Skip to content
Umbra

What it does

Umbra Agent works as an always-on orchestrator. Give it a task, and it handles the full loop: understanding your codebase, talking to the model, applying changes, running tests, reading errors, and retrying until the job is done.

Key capabilities:

  • Autonomous Harness Loop — runs your check script, reads failures, sends them back to the model, and iterates until the check passes. No babysitting.
  • Provider-agnostic — works with OpenAI, Anthropic, Mistral, Ollama (free, local), LM Studio (free, local), OpenCode Zen (free cloud), and any OpenAI-compatible endpoint. Switch anytime.
  • Daemon architecture — runs in the background via PM2. Queue tasks and manage sessions efficiently.
  • Persistent memory — one global SQLite database across all your projects. No leftover .sqlite files scattered around. Past solutions are indexed and recalled.
  • Local-first — your code and context stay on your machine. Nothing is sent to third-party services beyond the LLM provider you choose.