Orkas Multi-Agent Desktop App: A Commander and 16 Specialists Running on Your Own Machine.

The Orkas multi-agent desktop app puts a team on one computer instead of a single chatbot in a browser tab, running locally first. Tell its Commander what you need: it plans the work, splits it into to-dos, and hands each part to one of 16 built-in specialist agents, or to the coding and personal agents you already use. The Orkas AI GitHub repo, Orkas-AI/Orkas, had 6,105 stars at the time of checking (1 Nov 2026), with 369 forks, 146 watchers and 820 commits under an MIT licence. The repository was created on 29 April 2026. It has been building quietly since April; this explainer uses today's numbers, not a launch story.

Orkas is built to keep a project moving, not just answer one prompt. A project holds the goal, rules, memory and files, so a new conversation picks up where the last stopped, and finished work returns for review. Data stays on the computer by default. Stars measure attention, not quality.

How the Commander and 16 specialists divide the work

You start with a goal. The Commander breaks it into steps, picks the right agents, and handles general work itself when no specialist fits better. Delegation runs through three mechanisms the README names: dispatch_to, hand_off_to and run_worker, so one plan can span several agents while the Commander keeps track. One chat can fix a bug, review the change and draft the release note, each part handled by the agent suited to it.

Illustration created for INORYVA — a generic desktop workspace with several agent task panels, not a screenshot of the Orkas app.

The 16 specialists ship with the app, each with its own skills, memory and tools: DeepResearcher, SeoGeoAgent and StockAnalyser for research; ContentWriter, PptMaker, ImageStudio, VideoStudio, VoiceStudio and UIDesigner for content and media; ProductDeveloper and OfficeWorker for building and office work; and ECommerceResearcher, ECommerceWriter, ECommerceReviewer, ECommerceAnalyzer and ECommerceOperator for e-commerce. The README credits them with research reports, SEO audits, decks, videos, websites, Office files and product listings. SeoGeoAgent audits a URL for technical issues, content quality, Core Web Vitals and AI-answer citability, a narrower job than the claude-seo Claude Code SEO plugin, which runs a full SEO audit inside Claude Code. The README also points to a marketplace of more than 30 agents, and says the Commander can build a custom agent from a described role.

Work carries across days. Every project has a to-do list for people and agents; with Auto-advance on, the team takes the next to-do itself. Scheduled automations repeat routine work hourly, daily, weekly or monthly. Connectors cover Gmail, Calendar, Sheets and Drive, Notion, Slack, Shopify, Stripe and more; Orkas asks by default before sending, posting or deleting anything. A local Library indexes documents so answers can cite source passages, and durable notes keep preferences and project facts separate. The app can also build a small tool from a description and run it locally.

What it runs on, and which coding agents it works with

Packaged installers cover macOS 12 and later, on Apple Silicon and Intel, and Windows 10 and later. Linux runs from source today, for glibc 2.34 and later on x64 and arm64, with no installer yet; Alpine and other musl-based distributions are not supported. Running from source needs Node 22.12 or later. The packaged app asks the user to sign in on first launch; the source build has no account layer and boots straight into the app.

The Orkas vs Claude Code Codex question has a direct answer in the README: not either-or. Orkas runs installed command-line agents as local sessions next to its specialists: Claude Code, Codex and OpenCode for coding, OpenClaw and Hermes-Agent as personal agents. The Commander hands each its best part and keeps one plan across them. The README draws the same line for frameworks: LangChain and CrewAI are code-first tools for building agent software, while Orkas is a desktop app directed through chat. Developers tracking GitHub agent tooling may know the Impeccable GitHub skill, which gives coding agents a design vocabulary; Orkas is on the same shelf, but its job is coordination across a team.

Illustration created for INORYVA — files and notes stored on a single computer, not a screenshot of the Orkas app.

Local-first is drawn carefully in the README. Conversations, files, knowledge bases and custom agents stay on disk by default, and cloud sync is optional in the packaged app only. With the user's own keys, calls go straight from the machine to the provider; keys are encrypted locally and excluded from cloud sync. The built-in Orkas model runs through Orkas's managed service. Orkas is not itself a model: it orchestrates agents and calls the models chosen. The source build can run fully offline against a local endpoint; the packaged app needs sign-in, and cloud models need a connection. Because Orkas runs on the computer, automations need it awake and running; work does not continue while the machine is off.

Model choice stays open. The Commander and specialists share one setting: the built-in Orkas model, the user's own keys (Claude, OpenAI, Gemini, DeepSeek, Kimi, GLM, Qwen, MiniMax, Doubao) or a local endpoint, including Ollama or LM Studio. External agents keep their own model settings.

What is free, and what the paid plans add

The Orkas free credits and pricing position is short. The app is MIT licensed and free to use, and the packaged app includes 500 free credits for the built-in Orkas model, so no API key is needed to start. After those credits, the FAQ gives three routes: buy credits, choose a paid plan, or add the user's own provider keys, with provider charges billed separately. Paid plans start from $11.99/month; this article uses only the FAQ wording: paid plans from $11.99/month add monthly credits, cloud sync, more shared projects and app publishing. With the user's own keys, Orkas adds no markup.

Free and packaged are overlapping, different offers. Local apps built inside Orkas are free, while publishing a web app to a public link needs a paid plan. Shared projects, cloud sync and app publishing are packaged-app features not yet in the open-source build, per the README's own note.

Illustration created for INORYVA — a developer working beside a terminal window, not a screenshot of the Orkas app.

Who Orkas fits, and who should skip it

The README's fit test is blunt. Orkas is for people who want a team rather than one assistant, work that keeps moving across days, and data, keys and agents on their own disk. It is not for people who just want a single all-purpose chatbot, or a pure code library to embed in their own app: not a single chatbot, not a code framework, not a hosted service, as the FAQ puts it. In practice that means a developer or small team already using Claude Code or Codex, with research, content or store work attached to the same project, sharing one plan, memory and review step.

The skip cases follow. A reader who only needs a chatbot answer will find a team and a desktop install to be overhead. A Linux user wanting a one-click installer does not have one yet. A reader wanting design checks on finished pages needs a different job from UIDesigner, which builds editable HTML-first interfaces; the Hallmark GitHub skill checks finished pages against design rules. Not to be confused with Orca, a separate project. That trade pays off only when the work is ongoing, multi-part and worth coordinating.

Key Takeaways

• The Orkas multi-agent desktop app runs a Commander plus 16 built-in specialists on the user's machine, with Claude Code, Codex, OpenCode, OpenClaw and Hermes-Agent as local sessions, not replacements.
• Checked 1 Nov 2026: 6,105 stars, 369 forks, 146 watchers, 820 commits; MIT licence; created 29 April 2026. An explainer, not a launch story; no release date claimed.
• Projects, to-dos, Auto-advance and scheduled automations carry work across days. Conversations, files and keys stay on disk by default; cloud sync is optional and packaged-only.
• Packaged for macOS 12+ and Windows 10+; Linux from source, no installer yet. The source build needs no account; the packaged app signs in on first launch.
• Free: the MIT app plus 500 credits for the built-in model. Paid plans from $11.99/month add monthly credits, cloud sync, more shared projects and app publishing.
• For developers and small teams wanting a coordinated team for ongoing work on their own machine. Not a single chatbot, framework or hosted service.

Conclusion

Orkas is infrastructure for work that does not fit in one chat: a Commander, 16 specialists, the coding agents a developer already uses, and a project that remembers. The 6,105 stars checked on 1 November 2026 say the idea has an audience; they do not prove it works everywhere. Judge it on one real multi-part job. If the work is single and short, skip it. If it keeps moving across days, this is the shape of tool built for it.

Sources

• GitHub — Orkas-AI/Orkas (repository statistics, README, FAQ, download and platform details; stats and README re-checked 1 Nov 2026 ~03:07 PKT, page copy crawled 31 Oct 2026 21:57 UTC): https://github.com/Orkas-AI/Orkas