A private AI workspace
that runs on your own machine.
Orkas is an open-source, local-first private AI workspace. Projects, files, agent configs, and memory stay on your machine by default, while a lead agent and its sub-agents work inside the same workspace instead of a hosted chat thread.
A workspace is a place you keep, not a session you rent
People searching for a private AI workspace are past the single-chat stage. They have real projects, recurring work, and files they cannot hand to a hosted service — and they want the AI to work inside that material, not on a copy of it somewhere else.
Projects, not threads
Work is organized as projects with their own files, agents, and history, so you can leave and come back weeks later.
- projects
- history
Shared local context
Every agent reads and writes the same local project files, so what one agent produces is immediately usable by the next.
- local files
- handoff
Your keys, your provider
Use optional managed official models, or connect your own provider. With your own provider, encrypted keys stay local and calls go directly to it.
- BYO keys
- direct calls
Sync only if you ask
Optional Lite sync stores only what you choose to sync on Orkas servers for multi-device access. It never uploads model keys or proxies model traffic.
- optional
- multi-device
Private AI workspaces, compared
No cherry-picking — the last row is where the hosted products win.
| Orkas This site | Msty | Notion AI | ChatGPT | |
|---|---|---|---|---|
| Form factor | Local desktop app | Local desktop app | Cloud workspace | Cloud web + mobile |
| Open source | ●Yes (MIT), auditable | No | No | No |
| Where workspace data lives | ●Your machine by default | Your machine | Notion's servers | OpenAI's servers |
| Model access | ●Managed models or BYO provider, called directly | Local and cloud models | Built-in models only | OpenAI models only |
| Multiple agents in one workspace | ●Lead agent dispatches sub-agents in parallel or in series | Autonomous multi-step tasks | Single assistant | Single assistant with hosted agent mode |
| Recurring work becomes | ●Sub-agents and skills stored locally | Prompt workflows and personas | Built-in AI actions | Custom GPTs, hosted |
| Zero-setup convenience | Install the app, then set up a model | Install the app | ●Instant in the browser | ●Instant, nothing to install |
Notion AI and ChatGPT are genuinely easier to start with: nothing to install and no model to configure. Orkas makes the opposite trade — you run the workspace yourself, and in exchange the projects, files, agent memory, and model keys stay on your machine and a whole agent team can work inside them.
Set up a private workspace in four steps
Install the desktop app
Run Orkas on macOS or Windows. The workspace lives in a local folder you can see.
Choose a model path
Use managed official models or connect your provider by OAuth or API key.
Add your project
Point the workspace at real files, a codebase, or a local document library.
Direct the agent team
Describe the work in chat. The lead agent splits it across sub-agents that share the workspace.
Workspace-shaped work, not one-off questions
Research over a private library
Keep PDFs and notes in a local Library, then ask questions with citations.
Multi-step delivery
Research, draft, review, and revise as separate sub-agents handing files to each other.
Work inside a codebase
Reproduce bugs, edit files, and run tests against the repo on your own disk.
Private AI workspace questions
What is a private AI workspace?
A private AI workspace is a working environment — projects, files, agent configs, and memory — that AI agents operate inside while it stays under your control. Unlike a hosted chat thread, the workspace is a durable place you own rather than a session on someone else's server. In Orkas that workspace is a local desktop workspace, and its data stays on your machine by default.
How is a private AI workspace different from a private AI assistant?
A private AI assistant is one assistant you ask questions. A private AI workspace is the shared place several agents work in: a lead agent dispatches sub-agents, they read the same local project files, and the results accumulate as your own assets. Orkas is both, but the workspace model is what lets recurring work become reusable sub-agents and skills.
Does Orkas send my workspace files or model keys to its servers?
No by default. Workspace files, agent configs, chat history, and generated outputs stay on your machine. With your own provider, encrypted model keys stay local and model calls go directly to that provider rather than through Orkas. Optional Lite sync stores only the data you choose to sync on Orkas servers, and Orkas-managed official models use Orkas's model service when you choose them.
Can several agents work in the same private workspace at once?
Yes. A lead agent owns the conversation and can dispatch specialized sub-agents in parallel or in sequence. Each sub-agent gets its own context but works against the same local workspace, so files one agent produces are available to the next.
Is Orkas private AI workspace open source?
Yes. The Orkas desktop client is open source under the MIT license and the source is published on GitHub, so you can inspect how the workspace stores data, how keys are handled, and how agents are dispatched.
Set up a private AI workspace on your own machine
Free, open source, local-first. Choose managed official models or your own provider, and direct an agent team by chat.