简体中文
English
Website review
This review may include AI-generated content and is not individually verified. Domain, certificate and other facts come from the observed data on this page.
What is Osaurus?
Osaurus is a desktop AI agent that runs locally on your own computer rather than in a vendor's cloud, so the files and text you work with stay on your machine. Its tagline, "Privacy isn't extinct," frames the core pitch: capable AI assistance without shipping your documents to a remote service. It offers downloads for Mac and Windows (Windows labeled Beta) and ships with a local model called Raptor (shown as Raptor 0.6, with a "Meet Raptor 4B" section).
What it actually does
The page demonstrates agents working over real local files. In one example, a "Legal Helper" agent reads a master services agreement and a playbook, spots that the liability clause conflicts with internal rules, and drafts a tracked redline in Word that you approve or deny. In another, a "Practice Helper" reads several intake notes and produces a handoff summary for a care team. Agents appear to be organized into projects (Legal Helper, Contracts, Practice Helper, Books Helper, Research Helper) with settings for each.
- Local execution: the interface shows a "Local" badge and "Nothing left this Mac," reinforcing that processing happens on-device.
- File-grounded work: it reads and edits documents you point it at, such as .docx and .md files, and can write changes back with tracked edits.
- Approval step: suggested edits are presented for Approve or Deny before being applied.
- Visible reasoning: it shows brief "thinking" steps and which files it read, which helps you sanity-check its work.
Who it suits, and the trade-off
This is aimed at people handling sensitive material — legal, clinical, research, or general confidential documents — who want AI help without uploading files to a third party. The trade-off is the usual one for local models: you get privacy and offline-friendly operation, but on-device models like Raptor 4B are generally less capable than the largest cloud models, and performance depends on your hardware.
Next step
If privacy is your deciding factor, test it on a real, non-critical task first: point an agent at a sample contract or notes file, review its proposed edits, and judge whether the local model's quality is good enough for your workflow. Check Osaurus for current platform support and pricing details.
How does Osaurus keep my data private compared to cloud-based AI tools?
Osaurus keeps data private by running its AI agents locally on your own computer rather than in a remote cloud. According to its site, the agents "live on your computer, not in someone else's cloud," and whatever you use them for "stays with you." The page illustrates this with a local model called Raptor (shown as Raptor 0.6 and Raptor 4B) working on files such as an MSA document and practice intake notes, with the interface noting "Nothing left this Mac." That is the core difference: processing and file handling happen on your machine, so sensitive content does not need to be uploaded to a third-party service to be analyzed.
How that differs from cloud-based tools
| Aspect | Osaurus (local, per its page) | Typical cloud AI tool |
|---|---|---|
| Where processing happens | On your computer | On remote servers |
| What leaves your device | Described as staying with you | Prompts and attached files are sent to the provider |
| File access | Works with local files like MSA-v3.docx and intake PDFs | Usually requires uploading files |
| Data retention | Framed as staying on your Mac | Governed by the provider's policies |
| Setup | Download for Mac, Windows in beta | Usually browser-based, minimal setup |
| Trade-off | Depends on your hardware and local models | Depends on internet connection and provider limits |
A concrete reader scenario
A solo legal consultant reviewing a client's master services agreement could ask the agent to check a liability clause against an internal playbook and draft a redline. With a cloud tool, that contract and playbook would typically be uploaded to someone else's servers. With a local setup like the one shown, the work happens against files already on the machine, which matters when confidentiality obligations or client rules restrict third-party processing.
What to weigh
Local processing is not automatically better for every task. Very large models, heavy multitasking or weaker hardware can make local agents slower or less capable than cloud services, and you become responsible for your own backups, security and model updates. If your work involves confidential documents, regulated data or simply a preference for keeping files off external servers, local agents are worth testing. If you need maximum model capability or work from many devices, a cloud tool may still fit better.
A practical next step: pick one non-critical but sensitive document, run the same task locally and in a cloud tool you already use, and compare output quality, speed and how much of the file you are comfortable sending off-device. The download is available for Mac, with Windows in beta, per the site. You can also check Osaurus for current details.
Can Osaurus run on my Mac or Windows computer, and what are the system requirements?
Yes. Osaurus is offered as a download for both Mac and Windows, so it is not tied to a single desktop platform. The Windows build is labelled Beta on the page, which is the main practical difference between the two options: Mac users are downloading the standard release, while Windows users are opting into an early build that may change more often.
What the page actually confirms
- Platforms: Mac and Windows downloads are listed side by side.
- Windows status: marked Beta.
- Adoption signals: the page cites 232.9k downloads and 7.8k stars, which suggests an active user base rather than a brand-new experiment.
- Where the AI runs: the core claim is that agents "live on your computer, not in someone else's cloud," so the work happens locally rather than on a remote service.
System requirements
The page does not publish a minimum macOS or Windows version, processor, RAM or disk-space figure. Rather than guess, check the download page or the project's documentation for the exact requirements before installing, since local model inference is usually the part that dictates hardware.
How to decide
Local execution is the trade-off to weigh. Running models on your own machine is what keeps documents such as contracts and intake notes on your device, as the page's examples show. The counterweight is that performance depends on your hardware, and a Beta build on Windows is more likely to hit rough edges than the Mac release. If your machine is modest, expect smaller models to be the realistic option; if you mainly care about privacy for sensitive files, that trade-off is often worth accepting.
A sensible next step: download the build for your platform, then test it on one non-critical document before pointing it at anything confidential.
What kinds of tasks can Osaurus agents handle for legal, medical, or research work?
Osaurus agents are presented as local, file-aware assistants that read documents on your machine, reason over them, and draft or edit outputs in place. The page’s own examples point to three concrete families of work.
Legal and contract work
The sample session shows an agent reviewing a Master Services Agreement against an internal playbook, spotting a conflict, and proposing a tracked redline in Word — specifically flagging that an uncapped liability clause conflicts with a policy requiring a twelve-month cap, then editing the clause and recording the change as tracked. That maps to everyday contract review: checking clauses against a standard position, drafting redlines, and keeping an audit trail of what changed.
Medical and care coordination
A second example turns a folder of intake notes into a handoff summary for a care team, extracting symptoms, medication changes, and follow-ups from several files. The practical use is administrative summarization and handoff preparation — pulling structured facts out of unstructured notes — rather than diagnosis or clinical decision-making.
Research and document triage
The interface shows helper agents for contracts, practice, books, and research, plus a visible set of source files (PDFs, Markdown notes, Word documents). That suggests literature triage, cross-referencing sources, and drafting summaries grounded in a defined set of local documents.
What the design implies, and where the trade-off sits
Because the agents run on your computer and work against local files, the appeal is confidentiality: sensitive material stays on the machine. The trade-off is capability. The page names a compact model (Raptor 4B), which suits extraction, clause-checking, and summarization against a supplied playbook, but is a weaker fit for open-ended legal analysis or complex multi-source reasoning than larger cloud models. If your work is bounded and document-centric, local is a strong match; if it requires broad world knowledge or heavy reasoning, expect to verify more.
A practical test before committing: pick one recurring task — a liability-clause check against your playbook, or an intake-to-handoff summary — run it on real files, and judge whether the draft needs light editing or full rewriting. That single trial tells you more than any feature list.
Does Osaurus offer a free version or paid plans, and what do they cost?
Osaurus separates the software you download from whatever paid plans may exist: the site presents a Mac download (with a Windows beta) and a Pricing page, but the page evidence provided does not list any plan tiers or amounts. So the honest answer on cost is that a downloadable version exists and pricing is published separately — the specific figures aren't part of what's shown here.
What you can tell from the product itself
The free/local angle is structural, not just a tier. The agents run on your machine: the demo shows Raptor 4B reading MSA-v3.docx and Playbook.md, flagging that §9 conflicts with the playbook, and editing the document with tracked changes — all labelled "Local" with a note that nothing left the Mac. A second example turns intake notes into a care-team handoff from files in a local folder.
That matters for cost in a practical way. If inference happens on your hardware, you aren't paying per query to a cloud provider, and you aren't sending client contracts or patient notes to a third party. The trade-off is the usual one for local models: a smaller model such as Raptor 4B is well suited to structured, document-grounded tasks like redlining against a playbook or summarising notes, but it will not match the largest hosted models on open-ended reasoning. You also carry the hardware requirement.
How to decide
- If your blocker is confidentiality — law firms, clinics, anyone handling contracts or patient data — evaluate the local workflow first and treat any paid tier as an optional extra.
- If your blocker is raw capability on hard reasoning tasks, test the local model on your own documents before assuming it replaces a hosted assistant.
- If you need team administration, shared seats or central billing, check the Pricing page directly, since those features typically sit behind paid plans rather than the free download.
A concrete next step: download it, point one agent at a real folder of contracts or notes, and run a task you already know the right answer to. That tells you more about whether the free version is sufficient than any feature list.
How do I install and set up Osaurus to use local AI agents on my own files?
Download Osaurus from Osaurus and run the installer for your platform — the page offers a Mac build and a Windows Beta — then point an agent at a folder on your machine and let it read and edit those files locally. The setup model shown on the page is: pick an agent (Legal Helper, Practice Helper, Books Helper, Research Helper), attach or reference files, type a task, and approve or deny the proposed edits before they're written.
What the setup actually involves
The page's walkthrough suggests the working loop rather than a long configuration wizard:
- Install the app for macOS or Windows Beta.
- Choose or create an agent from the sidebar (the demo shows a "Practice Helper" and a "Legal Helper" under Agents, with Projects listed separately).
- Point it at local material. In the examples, the agent reads named files such as
MSA-v3.docxandPlaybook.md, or a whole folder like~/Practice/Intakecontaining an intake PDF and session notes. - State the task in plain language — "Check §9 against our playbook and draft a redline," or "Turn this week's intake notes into a handoff for the care team."
- Review the proposed change. The interface shows a diff (
− Liability is uncapped…/+ Liability is capped at twelve months of fees.) with Approve and Deny buttons, and notes the result was "tracked in Word." - Confirm where the work stayed. The status line reads "Raptor 0.6 · Local" and "Nothing left this Mac."
The privacy trade-off
The pitch is that agents "live on your computer, not in someone else's cloud," and the demo reinforces this with a local model label and a "nothing left this Mac" indicator. That is the main reason to choose this over a browser-based assistant for contracts, clinical notes or client files: sensitive text doesn't need to be uploaded to a third party.
The cost is capability and effort. A local setup is bounded by your own hardware, and the page's own framing — a "Raptor 4B" model — points to a compact model rather than a frontier one. Expect strong results on structured, repetitive work like extracting symptoms, medications and follow-ups from notes, or flagging a clause against a playbook rule; expect weaker performance on open-ended reasoning or very long documents. You also carry the setup work yourself.
A practical first run
Start with one narrow, low-risk task in a single folder — for example, summarising meeting notes into a fixed template — and check the output against the source files by hand. Once the results hold up, widen to tasks where the agent proposes edits you review, such as redlining a clause. Keep the approve/deny step as a real gate rather than a formality, especially for anything legal, medical or financial.
If your work involves material you would not paste into a cloud chat, this local-first design is the deciding factor. If you mainly need the strongest possible reasoning on messy, long inputs, compare against cloud-based options before committing. Pricing details are listed at Osaurus if you want to check tiers before installing.
Vercel
Google Workspace
User reviews (0)