Download
Logos is a universal AI desktop assistant driven by Large Language Models. Four modes share one window: Chat answers and analyzes files; Vision drives the screen through a provider's Computer Use interface with mouse and keyboard; Terminal runs an autonomous command-line agent with shell, file, web and MCP tools; Browser pilots a real Chromium browser via accessibility snapshots.
Bring your own key: OpenRouter, Anthropic, Google, OpenAI, xAI, Hugging Face, or any OpenAI/Anthropic-compatible endpoint such as LM Studio, Ollama or vLLM. Keys stay in the OS keychain, not plain text.
Sessions are journaled, so history and tool calls survive a restart, and a rolling summary keeps long sessions cheap. Terminal actions pass an access policy (ask, auto, full), Agent Skills and MCP servers extend the assistant, and a workspace panel shows context usage, activity, files and git status.
Windows builds (installer or portable) are on GitHub Releases; Linux runs from source. MIT licensed.
Features
- Four modes: Chat, Vision, Terminal, Browser
- Computer Use: sees and controls the screen
- Autonomous terminal agent: shell, files, web, MCP
- Chromium browser automation via accessibility snapshots
- OpenAI, Anthropic, Google, xAI, OpenRouter, Hugging Face, local servers
- API keys stored in the OS keychain
- Session journal, replay and rolling context summary
- Agent Skills (SKILL.md) and MCP servers
- Terminal access policy: ask, auto, full
- Windows installer and portable build
Project Samples
