Orkas is an open-source, local-first desktop application in which one "Commander" agent takes a plain-language goal and hands the work to a team of sixteen built-in specialist agents. It is aimed at two kinds of reader: someone who wants a run of work delegated rather than typed back and forth in a chat window, and someone who already pays for Claude Code or Codex and would like another program to direct them. Running it takes no orchestration code, which is the project's main argument against the developer frameworks it compares itself to.
What it does
You give the Commander one goal in plain language. It turns that goal into a to-do list and routes each item to the specialist that fits: researchers, writers, e-commerce helpers, and two creative agents, VideoStudio, which recreates a reference video using your own footage and assets, and ImageStudio, which produces posters, covers and infographics.
Two features sit underneath the agents. Orkas keeps a project memory across sessions, holding goals, rules and files, so the next session does not start blank. And a project can run on a schedule: hourly, daily or weekly.
The part the video is about is the hiring. The Commander can launch outside coding command-line tools as its own subordinates. The README names Claude Code, Codex, OpenCode, OpenClaw and Hermes-Agent, coordinated inside the same project alongside the sixteen built-in agents.
How it works
Direction happens entirely in chat. The project's own pitch is "a team, not a chat box": you describe the outcome, the Commander does the splitting and the assigning, and you are not writing graphs, chains or crew definitions the way LangChain or CrewAI would ask you to.
Models are pluggable. Orkas ships with a built-in model, and users can connect their own keys or subscriptions for OpenAI, Claude, Gemini, DeepSeek, Kimi and GLM, or point it at any OpenAI-compatible local endpoint. The project describes itself as local-first: by default your files and your model API calls stay on your own machine, and cloud sync is an optional extra in the paid, packaged app. Those are the project's own claims from its README, not figures anyone has measured independently.
Getting started
The source lives at github.com/Orkas-AI/Orkas under the MIT licence. The material here does not describe a download beyond that paid packaged build, so plan on cloning the repository and following its own setup notes. The project does not list hardware requirements: no operating system, chip, memory or GPU figure appears anywhere in the material, so do not assume it will run on whatever machine you have.
Paying for the intelligence is the part worth thinking about. The app itself is open source, but the agents still need a model behind them. You can use the built-in model, or bring your own key, which means a bill from OpenAI, Anthropic, Google or whichever provider you choose; running an OpenAI-compatible model on your own hardware is the way to avoid API fees, if the hardware is there. The README's pricing section mentions a free tier and a paid packaged app whose addition is cloud sync, but the saved copy is cut off mid-sentence, so read the current tiers on the project's own site rather than trusting a number quoted here. Claude Code and Codex are not bundled: each needs its own installation and its own account.
When to use it / when not
The main catch is the headline trick itself. Orkas coordinates Claude Code and Codex; it does not supply them. The demonstration that sells the app only works after you have installed those tools and paid for them separately, and whatever limits your accounts have apply inside Orkas too.
Use it instead of just asking Claude when the job is a standing one with several kinds of work in it: research, then writing, then an image, repeated weekly, with the project's rules remembered between runs. Don't use it if your task is one question or one patch, where a chat window or a single coding CLI will finish sooner with fewer moving parts.
Alternatives
Orkas positions itself against two things a reader already knows. One is the ordinary AI chatbot, a single chat box answering one prompt at a time, which has no to-do list, no scheduled runs and no project memory. The other is a developer orchestration framework such as LangChain or CrewAI, which can express far more complicated agent topologies but expects you to write code to do it. Orkas is not a replacement for a coding assistant: it treats Claude Code and Codex as team members rather than competitors.
Take this repository seriously if you already keep two or three coding agents installed and have wished something would run them against the same project at once, or if you want a multi-step routine that repeats on a schedule without any code to maintain. Treat the claims as the project's own until you have run it, budget for the model bill and for the separate coding-agent accounts, and check the repository for the platform and setup details this material does not cover.