Manage a 16-agent AI team on your desktop by stating your goals
Imagine moving away from the burden of pushing complex projects alone to a scene where multiple AIs divide roles on your computer. Orkas is a desktop app that takes your goals, breaks them down into concrete tasks, and executes them by combining its built-in 16 specialized agents with the coding tools you already use. All data stays on your computer, and finished results return for review.
Collaboration structure led by a Commander
The core of Orkas is an orchestrator called the 'Commander'. The Commander analyzes the goal you provide and decides which agent should handle which task. It handles simple tasks directly and delegates specialized parts to built-in agents like VideoStudio or DeepResearcher. It also runs external CLI agents such as Claude Code, Codex, OpenCode, OpenClaw, and Hermes-Agent as local sessions, making them work together under a single plan. This allows you to handle bug fixes, code reviews, and release note writing in a single flow within one chat window.
Sustainable project management beyond one-off answers
Orkas is designed to manage tasks that need to progress over time, going beyond a single question-and-answer exchange. Projects store goals, rules, memory, and files, so new conversations pick up where the previous one left off. Todo lists are assigned to both humans and agents, and if you enable the 'auto-proceed' feature, the team selects and completes the next task on its own before requesting review. You can also handle repetitive work through scheduled automation at time, day, week, or month intervals. Starting from v1.9, a feature was added that automatically fills in missed executions from the past 12 hours when the computer is turned back on after being off.
16 built-in experts and custom agents
The app comes with 16 specialized agents by default, each with their own skills, memory, and tools. They cover a wide range of areas, from video, audio, and image production to market research, SEO audits, stock analysis, content writing, PPT creation, UI design, code development, office document processing, and e-commerce operations. For example, VideoStudio can analyze reference videos to reproduce similar structures in your own content, while ECommerceAgents support the entire e-commerce process from product selection to review and sales data analysis.
Local-first design and freedom in model selection
Orkas follows a local-first principle by default. Chat logs, files, knowledge bases, and custom agents are all stored on your disk, and cloud synchronization is optional. Model selection is also flexible. You can use the built-in Orkas model or connect your own API keys for Claude, OpenAI, Gemini, DeepSeek, Kimi, GLM, Qwen, MiniMax, or Doubao. It also supports local model endpoints like Ollama or LM Studio, allowing you to run completely offline with local models without a network connection. When using your own keys, model calls are made directly from your computer to the provider, and Orkas does not add any markup.
Things to check before running and limitations
Orkas is not a single chatbot or code library, but a desktop app that configures an agent team to manage sustainable tasks. Therefore, it may not be suitable if you only want simple question-and-answer interactions or are looking for a code framework to embed in your own app. The packaged app supports macOS 12 and above, and Windows 10 and above, while Linux must be run from source. Running from source requires a Linux x64/arm64 environment with Node 22.12 or higher and glibc 2.34 or higher, and musl-based distributions like Alpine are not supported. The packaged app provides 500 credits for free, but cloud synchronization, shared projects, and web app publishing features are only available in paid plans or the packaged app. Source builds have no account hierarchy and enter the app directly, but do not support these extended features. Automation and auto-proceed only work when the computer is on and Orkas is running, so you should be aware that this is a personal desktop environment, not a 24-hour server environment.
