Omegon

Terminal-native AI agent harness

Single binary. Native Rust. Real memory. Parallel worktrees.
For operators who build.

Omegon is the agent harness for engineers who need more than a polite chatbot: persistent project memory, honest provider identity, design and spec lifecycles, and enough systems discipline to survive real work.

curl -fsSL https://omegon.styrene.dev/install.sh | sh

installs v0.15.10  ·  other versions & options →

or brew tap styrene-lab/tap && brew install omegon

Honest engine identity

Omegon shows the concrete runtime backend — Anthropic/Claude, OpenAI API, OpenAI/Codex, Ollama — instead of hiding routing behind mushy branding.

Persistent memory

Project knowledge survives across sessions. Durable facts, episodic recall, and typed memory sections. Not another markdown file pretending to be state.

Parallel execution

Decompose substantial work into parallel git worktree children with scoped execution, conflict detection, and merge back into the parent branch.

Design + spec lifecycle

Track design exploration in the design tree, then drive implementation with OpenSpec scenarios. The plan is part of the system, not a side document.

Real TUI instrumentation

Wrapped multiline editor, tool cards, memory telemetry, engine status, and limit metadata in the footer. It behaves like an instrument panel because that is what it is.

Single binary

Native Rust harness, signed releases, no Node.js runtime requirement, and direct provider integrations instead of subprocess cosplay.

~19 MBsingle binary
10inference providers
53agent tools
0runtime Node deps