Orkas is an open-source, local-first multi-agent AI desktop client, primarily aimed at independent developers, one-person businesses, and small teams. Users can directly command a team of AI agents through chat to deliver research results, documents, slides, videos, websites, and other content.
In terms of use, Orkas organizes research, writing, office work, operations, and other tasks into task lines completed collaboratively by multiple expert agents. Typical workflows include: choosing a market with evidence rather than intuition (DeepResearcher → ContentWriter → OfficeWorker), turning an approved product brief into articles, images, and videos (ContentWriter → ImageStudio → VideoStudio), drafting reports and review slides after verifying the numbers (OfficeWorker → PptMaker → ContentWriter), and automatically generating team weekly reports and refreshing project notes according to plan (Commander → OfficeWorker → ContentWriter).
Main capabilities include: frontier models and frontier agents; the ability to read PDFs and provide citations, fix bugs and submit PRs, analyze data, create apps or videos, organize inboxes and calendars, and connect external tools; and turning repetitive work into reusable agents and skills. The official emphasis is on three foundations: professional and reliable (clear ownership of responsibility from planning to delivery, retaining reusable work knowledge after each run, turning goals into executable plans and tracking verification), safe and controllable (workspace data stays local by default, credentials are protected locally, the scope of what each agent can read, invoke, and change can be defined, and confirmation is paused before high-risk operations), and open and extensible (the implementation can be inspected, self-hosting is optional, you can choose to host official models or connect your own model provider, and you can also invoke supported local agents in the same workspace and conversation, and connect data sources, tools, and community capabilities through supported integrations).
In terms of usage, you can start for free after installing the desktop app: you can directly use Orkas-hosted official models, or connect your own model provider via OAuth or API Key, including OpenAI, Claude, Gemini, DeepSeek, Kimi, and GLM; calls from your own provider are sent directly to that provider. The official download page provides macOS and Windows versions; the open-source repository also supports Linux, which requires building it yourself. The interface offers 18 language options and can be switched at any time.
Use cases include: market research, report and slide creation, PDF reading with citations, fixing bugs and submitting PRs, data analysis, creating apps or videos, organizing your inbox and calendar, connecting external tools, and turning repetitive work into reusable agents and skills.
Regarding limits and billing: the desktop app is free and includes 500 complimentary credits for hosted official models; usage with your own provider is billed by that provider and does not consume credits. Credits only meter Orkas hosted model services. Paid plans add monthly credits, multi-device sync, and cloud storage; Lite includes 5,000 credits per month, Pro includes 10,000 credits per month, and Max includes 50,000 credits per month. Lifetime credit packs can also be purchased as a one-time purchase (monthly credits are consumed first). Consumption follows each provider's official pricing: approximately 100-300 credits per million input tokens, 2-6 credits per million cached input tokens, and 200-600 credits per million output tokens. Remote control and team collaboration are still planned. By default, configuration, history, the knowledge base, and generated files remain on the local machine; bring-your-own keys are stored locally, and requests are sent directly to the provider; hosted models and optional cloud features use Orkas services only for the selected features. The desktop client is open source under the MIT License at github.com/Orkas-AI/Orkas, so you can read the source code, build it yourself, and audit network requests. The open-source desktop version can be downloaded and self-hosted for free, without an Orkas account; the official distribution adds optional hosted services (hosted models, multi-device sync, cloud storage), which require an Orkas account. Note that Orkas is a desktop client rather than a server; self-hosting means running the app on your own machine, not deploying a server for a team.
