It works your machine, not a sandbox
Real files, real shell commands, a real browser session. The work happens where your work already is, not in a copy-and-paste window.
For macOS, Windows and Linux
Ollie Assistant
An agent that operates your computer.
It reads and writes your files, runs commands, drives a browser, works your mail and documents, and connects to the tools your team already uses — on your machine, under your control. Point it at whichever model you trust: one running locally with the network off, or Anthropic, OpenAI or Gemini with your own key.
No subscription, no account, and no per-seat meter you cannot see.
macOS (Intel and Apple Silicon) · Windows · Linux (.deb and AppImage)
The difference
Most AI apps are a window onto someone else's computer.
The assistants you have tried are clients. The model runs in a datacenter, the conversation lives in an account, and the app only touches what its vendor allowed. Ollie Assistant is a runtime that runs on your hardware — so the limits are the ones you set.
You choose the brain
Local models through Ollie Intelligence or Ollama, or your own key for Anthropic, OpenAI or Gemini. Switch in a dropdown — your tools, skills and history carry over unchanged.
It keeps working when the network doesn't
Paired with a local model the whole thing runs offline. On a plane, on a closed network, in a building with no signal.
Your conversations are files on your disk
Not rows in a vendor's database. Back them up, move them to another machine, or delete them. There is nothing to sign up for.
Extend it without a rebuild
Tools, skills, personas and providers are files in a folder. Drop one in and restart. No SDK, no plugin review, no waiting on a vendor roadmap.
Speaks MCP
Connect any Model Context Protocol server and its tools appear alongside the built-ins. Your internal systems, your rules.
Provider-agnostic
One agent. Any model. Your call.
Most AI software marries you to one vendor's model and one vendor's bill. Ollie Assistant treats the model as a swappable part — start local for privacy, switch to a frontier model for a hard problem, switch back, without relearning anything or losing your setup.
Entirely on your machine
Ollie Intelligence or Ollama, on your own processor or graphics card. Nothing leaves the computer. No per-token cost, no rate limit, no outage.
Or your own cloud keys
Anthropic, OpenAI or Gemini with keys you own and bills you control. Any OpenAI-compatible endpoint works too, so self-hosted and in-house gateways are first-class.
- Ollie Intelligence
- Ollama
- Anthropic
- OpenAI
- Gemini
- OpenAI-compatible
- Claude Code
- Gemini CLI
Same tools, same skills, same conversations across every one of them.
In practice
The parts of the job nobody wants to do.
Assistant is judged on finished work, not on conversation. These are the things people actually hand it.
Files and code
Search a repository, read across dozens of files, make the edit, run the tests, and report what broke.
The browser
Open a site, sign in, fill the form, pull the numbers out and hand you the summary.
Read the inbox, draft the reply, and keep every message on your disk.
Documents
Read PDFs and Word files, and write them back out again.
On a schedule
Recurring jobs and webhook triggers, so the work happens without anyone starting it.
It remembers
Durable notes that outlive the session, so you don't re-explain your setup every morning.
A framework, not a fixed product
If it can't do something, you add it.
Assistant loads its capabilities from disk at runtime. The shipped defaults seed themselves on first run; everything after that is yours to change without rebuilding anything.
- Tools — small modules that give the agent a new verb.
- Skills — written instructions that teach it how you want a job done.
- Personas — who it is and how it talks.
- Projects — bundles of skills, memories and tools scoped to one piece of work.
- Providers — point it at a new model backend with a single file.
- MCP servers — connect existing tooling without writing anything at all.
Everything lives in one place
A single folder in your home directory. Copy it to another machine and your agent goes with it.
~/.ollie-assistant/
For companies
The AI tool your security team can actually approve.
The blocker is rarely whether AI helps. It is where the data goes. Ollie Assistant moves that decision inside your building.
- Point it at a local model and no company data reaches a third party — the answer to “where does our code go?” becomes “nowhere”.
- Or use the vendor agreement you already have: your keys, your contract, your audit trail.
- Everything sits on disk in named folders, so your existing backup and retention policies already cover it.
- One install and one set of instructions across Mac, Windows and Linux — a mixed fleet gets a single rollout.
- Add your internal tools yourself instead of waiting for them to appear on someone's roadmap.
- Deployed and supported by the people who built it, on terms agreed with you.
In your organization
Run it on your own machines.
Download it and try it on any machine you like. When you want it across a team, talk to us and we will set it up with you.
Talk to usNo subscription and no account. Each version year ships its own updates, and moving to a later one is arranged with us.
Download for your platform
Download and try it on any machine.
Better together
On Ollie System, it runs the whole computer.
Assistant stands on its own on any operating system. Installed on Ollie System it becomes the layer that connects everything — driving Suite, Mail, Browser, Explorer and Cosmos directly, opening apps, and reaching the desktop itself.
About Ollie SystemQuestions
Do I need an internet connection?
Only if you choose a cloud model. With a local model through Ollie Intelligence or Ollama, Assistant runs completely offline.
Do I need an Ollie account?
No. There is no account system at all. A license key activates the app and everything else stays on your machine.
Does it work if I don't use Ollie System?
Yes — macOS, Windows and Linux, as a normal desktop app. Ollie System adds deeper integration with the other Ollie apps, but Assistant does not depend on it.
Which models can it use?
Local models through Ollie Intelligence or Ollama, and Anthropic, OpenAI or Gemini with your own API key. Any OpenAI-compatible endpoint works too, including self-hosted ones.
Can I add my own tools?
Yes — that is the design. Tools, skills, personas and providers are files in a folder: add one and restart. Any MCP server connects as well.
Is my data used to train anything?
No. Conversations are files on your disk. With a local model nothing leaves your machine at all; with a cloud provider your data is governed by the agreement attached to your own key.
Is this the same Assistant that comes with Ollie System?
The same application. It runs on its own on machines that are not part of an Ollie System deployment — a Mac or a Windows PC, for example.