Open-source AI agent orchestration,
without the orchestration code.
Most open-source orchestration means writing a program: agents, tasks, graphs, retries. Orkas is the finished app instead. A Commander agent turns your goal into a plan, dispatches specialist sub-agents in parallel or in sequence, and brings the results back in one thread — running on your own machine, with the source on GitHub.
Orchestration you can read, run, and direct by chat
People searching for open-source AI agent orchestration usually hit the same wall: the open options are libraries, and the products that just work are closed. Orkas is both open and finished.
Read how it orchestrates
The desktop client is open source under the MIT license, so the planning and dispatch logic is something you can inspect rather than infer.
- open source
- MIT
No graphs, no glue code
Describe the outcome you want in plain language. The Commander decides the steps, the order, and which specialist handles each one.
- chat-directed
- no code
Runs on your machine
Orchestration happens locally on macOS and Windows. Chats, agent configs, generated files, and API keys stay on your machine by default.
- local-first
- private
Orchestrate your coding agents
One Commander can drive Claude Code, Codex, and OpenClaw alongside its own specialists, so you stop switching between terminals.
- Claude Code
- Codex
From one sentence to a coordinated team
Describe the goal
Say what you want in chat. No task definitions, no YAML, no graph to draw first.
The Commander plans
A lead agent breaks the goal into steps and picks which specialist should own each one.
Specialists run
Sub-agents work in parallel or in sequence, calling skills to do research, code, documents, and data.
Results come back to you
Everything lands in one thread, on your machine, where you can redirect the work mid-flight.
Choose Orkas when you want the orchestration, not the plumbing
Open and inspectable
Read how planning and dispatch really work instead of trusting a hosted black box.
Ready on day one
Install the app and direct a team by chat, with no framework to learn or maintain.
Model choice, no lock-in
Use optional managed official models or connect your own provider, and switch anytime.
Open-source AI agent orchestration questions
What is open-source AI agent orchestration?
AI agent orchestration is coordinating several AI agents so they work toward one goal: planning the steps, deciding what runs in parallel or in sequence, passing results between agents, and recovering when a step fails. It is open source when you can read and run that coordination logic yourself instead of trusting a hosted service. Orkas does this from a single chat, and its desktop client is on GitHub under the MIT license.
Can I orchestrate AI agents without writing orchestration code?
Yes. You describe the goal in chat and a Commander agent turns it into a plan, dispatches specialist sub-agents in parallel or in sequence, and brings the results back in one thread. There are no graphs to draw, no task queues to wire, and no Python glue to maintain.
How is Orkas different from open-source agent frameworks like CrewAI or LangGraph?
CrewAI and LangGraph are code-first libraries: you write and maintain the program that defines the agents, tasks, and control flow, then run it inside your own service. Orkas is a finished desktop app you talk to, so the orchestration is already built and you direct it by chat. A framework is still the better fit when you need programmatic control or want orchestration embedded inside your own product.
Can Orkas orchestrate coding agents like Claude Code and Codex?
Yes. One Commander can drive external coding agents such as Claude Code, Codex, and OpenClaw alongside its own specialist sub-agents, so you direct them from one conversation instead of switching between terminals.
Which models can I use for agent orchestration in Orkas?
You can use optional Orkas-managed official models, or connect your own provider through OAuth or an API key, including OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Kimi, and GLM. With your own provider, traffic goes directly to it and is not proxied through Orkas.
Try open-source agent orchestration
Free, local-first, and open source. Use optional managed official models or connect your own provider, and direct your AI team by chat.