Orkas Review: The Free Open-Source AI Team App — and the $5.99 Catch

Your AI tools are scattered across a dozen tabs: a chatbot here, a coding agent in the terminal, a research helper somewhere else. Orkas wants to be the room where they all work together. It is an open-source, MIT-licensed desktop app that presents itself as a team rather than a chatbot: you tell a Commander what you need, and it plans the work, splits it between specialist agents, and brings finished results back for your review.

The pitch is landing. The Orkas-AI/Orkas repository has passed 7,200 GitHub stars and appeared on GitHub trending trackers on November 1. One correction up front, though: this is not a brand-new launch. The repository dates to April 2026, so the surge is about attention, not a release. Here is what Orkas actually is, what the free tier covers, what the paid plans buy, and where the fine print sits.

Official links: GitHub repository, orkas.ai, and pricing page.

What Orkas actually is

The README's own one-liner is unusually honest for the category: "An open-source team of AI agents on your desktop, turning your goals into finished work and keeping it moving." The emphasis is on continuity. A project holds the goal, the rules, the memory and the files, so each new conversation picks up where the last one ended instead of starting from a blank prompt.

The Commander is the coordination layer. For small jobs it finishes the work itself; for bigger ones it plans the steps and hands them to specialists. Sixteen of those specialists ship with the app:

The move that separates Orkas from a walled garden is that it also runs the coding and personal agents you already have installed: Claude Code, Codex, OpenCode, OpenClaw and Hermes-Agent, as local sessions next to its own specialists. It does not replace those tools; it puts them on the same team. A marketplace with 30+ additional agents exists, and you can describe a role and have the Commander build a custom agent for it.

For work that never ends, there are projects, to-do lists shared between people and agents, an Auto-advance mode that lets the team pick up the next to-do by itself, and scheduled automations that run hourly, daily, weekly or monthly. Connectors cover Gmail, Google Calendar, Sheets, Drive, Notion, Slack, Shopify, Stripe and more; by default, the app asks before it sends, posts or deletes anything. The interface ships in 18 languages, and the official downloads are for macOS and Windows, with Linux supported if you build from source.

What it costs

The app itself is free. What costs money is the managed model service, metered in credits, plus the hosted extras (multi-device sync, more cloud storage, web-app publishing). Orkas's own pricing page, read live for this article, lists:

GBP and CAD figures are approximations at November 2, 2026 market rates, verified at publication. Managed models include Claude Opus 5.5, GPT-6.1 Sol and Kimi K3, with more providers (OpenAI, Anthropic, Google Gemini, DeepSeek, Zhipu GLM, Moonshot/Kimi, MiniMax, Doubao, OpenRouter) available through your own keys.

Credit consumption, per the pricing page: roughly 60–120 credits per million input tokens, 6–12 for cached input, 360–540 for output; 0.5 credits per web search; 12 per image; 50–100 per second of video; 0.125 per second of speech recognition or synthesis. Lifetime credit packs sell for $3.99 (standard users) or $2.99 (members), roughly 70–350 chats, 2,000 searches, 83 images, 10–20 seconds of video, or 2+ hours of speech per pack. Monthly credits refresh at renewal and do not roll over; lifetime credits stay valid and are spent after monthly ones. Enterprise pricing is custom.

The important fine print: credits only meter the Orkas-managed model service. If you connect your own provider key, those calls go straight to your provider and are billed by it: they use no credits at all. If you already pay for Claude or OpenAI, the cheapest way to run Orkas is probably your own keys, not a plan.

The catches, stated plainly

The open-source build is not the packaged app. Shared projects, cloud sync and web-app publishing exist only in the official macOS/Windows release; the open-source build, which is the one Linux users compile themselves, does not support them yet. The MIT license covers the client, not the hosted services.

Self-hosting means running a desktop client, not a team server. Orkas's own FAQ is explicit about this: it is a desktop client, so "self-hosting" means running the app on your own machine rather than deploying a server for a team. If you expected multi-user infrastructure, that is not what this is.

The 500 welcome credits need a desktop-app sign-in. New users claim the 500 bonus credits by signing in inside the desktop app; signing in on the website does not grant them. This is the kind of detail that costs people their free trial of the managed models.

The pricing page contradicts itself. The plan cards list 1,000 / 5,000 / 10,000 monthly credits for Lite / Pro / Max, but the page's own FAQ says each plan includes "0–50,000 monthly credits": those two statements cannot both be right, and the page was also displaying a Stripe sandbox "test cards only, no real charge" notice when I read it. Treat the card numbers as the working figures and double-check at checkout.

The "30/44/48% less than provider API pricing" figures are Orkas's own arithmetic. Credit costs are derived from provider pricing and can change when providers change theirs. There is no independent audit of the discount; it is the project's claim about its own markup.

The trending spike is not a launch. The repository was created in April 2026. The November 1 trending appearance is a surge of attention, roughly a thousand new stars a day around that date per the tracker, not a version release. Star counts measure attention, not reliability.

Voice input and connectors have privacy edges. Voice recognition sends audio to Orkas and spends the key owner's credits. The "asks before it sends, posts or deletes" connector behavior is the project's own description; no independent security audit exists. Files, chats and API keys stay on your machine by default, and keys are stored encrypted locally; again, the project's claims, stated for what they are.

Setting it up in minutes

1. Download the app for macOS or Windows from orkas.ai, or clone the repository and build it yourself for Linux.
2. Sign in inside the desktop app to claim the 500 bonus credits: this step does not work from the website.
3. Choose your model path. Start on the managed models with the bonus credits, or add your own provider keys (Claude, OpenAI, Gemini, DeepSeek, Kimi, GLM, Qwen, MiniMax, Doubao, OpenRouter, or a local OpenAI-compatible endpoint). Own-key calls bypass credits entirely.
4. Give the Commander a goal in plain language. For bigger work it will plan the steps and pull in specialists; you can also browse the agent marketplace or describe a custom agent role and let the Commander build it.
5. Make it recurring. Turn repeated work into scheduled automations (hourly to monthly) and switch on Auto-advance so the team picks up the next to-do on its own. Finished work comes back for your review before anything goes out.

Orkas vs the alternatives

Orkas is entering a crowded week: this site has recently covered OpenClaw, the assistant you text from WhatsApp, Claude Code, Anthropic's terminal coding agent, and claude-mem, the memory layer for coding agents. How does Orkas differ?

The honest way to frame it: Orkas is the team room, not the worker. If you already live in Claude Code or OpenClaw, Orkas's value is coordination: projects, memory, scheduled runs, rather than replacing those tools. And because it can run them as local sessions, it is closer to a complement than a competitor.

Pro tips and common mistakes

Bring your own keys first. If you already pay Anthropic, OpenAI or Google for API access, connect those keys before buying credits. Credits only meter Orkas-managed models; your own keys cost you nothing extra inside Orkas.

Stay local if that is the point. The local-first pitch only holds while cloud sync stays off and you skip the hosted extras. Turn on sync and your project data lives on Orkas servers: that is the trade, stated plainly in their own FAQ.

Read the discount math skeptically. "44% less than provider API pricing" applies to managed-model credits, and the page's own FAQ disagrees with its plan cards about how many credits each plan includes. Trust the card numbers and recheck at checkout.

Do not confuse the builds. The open-source build lacks shared projects, cloud sync and web-app publishing. If those matter to you, you need the packaged release, and on Linux you are building from source either way.

Check connector permissions per service. The default ask-before-send behavior is the project's claim. Before connecting Gmail or Stripe, decide which permissions each service actually needs, because the agent's convenience scales with its access.

Is Orkas really free?

The desktop app is free and includes 500 bonus credits for Orkas-managed models, claimed by signing in inside the app. Bring-your-own-key usage is billed by your provider, not Orkas. Paid plans ($5.99–$29.99/month) add monthly credits, multi-device sync, more storage and web-app publishing.

Is Orkas open source?

Yes — the desktop client is MIT-licensed at github.com/Orkas-AI/Orkas, where you can read and audit the code. The hosted services (managed models, cloud sync, storage) are proprietary. The open-source build currently lacks shared projects, cloud sync and web-app publishing.

What do Orkas credits actually cost?

500 credits come free with the app. Lifetime packs cost $3.99 (or $2.99 for members) and cover roughly 70–350 chats or 2,000 web searches each. Monthly plans bundle 1,000–10,000 credits for $5.99–$29.99. Consumption runs about 60–120 credits per million input tokens. Monthly credits do not roll over.

Can Orkas use my existing Claude Code or OpenCode?

Yes. Orkas runs your installed CLI agents: Claude Code, Codex, OpenCode, OpenClaw and Hermes-Agent, as local sessions alongside its own specialists. It coordinates them rather than replacing them, which makes it a complement to tools you already use.

Does Orkas work on Linux?

The official download page offers macOS and Windows builds. Linux is supported through the open-source repository, which you build yourself, but note the open-source build does not yet support shared projects, cloud sync or web-app publishing.

Orkas is the most literal take yet on the "AI team" idea: not one assistant, but a Commander with staff, memory and a calendar. The app is free to start, the MIT license is real, and the trending spike is earned attention for a working product. The catches are all in the pricing page's fine print: contradictory credit allowances, credits that only cover managed models, and hosted features the open-source build does not have. Read those before you commit, and the tool is worth a look. You can start with the free download and decide whether the team earns its keep.

This is general information, not financial or procurement advice.

Sources

• Orkas GitHub repository and README (official): github.com/Orkas-AI/Orkas
• Orkas homepage (official): orkas.ai
• Orkas pricing page (official): orkas.ai/pricing
• GitHub API repository data (stars, forks, license, dates)
• OctoCounts GitHub trending tracker, November 1, 2026
• opensourcedex.com Orkas tool listing
• inoryva.blogspot.com Orkas explainer, October 31, 2026

Internal Link Opportunities

• OpenClaw review (WhatsApp-controlled assistant) — linked
• Claude Code explainer — linked
• claude-mem memory-layer review — linked
• Future: 7 best AI coding agents roundup (already covers the CLI agents Orkas orchestrates)