Context-Aware
Maintains deep context across your entire workflow, adapting to patterns as you work.
Features
Maintains deep context across your entire workflow, adapting to patterns as you work.
Integrates directly with your existing tools — no wrappers, no friction, just execution.
Learns your codebase structure, preferences, and decisions across every session.
Drop Umbra into any project and it starts working. No YAML, no schemas to define.
Quickstart
No config files. No setup wizard. Pick your platform, run two commands, and Umbra is reading your project.
Read the full documentationwinget install Umbra.Agent umbra curl -fsSL https://umbra.expert/install.sh | sh umbra brew install umbra umbra Changelog
The brief
Umbra is an AI agent that runs locally in your terminal. It reads your project, understands your codebase, and works silently in the background — running commands, editing files, and surfacing context without breaking your flow.
Windows (via winget), macOS (via Homebrew), and Linux (via the install script). All three platforms are first-class — no feature gaps between them.
Only the context needed to answer your query is sent to the model. Your full codebase stays local. You can also configure Umbra to run against a local model endpoint for fully offline operation.
Yes. Point Umbra at any OpenAI-compatible local server (Ollama, LM Studio, etc.) and it works entirely offline. Set the endpoint in your config or via environment variable.
Umbra is a background agent, not a chat interface. It watches your project state, runs tools autonomously, and acts when needed — instead of waiting for you to ask. Think of it as the shadow behind your workflow, not another tab to switch to.
The core agent is free and open source. You bring your own model API key. A hosted tier with team features and cloud sync is planned for later in 2026.