Omegon
Terminal-native AI agent harness — for operators who build. Single binary. Ten providers. Zero runtime dependencies.
What is Omegon?
Omegon is a ~19MB Rust binary that provides a complete AI-assisted development environment. It connects to 10 inference providers through native Rust clients (no subprocess shims, no Node.js at runtime), manages persistent project memory across sessions, decomposes work into parallel git worktree children, and tracks design decisions in a knowledge graph — all from your terminal.
It is not a wrapper around curl. It is not an IDE plugin. It is not
a chatbot with a file picker. It is a systems engineering harness that
happens to use LLMs as one of its subsystems.
Three-Axis Routing Model
Omegon organizes inference along three independent axes, giving operators fine-grained control over capability, cost, and context:
| Axis | Values | Controls |
|---|---|---|
| Capability Tier | local → retribution → victory → gloriana | Which model family to use |
| Thinking Level | off → minimal → low → medium → high | Extended reasoning budget |
| Context Class | Squad (128k) → Maniple (272k) → Clan (400k) → Legion (1M) | Context window capacity |
The agent adjusts these axes autonomously based on task complexity, or operators
can override with /model, /think, and /context.
See Three-Axis Model for details.
Key Features
- 10 Inference Providers — Anthropic, OpenAI, Codex, OpenRouter, Groq, xAI, Mistral, Cerebras, HuggingFace, Ollama. Native Rust clients with automatic failover.
- Parallel Execution — Cleave tasks into parallel worktree children with automatic merge, conflict detection, and adversarial review.
- Project Memory — Persistent facts, episodes, semantic recall, and knowledge graphs across sessions. Not a markdown file — a real memory system.
- Design Tree — knowledge graph tracking architecture decisions, dependencies, and implementation status.
- OpenSpec Lifecycle — Spec-driven development: propose, spec, fast-forward, cleave, assess, archive.
- Live Dashboard — Terminal TUI with a wrapped multiline editor, segment rendering with model tags, context and memory telemetry, plus a web dashboard over WebSocket.
- Plugins — Personas with organic memory growth, tones, and memory layers. Not prompt templates — living cognitive profiles.
- Sessions — Auto-save and resume, plus a session log tool and narrative history for orientation across runs.
- Codebase Search — Ranked concept search across code and project knowledge via
codebase_searchandcodebase_index. - Signed Releases — Cosign signatures, CycloneDX SBOM, GitHub Attestations, and installer verification before update.
Get Working Fast
If your goal is a first successful run rather than a deep tour of Omegon internals, start with the provider happy paths:
- Choose Your Path — Claude subscription, API key, or local Ollama
- Claude Subscription Path — fastest “it works” route for many operators
- API Key Path — direct setup for key-based providers
- Local Ollama Path — on-device inference without external API keys
See Installation for all methods, and Quick Start if you want a broader first-session walkthrough.
At a Glance
| Runtime | Single ~19 MB Rust binary, 6 crates |
| Providers | 10 inference + 3 search, native Rust clients |
| Agent tools | 53 structured tools (file ops, memory, design, search, codebase retrieval, and inference control) |
| License | BSL 1.1 (converts to MIT on 2031-03-19) |