Orkas: 16 AI Specialists on Your Desktop, Commanded by One Coordinator
Talk to Claude Code for three hours, finish your research report, close the terminal. Reopen it tomorrow and it asks: "What were we working on?"
That's the fatal flaw of every chat-based agent: the session is the context, and once it's gone, it's gone.
Orkas's angle isn't "a stronger model" — it's "a more persistent workspace." Work is organized into projects; each project holds goals, rules, memory, and files. Every new conversation picks up where the last one left off, with no need to re-explain background.
Commander: Doesn't Do the Work, Just Breaks It Down
The core of Orkas is an agent called Commander. You don't talk to 16 specialists individually — you only talk to Commander.
Say: "Do a competitive analysis, then turn it into a deck and a short video."
Commander does three things: 1. Decompose: turns the goal into a to-do list (research competitors → write report → build deck → produce video) 2. Assign: routes each to-do to the right specialist (DeepResearcher for research, ContentWriter for the report, PptMaker for the deck, VideoStudio for the video) 3. Auto-advance: in Auto-advance mode, the next to-do starts automatically once the previous one finishes
Think of it as a project manager. Commander doesn't write code, do research, or edit video — it only breaks work apart, delegates it, and tracks progress.
16 Built-In Specialists
Orkas ships with 16 specialist agents, each with its own skills, memory, and tools:
| Domain | Agent | Output | |--------|-------|--------| | Research & analysis | DeepResearcher | Research reports with source verification | | Research & analysis | SeoGeoAgent | Technical SEO audits + GEO citation optimization | | Research & analysis | StockAnalyser | Stock research, factor screening, backtesting | | Content & media | ContentWriter | Articles, newsletters, case studies | | Content & media | PptMaker | Editable PPTX decks | | Content & media | ImageStudio | Posters, covers, infographics | | Content & media | VideoStudio | Voiceover, AI anchors, editing | | Office files | OfficeWorker | Word/Excel/PPT/PDF | | E-commerce | 5 e-commerce agents | Product selection, listings, review analysis, sales analysis |
If 16 isn't enough, browse 30+ community agents in the agent marketplace, or describe a role and let Commander build a custom agent for you.
Plugging In Your Coding Agents
Orkas doesn't want to replace your coding tools — it wants to orchestrate them. Commander can invoke:
- Claude Code
- Codex
- OpenCode
- OpenClaw
- Hermes-Agent
- Built-in Orkas model with 500 free credits, no API key required
- Bring your own key: Claude, OpenAI, Gemini, DeepSeek, Kimi, GLM, Qwen, MiniMax, Doubao
- Local OpenAI-compatible endpoints are also supported
- Scrape competitor pricing into a spreadsheet every morning at 8
- Generate last week's SEO audit report every Monday
- Run a monthly portfolio risk check
- Desktop app: macOS and Windows only, no Linux version (unfriendly to server use cases)
- Early stage: quality across the 16 agents is uneven; each one's output needs real testing
- Model dependency: built-in model capabilities are unknown; your own keys likely perform better but cost more
- JavaScript stack: may not be optimal for performance-sensitive scenarios
- GitHub: https://github.com/Orkas-AI/Orkas
- License: MIT
- Platforms: macOS + Windows
- Stars today: 1,118
- Built-in agents: 16
- Integrated coding agents: Claude Code / Codex / OpenCode / OpenClaw / Hermes
So you can have Orkas do market research (DeepResearcher), then pass the results to Claude Code for implementation — all within one project, sharing the same memory.
Local-First
All data stays on your machine. Orkas doesn't upload project files, conversation history, or agent memory to any server.
Models:
Scheduled Automations
This is the feature that separates Orkas from "yet another chatbot": scheduled automations.
Tasks can run hourly, daily, weekly, or monthly. Examples:
This isn't "reminding you to do it" — the agent does it itself and hands you the results for review.
Why This Matters
Orkas represents an emerging paradigm shift: from chat to workspace.
Most current AI tools use a prompt-response interaction model — complex tasks require you to manually decompose, chain, and manage context. Orkas's model is "one project" — you give a goal, Commander decomposes, assigns, advances, and remembers. When you come back, you see progress, not an empty conversation.
This aligns with OpenBot's "agent desktop workspace" concept: agents don't need bigger context windows, they need more persistent work environments. Chat is ephemeral; a workspace is continuous.
Limitations and Risks
Data
Nothing Orkas does is new on its own — multi-agent coordination, project memory, scheduled tasks — but packaging it into a local-first desktop app that non-technical users can actually use is new. The three-layer architecture of Commander + 16 specialists + plugged-in coding agents turns the "AI team" concept into something you can download.