What Is Osaurus? Local AI Agents That Run on Your Own Computer

Osaurus is a desktop application that runs AI agents locally on your Mac or Windows PC instead of in a cloud service. You download it, point an agent at files on your machine, and the work — reading documents, drafting edits, summarizing notes — happens on your computer. If your main concern is keeping sensitive files off third-party servers, that's the problem Osaurus is built to solve. If you need frontier-model reasoning on very large or complex tasks, a local app is a different tradeoff, and you should weigh that before committing.

The core idea: agents that live on your machine

The site's framing is direct: "Powerful AI agents that live on your computer, not in someone else's cloud. Whatever you use them for, it stays with you." The tagline is "Privacy isn't extinct."

Practically, that means the app reads and writes files in folders you grant it access to, and the agent's work product stays in those local locations. The homepage example shows an agent reading MSA-v3.docx and Playbook.md, flagging a conflict, and proposing a tracked edit — all described as happening on the Mac.

What you can actually do with it

The homepage shows two concrete agent workflows.

Contract review and redlining

A "Legal Helper" agent is shown reviewing a Master Services Agreement:

• It reads the contract and a separate playbook file.
• It identifies that §9 (uncapped liability) conflicts with a rule in the playbook.
• It proposes a specific edit: changing "Liability is uncapped for all claims" to "Liability is capped at twelve months of fees."
• The change is approved and tracked in Word.

This is a useful illustration of the pattern: the agent compares a document against your own rules file, rather than against a generic standard.

Summarizing notes into a handoff

A "Practice Helper" agent is shown turning a week's intake notes into a handoff document:

• It reads three files in a Practice/Intake folder.
• It pulls out symptoms, medication changes, and follow-ups.
• It produces handoff-summary.md with counts (3 patients, 1 medication change, 2 follow-ups due Friday).

The takeaway is that Osaurus is aimed at document-heavy, file-based work where the inputs already sit on your disk.

Platform and availability

The Windows build is explicitly labeled beta, so if you're on Windows, expect rougher edges than on Mac.

The model: Raptor

The interface references "Raptor 0.6," and the site has a section titled "Meet Raptor 4B." That naming suggests a compact local model (the "4B" implies roughly 4 billion parameters). A model that size can run on a personal computer, which is what makes the local-only approach feasible — but it also means the reasoning ceiling is lower than a large cloud model. For structured, file-grounded tasks like the contract and intake examples, that's often enough. For open-ended or highly complex reasoning, it may not be.

Who this fits — and who it doesn't

A good fit if:

• Your work involves confidential documents (legal, clinical, financial) that you'd rather not upload.
• Your tasks are file-based: review, compare, summarize, draft edits.
• You want the output to land in local files you already use.

Probably not the right fit if:

• You need the strongest available reasoning on hard, novel problems.
• You rely on cloud integrations or collaboration features.
• You're on Windows and need a stable, non-beta experience.

What to check before downloading

The site links to a pricing page, but the homepage itself doesn't state cost, licensing terms, or whether a login is required. Those details aren't confirmed here, so verify them on the pricing page before you commit. Also confirm the hardware requirements for running Raptor locally — a 4B-class model needs a reasonably capable machine, and the site doesn't spell out minimum specs in the material available.

The clearest way to decide: if your priority is keeping sensitive files on your own machine and your tasks are document-centric, Osaurus is designed exactly for that. If your priority is maximum reasoning power regardless of where it runs, look elsewhere.