Orkas provides an open-source, local-first desktop AI client enabling users to orchestrate a team of AI agents for complex tasks. Its core innovation is a "Commander" LLM that dispatches specialist agents, facilitating parallel or sequential execution. Agents self-evolve via reflection and skill crystallization, offering a powerful, privacy-centric AI workforce managed locally, supporting diverse LLM providers without vendor lock-in.

How It Works

The system features a Commander LLM that decomposes goals and coordinates specialist agents. "Visibility slicing" manages context, ensuring agents only see relevant dialogue segments for optimized token usage and privacy. All dispatches use a single scheduling primitive, with collaborative progress tracked in plan.md. Agents improve through post-task reflection, updating their meta/ directory and crystallizing reusable private skills via skill_manage.

Quick Start & Requirements

• Installation: Production installers available for macOS and Windows via orkas.ai. Run from source using Node.js 20+, Python 3, and bash.
• Setup: Clone repo, run ./run.sh or run.cmd (auto-installs dependencies, downloads ~95MB embedding model). Configure AI Providers in Settings with API keys/OAuth.
• Docs: Core design details in CLAUDE.md.

Highlighted Details

• Local-First: All data, keys, and agents remain on the user's disk; model calls route directly to providers.
• BYO LLM Keys: Supports Claude, OpenAI, Gemini, DeepSeek, Kimi, GLM, Qwen, MiniMax, Doubao, enabling flexible provider mixing.
• Agent Self-Evolution: Agents autonomously improve via reflection and skill crystallization.
• CLI Agent Integration: Orchestrates external CLI coding agents (Claude Code, Codex, etc.) and tools like HyperFrames.

Maintenance & Community

The provided README does not detail specific maintenance contributors, sponsorships, or community channels.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Permissive MIT license supports commercial use. Local-first design enhances data privacy.

Limitations & Caveats

Orkas is not a standalone chatbot, code library, or hosted SaaS. Users must provide LLM API keys or configure local models. Network connectivity is needed for model inference unless a local model is used.