A private AI assistant that keeps
your work under your control.
Orkas is an open-source, local-first desktop AI assistant. Workspace data stays on your machine by default. Choose optional managed official models or your own provider; your own-provider calls go directly to it.
Private is not just "no cloud". It is control.
People searching for a private AI assistant usually want to work with real files, code, notes, and credentials without handing the whole workspace to a hosted chatbot.
Files stay local
Use documents, PDFs, generated files, and local knowledge bases without making Orkas the default host for your workspace.
- local files
- library
Choose model access
Use optional managed official models or connect OpenAI, Claude, Gemini, or another provider. Calls using your own provider go directly to it.
- OAuth
- API key
Agent data stays under control
Agent configs, chat history, memory, and outputs live locally by default, so recurring workflows remain your asset.
- agents
- workflows
Cloud sync is optional
Lite sync is available when you need multiple devices. It does not upload model keys or route your model calls through Orkas.
- sync
- optional
Private AI assistants, compared
"Private" means different things to different products. The last row is where the hosted options win.
| Orkas This site | Proton Lumo | Jan | ChatGPT | |
|---|---|---|---|---|
| Form factor | Local desktop app | Hosted service with web and mobile apps | Local desktop app | Cloud web + mobile |
| Open source | ●Yes (MIT), client fully auditable | Apps are open source; the hosted service is not | ●Yes | No |
| Where your files live | ●Your machine by default | Proton's servers, encrypted | ●Your machine | OpenAI's servers |
| Model access | ●Managed models or your own provider, called directly | Proton-hosted models only | Local models, or a provider you connect | OpenAI models only |
| Agent model | ●Lead agent plus a sub-agent team | Single assistant | Single assistant | Single assistant with hosted agent mode |
| Zero-setup convenience | Install the app, then set up a model | ●Instant in the browser | Install the app, then get a model | ●Instant, nothing to install |
Proton Lumo is a strong choice if you want privacy without running anything yourself, and Jan is excellent if you mainly want local models in a single chat. Orkas is aimed at the case where the private material is real project work — files, a codebase, recurring workflows — and one assistant is not enough, so a lead agent splits the job across sub-agents that all stay on your machine.
A local-first control plane for AI agents
Install the desktop app
Run Orkas on macOS or Windows, not inside a hosted web workspace.
Choose a model path
Use managed official models or add your provider through OAuth or an API key.
Choose permissions
Approval modes prompt for sensitive local operations; routine operations may run without prompting.
Keep outputs local
Generated files, notes, and reusable workflows remain on your machine by default.
A private AI assistant should be able to answer "where is my data" item by item
"Local-first" is only meaningful if it is specific. This is what stays on your machine, what leaves it, and what changes when you turn on optional sync.
| Data | By default | If you turn on Lite sync |
|---|---|---|
| Workspace files and generated outputs | ●On your machine | Only what you choose to sync is stored on Orkas servers |
| Chat history | ●On your machine | Synced for multi-device access |
| Agent configs and skills | ●On your machine | Synced for multi-device access |
| Library index and embeddings | ●Built on your machine with a local embedding model | Unchanged — indexing stays local |
| Your own model keys | ●Encrypted and stored locally | ●Never uploaded |
| Model requests using your own provider | ●Sent directly to that provider, not proxied by Orkas | ●Unchanged — sync does not route model traffic |
| Model requests using Orkas-managed official models | Handled by the Orkas model service, only when you choose those models | Unchanged |
Two things are worth stating plainly. Sync is off unless you turn it on, and turning it on never uploads model keys or routes model traffic through Orkas. And choosing an Orkas-managed official model is the one case where a model request reaches Orkas infrastructure at all — using your own provider key never does.
Work with sensitive context without moving it into a hosted chatbot
Chat with documents
Add PDFs and files to a local Library, then ask questions with citations.
Work in a codebase
Let repo-aware agents reproduce bugs, edit files, run tests, and explain diffs.
Reuse recurring work
Turn repeat tasks into sub-agents you can summon again by chat.
Private AI assistant questions
What is a private AI assistant?
A private AI assistant is an AI app designed so your files, chats, credentials, and workflow data remain under your control instead of living by default in a hosted chatbot or model-proxy service.
Does Orkas upload my files or model keys?
Workspace files stay on your machine by default. With your own provider, encrypted model keys stay local and calls go directly to that provider; official models use Orkas's managed model service.
Can a private AI assistant still use cloud models?
Yes. Orkas can use optional managed official models or providers you connect through OAuth or an API key. With your own provider, its model traffic goes directly to it.
What changes if I enable Orkas cloud sync?
Optional Lite sync stores the data you choose to sync on Orkas servers so it is available across devices. It does not proxy your model traffic or upload your model keys.
Is Orkas a private AI file assistant?
Yes. Orkas works directly on files that stay on your machine: add PDFs, documents, notes, or a codebase to a local Library or project, then ask questions and get answers with citations. Because it is a desktop app rather than an upload-first web service, a personal AI with private files does not require moving those files into a hosted account first.
Can I run a private AI assistant server?
Orkas is a desktop application, not a self-hosted server you deploy for a team. Each person installs Orkas on their own machine, and that machine is where the workspace and encrypted model keys live. If you want a shared local workspace where several agents work on the same project, that is the private AI workspace model rather than a server.
Try a private AI assistant that starts on your machine
Free, open source, local-first. Choose managed official models or your own provider and direct your AI team by chat.