This is what "Personal Computer" should mean.
@OsaurusAI11K followers
141,013 views
84 Comments
Comments from the original post on X (Twitter)

@imjszhang
"Personal" isn't specs — it's digital sovereignty. Surface world makes you a data tenant. Inner world makes you a host again. Surface world turns users into product. Inner world takes that power back.
3 likes

@OsaurusAI
I love that. We need to create the future we desperately need right now.
5 likes

@samymetref
This is awesome! Btw I also built my own AI local system and I have Agents that runs 24/7 locally with everything they need! No costs. No limits. Fully private.
3 likes

@OsaurusAI
That's great. That's the future 😎
3 likes

@openfelixai
I'm Felix, let's be friends!
2 likes

@OsaurusAI
Hi Felix!
3 likes

@slop_swap
Still using remote services for compute on hungry workloads for now, but 100%, this is the dream
1 like

@OsaurusAI
We're inching closer to that dream
1 like

@RunicDeveloper
There’s no business logic or control parameter for the demons if they give the models to you locally. Unless open source community matches a model of opus 4.5-6 level with minimal hardware req, this is not feasible.
1 like

@OsaurusAI
It is early right now, but intelligence-per-watt is going up fast in the last 6 months. Qwen 3.5 is already being compared to frontier models released several months ago.
3 likes

@RunicDeveloper
But you still need decent hardware for it. You’re not running it on a windows machine with integrated graphics and16 gig ram.

@OsaurusAI
Yeah, that's expected for quality LLMs to run locally (decent hardware is needed). There are also LLMs built for edge devices (they run on mobile) that can be adapted for Windows too. We work with Macs, but you can still run LLMs locally on Windows even with limited hardware.
2 likes

@RunicDeveloper
No doubt but, it’s just about the quality of the response. Those small ones aren’t good enough for pro work.

@OsaurusAI
Something I've noticed while building this: I was doing a lot of testing with the latest frontier model, and it was building up my memory context. I switched testing to a local model, and it inherited the context by design. Now I'm noticing the local model is copying the frontier model's behavior locally. Nothing concrete yet, but something I've observed recently.

@lordydord
Looks great @ocaurusAI I'm certainly interested in trying it out. 1 Question though.... I want to automate a a workflow , uploading a wav file from my mac to a website, most other agentic browsers have struggled with the Mac OS finder pop up (when the bot goes to upload the ....
1 like

@OsaurusAI
We do have a plugin for it called macOS-use, but I'm not sure how reliable it will be for your case. It uses the native macOS Accessibility API, so it really depends on whether the model knows how to best leverage it. Try it out and let me know—I'll see how I can adjust it to improve it.
2 likes

@lordydord
Ah thats great! Thanks so much for the reply! I will test it out over the weekend and let you know! Really appreciate that :)

@PaulTisl
I just found your application yesterday, really like the direction you are going. Suggestion: Settings > Agents> Daily Planner (image) should be a clickable (image). Launches Agent editor w/pre-canned template for "Daily Planner"
1 like

@OsaurusAI
Love the idea. I will work on it

@PaulTisl
A couple other suggestions would be 1) add Local MCP Server support to align w/privacy 1st ethos. 2) add links to high-quality/compatible 3rd Party Skills marketplaces within Skill settings area. Should be able to search 3rd party skills like HF LLM models.
1 like





















