Orkas
Orkas · Orkas: Open-Source Desktop AI Team Where a Commander Runs 16 Specialist Agents Plus Claude Code and Codex
Orkas is a free, MIT-licensed desktop app for macOS and Windows in which a Commander agent turns a goal into to-dos and hands the work to 16 built-in specialist agents for research, writing, slides, images, video, office files, code and e-commerce, plus Claude Code, Codex, OpenCode, OpenClaw and Hermes-Agent if you have them installed. It suits solo founders, store owners and small teams who want finished deliverables and recurring work handled on their own computer, using the built-in Orkas models (500 free credits) or their own API keys.
Best for
Solo founders, online store owners, agencies and small teams on macOS or Windows who want one app that produces finished content, research, decks, videos and store analysis, keeps recurring work running on a schedule, and keeps files and data on their own computer, with the choice of managed models or their own keys
Not ideal for
Teams that need agents running around the clock while laptops are off, Linux users who want an installer, organizations that need compliance documentation or a named vendor under their own jurisdiction, and developers who want a code-first framework to embed in their own app
Who it's for
Solo founders, e-commerce store owners, agencies, creators and small teams who want a team of AI agents producing finished work and running recurring tasks from their own desktop
Capabilities
- A Commander agent that breaks a goal into steps and to-dos, does general work itself, and delegates to specialists by dispatching a task, handing off the conversation, or running an isolated anonymous worker
- 16 built-in specialist agents, each with its own skills, memory and tools: DeepResearcher, SeoGeoAgent, StockAnalyser, ContentWriter, PptMaker, ImageStudio, VideoStudio, VoiceStudio, UIDesigner, ProductDeveloper, OfficeWorker and five e-commerce agents
- A five-agent e-commerce team for product selection, listing copy, review and reputation analysis, sales and inventory analysis from your exports, and catalog, price and promotion changes made only through tools you connect
- Runs Claude Code, Codex and OpenCode, plus OpenClaw and Hermes-Agent, as local CLI sessions in the same chat, with native CLI approval requests forwarded to an Orkas confirmation dialog
- Projects that keep goals and rules, memory, files, agents and Library sources together, so each new conversation starts from the same context
- A to-do board shared by people and agents, with an optional Auto-advance mode that picks up the next pending to-do when its dependencies are met and returns finished work for review
- Automations that run a task once later or every few hours, daily, weekly or monthly, with multiple weekdays or month dates, an end date or run-count limit, and catch-up of the latest missed run within 12 hours
- A local Library that indexes PDFs and documents in SQLite with an on-device embedding model and answers with citations to source passages
- Connectors for Gmail, Google Calendar, Sheets and Drive, Notion, Slack, Shopify, Stripe and more, plus custom MCP servers, with permission levels that ask before sensitive actions
- Builds small tools such as quote calculators and booking forms as local apps inside Orkas; paid plans can publish compatible ones as web apps on a public link
- One model setting for the Commander and built-in agents: the built-in Orkas-1.5 and Orkas-1.5 Pro models, your own Claude, OpenAI, Gemini, DeepSeek, Kimi, GLM, Qwen, MiniMax or Doubao keys, or a local OpenAI- or Anthropic-compatible endpoint such as Ollama or LM Studio
- Background reviews while the app is idle that can update each agent's self-assessment, reusable methods and learned skills
- Create custom agents by describing a role, install agents and skills from a marketplace, or import skills from a URL, folder or ZIP
Limitations
- Runs on your computer, so Auto-advance and automations need the computer awake and Orkas running; there is no hosted always-on option
- Packaged installers exist only for macOS 12+ and Windows 10+; Linux (glibc 2.34+, x64 or arm64) must run from source with Node 22.12+, and musl-based distributions such as Alpine are not supported
- The packaged app requires signing in to an Orkas account, and the free edition sends limited usage analytics
- Shared projects, cloud sync and web-app publishing are only in the packaged app; the MIT-licensed source build does not support them yet
- The built-in Orkas models, voice input and other managed services run through Orkas servers and use credits; only calls made with your own provider keys go directly from your machine to the provider
- By default agents can reach all files and ask only before sensitive file or shell operations; the all-files auto mode and the Trusted connector level skip those confirmations
- Published web apps cannot use the desktop AI, local files or connectors, and publishing requires a paid plan
- A young project: the repository was created in April 2026 and ships releases every one to two weeks
- The terms of service are governed by the laws of the People's Republic of China, and the website does not name the operating company
Use cases
- Turning one request into a product launch kit: a 30-second video with captions, product page copy and a week of social posts with images
- Running a Monday automation that reads store orders, reviews and stock and lists the three things to fix, with review reply drafts held for approval
- Preparing a client's weekly update deck in an agency template from the week's ad-account exports
- Having ProductDeveloper or Claude Code fix a bug with tests while ContentWriter drafts the changelog and launch post in the same project
- Answering questions from a folder of policies, manuals or research papers with citations back to the source passages
- Building a wholesale quote calculator and publishing it as a web app customers can open from a link
Our take
Orkas packages a lot into a free, MIT-licensed desktop app: a Commander that delegates, sixteen ready-made specialists including a rare dedicated e-commerce team, and the project, to-do and automation layer that most single-task agents lack. For a store owner or small agency that wants weekly reports, content and review replies produced on their own machine, that combination is unusual, and the option to bring Claude Code or Codex into the same plan makes it useful for small product teams too. The open-source label needs reading carefully, though: the code is MIT, but collaboration, cloud sync and publishing are only in the packaged app, which needs an account, and the built-in models run through Orkas's servers. Add a desktop-only runtime that stops when the laptop sleeps, a project only months old, an unnamed operator, and terms governed by Chinese law, and it is a strong pick for individuals and small teams experimenting with an AI team, but not yet something to build regulated or business-critical workflows on.
Who should use it
Solo founders, Shopify and other online store owners, agencies and creators on macOS or Windows who want weekly reports, content, decks, videos and store analysis produced by a team of agents on their own computer, and developers who want Claude Code or Codex working inside the same plan as research and writing agents.
Who should skip it
Teams that need automations running while their computers are off, Linux users who want a packaged installer, organizations that require compliance documentation or a named vendor, and developers looking for a framework to build agents in code.
Strengths
- MIT-licensed code and a free app with 500 credits, so you can start without an API key
- A ready-made team of 16 specialists, including a dedicated e-commerce team, rather than a single general agent
- Projects, a shared to-do board, Auto-advance and scheduled automations keep multi-day work moving
- Runs Claude Code, Codex, OpenCode, OpenClaw and Hermes-Agent alongside its own agents in one plan
- Local-first: files, memory and an on-device document index stay on your computer, and your own keys call providers directly
- Works with many model providers or a local endpoint, and paid plans are inexpensive
Weaknesses
- Automations stop when your computer sleeps or Orkas is quit
- Collaboration, cloud sync and app publishing are missing from the open-source build
- The packaged app requires an Orkas account and sends usage analytics on the free edition
- No Linux installer; Linux means a source build
- Very young project, with an unnamed operator and terms governed by PRC law
Orkas pricing
Free
$0
- 500 one-time credits for the built-in Orkas models, claimed by signing in to the desktop app
- Commander, specialist agents and local CLI agents
- Built-in models or your own provider keys
- Up to 5 shared projects with 5 members each and 100 MB of cloud storage
Lite
$5.99
Billed monthly
- 1,000 monthly credits at the listed price; $3.99 with no monthly credits, up to $77.99 with 50,000
- Cloud sync across unlimited devices and 1 GB of cloud storage
- Up to 50 shared projects with 50 members each
- Publish up to 50 web apps
Pro
$16.99
Billed monthly
- 5,000 monthly credits at the listed price; $8.99 with no monthly credits, up to $82.99 with 50,000
- Cloud sync and 5 GB of cloud storage
- Up to 50 shared projects with 50 members each
- Publish up to 50 web apps
Max
$29.99
Billed monthly
- 10,000 monthly credits at the listed price; $14.99 with no monthly credits, up to $88.99 with 50,000
- Cloud sync and 10 GB of cloud storage
- Up to 50 shared projects with 50 members each
- Publish up to 50 web apps
Enterprise
Custom
- Plans shaped around team size and usage
- Fits purchasing requirements
Free tier limits: 500 one-time credits for Orkas-managed models and services, up to 5 shared projects with 5 members each, and 100 MB of cloud storage; no personal multi-device sync or web-app publishing. Work done with your own provider keys does not use credits.
Note: Credits pay for Orkas-managed models, web search, image and video generation, speech and some connector calls; usage is calculated from the providers' official prices, which Orkas says its managed model rates undercut. The prices shown are the listed defaults: each plan has a monthly credit selector (0, 1,000, 5,000, 10,000 or 50,000 credits) that sets the price, so the plans differ mainly in cloud storage, and a one-month purchase without auto-renewal costs $1 more ($6.99, $17.99 and $30.99 at the defaults). Monthly credits do not roll over; lifetime credit packs start at $3.99 for 1,000 credits ($2.99 for members). The project's README says paid plans start at $11.99 a month (the current Lite price with 5,000 monthly credits) and lists lower web-app publishing limits per plan than the pricing page, which is used here.
Technical specs
Modalities
Text, Image, Video, Voice, Documents
Available models
Where Orkas excels
A weekly store brief for a small online shop
ECommerceAnalyzer can read sales, refund and inventory exports on a Monday automation and list what needs attention, while ECommerceReviewer drafts review replies that wait for approval.
A launch kit from one request
The Commander can split a product launch into a video for VideoStudio, page copy for ContentWriter and social images for ImageStudio, then combine the results in one project.
Recurring client reporting at an agency
A project keeps the client's template, rules and files, so a Friday automation can have OfficeWorker clean the week's numbers and PptMaker update the deck without re-briefing.
Coding and release work in one plan
Claude Code or Codex can fix a bug as a local CLI session while built-in agents write the changelog and launch post, coordinated by the same Commander.
Orkas vs. competitors
Orkas vs. Manus
Manus is a closed, cloud-first agent that plans and runs tasks in its own sandbox, so work can continue while your computer is off. Orkas does the same goal-to-deliverable job on your desktop with MIT-licensed code and local data by default, and adds projects, to-dos, scheduled automations and a fixed team of named specialists.
Orkas vs. Claude Cowork
Claude Cowork is Anthropic's agent for everyday work inside the Claude app, on a paid Claude plan and Claude models. Orkas is free to start, works with many model providers or a local endpoint, keeps data on your machine by default, and can run Claude Code alongside its own specialists, but it needs your computer awake for scheduled work.
Orkas vs. WorkBuddy
WorkBuddy is Tencent's desktop agent that turns one-sentence requests into reports, decks and spreadsheets from local files, and can take tasks from messaging apps. Orkas covers similar office deliverables but is open source, adds video, voice, image and e-commerce specialists, and is organized around projects, to-dos and automations.
Orkas vs. OpenMausBot
OpenMausBot is an Apache-2.0 desktop app that runs a roster of chat-contact bots on agent CLIs you already have, such as Claude Code, Codex or Grok, with optional paid always-on Cloud plans. Orkas ships its own specialist agents and built-in models, so it works without any agent CLI installed, and treats CLI agents as optional team members.
Frequently asked questions
What is Orkas?
Orkas is an open-source, local-first multi-agent desktop app. You tell its Commander what you need, and it plans the work, does general tasks itself and brings in specialist agents, such as DeepResearcher, ContentWriter, PptMaker, VideoStudio or the e-commerce team, plus Claude Code, Codex and other CLI agents you have installed. Projects, to-dos and scheduled automations keep the work going across days.
Is Orkas free and open source?
Yes. The code is on GitHub at Orkas-AI/Orkas under the MIT license, and the packaged app is free to download with 500 credits for the built-in Orkas models. Two editions differ: the packaged macOS and Windows app requires an Orkas account and includes shared projects, cloud sync and web-app publishing, while the source build has no account layer but does not support those three features yet.
How much does Orkas cost?
The app is free with 500 one-time credits. Orkas's pricing page lists Lite at $5.99 a month with 1,000 monthly credits, Pro at $16.99 with 5,000 and Max at $29.99 with 10,000. Each plan lets you choose from 0 to 50,000 monthly credits, which changes the price: the cheapest option is Lite with no monthly credits at $3.99 a month. One-month non-renewing purchases cost $1 more. Paid plans add cloud sync across devices, 1 to 10 GB of cloud storage, up to 50 shared projects and web-app publishing. Credit packs start at $3.99 for 1,000 credits, and Enterprise pricing is custom. Services you run on your own API keys do not use credits.
Do I need an API key to use Orkas?
No. The packaged app includes 500 credits for the built-in Orkas models (Orkas-1.5 and Orkas-1.5 Pro). After that you can buy credits, subscribe, or add your own keys for providers such as Anthropic, OpenAI, Google Gemini, DeepSeek, Kimi, Qwen or MiniMax. You can also point it at a local OpenAI- or Anthropic-compatible endpoint such as Ollama or LM Studio. External CLI agents like Claude Code keep their own model settings.
Does my data leave my computer?
By default, conversations, generated files, agent configurations, memory and the Library index stay on your disk, and your own API keys are encrypted locally with AES-256-GCM and excluded from sync. With your own keys, model calls go directly from your machine to the provider. The built-in Orkas models, voice input, optional cloud sync and shared projects go through Orkas servers, connector calls go to the services you authorize, and the free edition sends limited usage analytics.
Does Orkas keep working when my computer is off?
No. Orkas runs on your computer, so Auto-advance and scheduled automations need the computer awake and Orkas running. Closing the window keeps it running in the background, and when the computer wakes or Orkas restarts, each automation catches up its most recent missed run within 12 hours.
Orkas vs Manus: what is the difference?
Both take a goal and return finished deliverables such as reports, decks and media. Manus plans and runs tasks in its own cloud by default, so cloud tasks can keep going while your computer is off. Orkas runs its agents on your computer, keeps data local by default, is MIT-licensed, and organizes work into projects, to-dos and automations, with a built-in e-commerce team and support for running Claude Code or Codex as team members.
Integrations & fit
Alternatives to consider
Manus
Professionals and teams who need AI to complete defined tasks end-to-end — research reports, data extraction, scheduled automation, presentations — rather than assist with each step manually
Freemium
Anthropic
Professionals with recurring document, spreadsheet, or research workflows who already pay for Claude and want them done autonomously
Paid
Tencent (Tencent Cloud)
Office professionals, analysts and small teams who produce reports, decks and spreadsheets from files on their own computer, want an agent that can be tasked from WeChat, Slack, Telegram or Discord, and prefer a free tier with low-cost credit plans; especially teams already on Tencent tools such as WeCom, Tencent Docs or QQ Mail
Freemium
MausBot (Supamaus Software Private Limited)
Individuals and small teams who already pay for Claude, ChatGPT, SuperGrok or another supported agent, are comfortable installing and signing in to an agent CLI, and want several named bots with their own models, computers, connected apps and schedules in one chat app on their own machine, with the option to move them to an always-on hosted Cloud later
FreemiumAbout Orkas
Orkas is built around a Commander. You describe the outcome you want; the Commander handles general analysis, writing and file work itself, and routes specialist work to named agents through three tools: `dispatch_to` runs an agent and returns its result, `hand_off_to` passes the rest of the conversation to an agent, and `run_worker` runs an isolated anonymous sub-task. Sixteen specialists ship with the app: DeepResearcher, SeoGeoAgent and StockAnalyser for research; ContentWriter, PptMaker, ImageStudio, VideoStudio, VoiceStudio and UIDesigner for content and media; ProductDeveloper and OfficeWorker for code and Word, Excel, PowerPoint and PDF files; and a five-agent e-commerce team for product research, listings, review analysis, sales analysis and catalog operations. Claude Code, Codex and OpenCode, and the personal agents OpenClaw and Hermes-Agent, can join the same conversation as local CLI sessions with their own model settings. What separates Orkas from a one-shot task agent is the layer that keeps work going: projects hold goals, rules, memory and files; a to-do board splits work between people and agents, with an optional Auto-advance mode that takes the next item by itself; and automations repeat any task hourly, daily, weekly or monthly. A local Library indexes your documents with an on-device embedding model, connectors cover Gmail, Google Workspace, Notion, Slack, Shopify, Stripe and custom MCP servers, and Orkas can build small local apps, which paid plans can publish as public web apps. Conversations, files, memory and agents stay on disk by default. With your own keys, model calls go straight to the provider; the built-in Orkas models, voice input, cloud sync and shared projects go through Orkas servers. The code is MIT-licensed, but the open-source build lacks shared projects, cloud sync and web-app publishing, which are only in the packaged app, and the packaged app requires an Orkas account. Paid plans (Lite, Pro and Max, from $3.99 a month without monthly credits or $5.99 with 1,000) mainly buy monthly credits for the managed models and services, more cloud storage, multi-device sync and app publishing. The tradeoffs: everything runs on your computer, so automations stop when it sleeps or Orkas is quit; Linux has no installer and needs a source build; the project only started in April 2026; and the terms of service are governed by the laws of the People's Republic of China.
Updates from Orkas
Orkas 2026.11.4 lets you pick several weekdays or month dates for a recurring automation, with clearer schedule summaries, adds GPT-6.1 Terra to supported provider and model settings, and improves product-page capture and video-composition recovery. Fixes cover recovering queued messages and interrupted conversations and keeping search results complete through cache rebuilds.
Orkas 2026.10.27 lets you configure public image and video models with your own provider keys alongside Orkas's managed services, connect custom models through the OpenAI Responses, Chat Completions or Anthropic protocols, and create, edit and preview local apps separately from publishing.
Are you the founder? Claim this listing →