the payload sent to each cloud backend; that depends on the selected backend. The README includes a statement that session data is not sent back as telemetry.The README says old sessions are preserved in SQLite and credentials live in a dedicated permission-locked file excluded from version control and Memory Backup archives. The README describes browser fetching, Playwright automation and a Browser Companion using your signed-in Chrome profile. The README points to docs/security.md for the full security model.