Orkas: Coordinate AI Agent Teams from a Desktop App

Overview

Orkas is a desktop workspace where a Commander agent interprets a goal, plans the work, and coordinates specialist agents. It addresses the gap between single-assistant chat and code-first orchestration frameworks: users can delegate multi-step tasks through a graphical app without building an agent system themselves.

The app runs on the user's machine and can connect agents to different model providers, local model endpoints, and supported coding or personal-agent CLIs. It is a fit for people who want several agents to produce connected outputs, such as research followed by a report or slide deck, while keeping conversations, keys, and files local.

Key Features

• Commander plans tasks and delegates work to specialist agents, including parallel or sequential work.
• Nine specialist agents are included, with additional agents available through a marketplace.
• Connects to multiple model providers or OpenAI-compatible local endpoints, with provider choice configurable per agent.
• Runs supported external CLI agents as local subprocesses.
• Stores conversations, files, API keys, knowledge bases, and custom agents on the local machine.
• Agents maintain private skills and memory, and can reflect on task outcomes to improve future work.
• Provides local document retrieval using full-text search and on-device embeddings.
• Offers desktop builds for macOS and Windows, with Linux use from source.

Use Cases

• Researchers can delegate source gathering and verification, then hand findings to a writing agent for a cited report.
• Teams preparing presentations can turn a topic or document into a deck after research and drafting steps.
• Developers can use the Commander to break down a product specification and delegate repository-aware coding tasks to agents or supported coding CLIs.
• People handling sensitive documents can analyze files and search a local knowledge base without sending the documents to Orkas servers.
• Individuals who automate recurring work can coordinate agents to prepare reports or process files, provided they have configured suitable model access.

Getting Started

For a source build, install Node.js 22.12 or later and run:

git clone https://github.com/Orkas-AI/Orkas.git
cd Orkas
./run.sh

On Windows, use run.cmd instead. See the README for setup details, platform requirements, and packaged downloads.

Alternatives

• maka: Maka spans desktop, terminal, and evaluation workflows, with an append-only event log for reviewing agent actions.
• orca: Orca focuses on coordinating coding agents in isolated Git worktrees, not broader research and document workflows.
• OpenHuman: OpenHuman is a general-purpose agent harness with browser, terminal, and library interfaces, rather than a desktop-first multi-agent chat.

Considerations

• Orkas orchestrates models and tools; it is not itself a language model. Most use requires users to configure their own model keys or a local model endpoint, and model-provider usage may incur separate costs.
• Although data and keys are stored locally, hosted model calls still send prompts to the selected provider.
• The repository documents packaged installers for macOS and Windows. Linux currently runs from source and requires glibc 2.34 or later; musl-based distributions such as Alpine are not supported.
• A source setup needs Node.js 22.12 or later. Python 3 and a C/C++ build toolchain may be needed if a native dependency must be built locally.
• The app coordinates multiple agents and external CLIs, so setup and task behavior depend on the configured providers, tools, and agents.