The native macOS harness for AI agents. Any model, persistent memory, autonomous execution, cryptographic identity. Built in Swift. Fully offline. Open source.
Replies
Best
Love that Osaurus is Mac-native and fully offline, running open source AI locally means your data never leaves your machine which is a refreshing take on privacy-first computing.
@ilko_kacharov Thank you for your feedback! We believe privacy means freedom, and users should have full ownership of their AI. Our principles guide us when we're building Osaurus.
Report
the approval-gate-before-every-action bit is what i'd actually trust daily. same constraint on our end, on-device voice checkin, nothing leaves the phone. does the approval prompt get less frequent as it learns what you always say yes to, or is every action a fresh ask forever?
@sabber_ahamed You have full control over it, but you can also choose "Always Allow" which will skip next time. I would suggest for destructive actions, do not set "Always Allow"
Congrats Terence! Approval before each action is the detail that won me over, that’s the exact rule I enforce on every agent I build at work. My question is about the floor: what’s the smallest Mac this genuinely works on for agent tasks? Chat is easy for local models, but tool calling and multi-step work is usually where small models fall apart. Curious what model size you’d call the honest minimum for Work Mode on, say, a 16GB M2.
@ridhwikvinod Thank you! I would say 16GB would fit Gemma 4 e4b, which is not the most smartest but is pretty reliable at tool calling. For best experience, I would suggest 24GB.
Report
the fully-offline bit is the whole selling point for me tbh... most agent tools are just api wrappers that fall over the second the wifi does. running local on the mac with persistent memory is an actual edge, not a checkbox... swift instead of yet another electron thing is a nice touch too.
@alex_watson2110Â Exactly. Electron is purely for companies to expand their reach, for Mac users, we're getting the shorter end of the stick. Mac users deserve better. That's why we built it in Swift.
Report
The Mac-native angle is the part that stands out. Ollama works, but it always feels like it is renting space on the machine rather than living on it. What I would want to know is how model management holds up once people start pulling in the bigger local models, is there a clean way to swap them without eating all the RAM. Either way, local-first AI getting a friendlier front door is good for the whole space.
@liana_preston Thank you for the feedback, totally agree that local-first AI needs to get more attention. As AI enthusiasts, we need more choices on how to use our AI, and in my opinion, owning your AI is the best option. We allow you to own your AI without having to rent them.
Our harness has been tuned to work with local models efficiently, if you have a larger mac, we support multi model residency (allowing you to warm up multiple models at a time), but if you have a smaller one, we make it efficient to load and unload them.
Our core idea is that harness is what makes local AI useful. We spent our time optimizing our harness for local models, and it's capable of doing complex agentic tasks.
Report
Congrats on the launch! With the approval gate on every action, does that mean autonomous runs stop and wait a lot, or is it smart about only flagging the risky stuff like sending a message or running code?
@irahimiam It's about having control, and you will be asked each step, but you can always choose "Always Allow" for specific actions you feel comfortable with.
Report
Downloading to check this out now! Just the branding of the dinosaurs is unique enough to pique my interest, would love to know where you got the idea to use them. Make the internet fun again!
My son was getting into dinosaurs (he's turned 3), and I wanted to build something fun with AI. I built Dinoki (pixel dinosaur desktop AI companion), and realized there's a growing need to run local models (for costs and privacy). Dinoki will be central to what we do, so when we built Osaurus, we decided to keep the same branding.
Report
Local-first agents is the direction I keep hoping wins. The cryptographic identity part is interesting — most agent frameworks treat identity as an afterthought. Curious how heavy it gets memory-wise with a couple of agents running on an M-series Air?
@alex_tomilin How big is your Air? I would suggest having minimum RAM of 16GB, but 24GB is recommended. Depending on the model you choose but it could take up anywhere between 8GB ~ 12GB RAM during active usage.
Report
the approval gate plus "always allow" per action is the right default, but the part I'm actually curious about is the open source contribution angle - if community members start shipping shareable agent skills/plugins that get access to iMessage or Contacts, is there any review step before those get trusted, or is it on the user to audit what a downloaded skill can touch before installing it
@galdayan We accept contributions but we make sure it's strict to our code standards. We have artifacts that allows agents to run through series of checks before making the PR. We set our standards high for developer contributions
Replies
Love that Osaurus is Mac-native and fully offline, running open source AI locally means your data never leaves your machine which is a refreshing take on privacy-first computing.
Osaurus
@ilko_kacharov Thank you for your feedback! We believe privacy means freedom, and users should have full ownership of their AI. Our principles guide us when we're building Osaurus.
the approval-gate-before-every-action bit is what i'd actually trust daily. same constraint on our end, on-device voice checkin, nothing leaves the phone. does the approval prompt get less frequent as it learns what you always say yes to, or is every action a fresh ask forever?
Osaurus
@sabber_ahamed You have full control over it, but you can also choose "Always Allow" which will skip next time. I would suggest for destructive actions, do not set "Always Allow"
DiffSense
Third and final question 🙏 Does Osaurus have a cli input? Like is it possible to invoke an action via cli to it? If you know what I mean?
Osaurus
@conduit_design Osaurus does have CLI input! You can read more about it here: https://docs.osaurus.ai/cli
You can invoke actions using our MCP server.
Osaurus
@ridhwikvinod Thank you! I would say 16GB would fit Gemma 4 e4b, which is not the most smartest but is pretty reliable at tool calling. For best experience, I would suggest 24GB.
the fully-offline bit is the whole selling point for me tbh... most agent tools are just api wrappers that fall over the second the wifi does. running local on the mac with persistent memory is an actual edge, not a checkbox... swift instead of yet another electron thing is a nice touch too.
Osaurus
@alex_watson2110Â Exactly. Electron is purely for companies to expand their reach, for Mac users, we're getting the shorter end of the stick. Mac users deserve better. That's why we built it in Swift.
The Mac-native angle is the part that stands out. Ollama works, but it always feels like it is renting space on the machine rather than living on it. What I would want to know is how model management holds up once people start pulling in the bigger local models, is there a clean way to swap them without eating all the RAM. Either way, local-first AI getting a friendlier front door is good for the whole space.
Osaurus
@liana_preston Thank you for the feedback, totally agree that local-first AI needs to get more attention. As AI enthusiasts, we need more choices on how to use our AI, and in my opinion, owning your AI is the best option. We allow you to own your AI without having to rent them.
Our harness has been tuned to work with local models efficiently, if you have a larger mac, we support multi model residency (allowing you to warm up multiple models at a time), but if you have a smaller one, we make it efficient to load and unload them.
Our core idea is that harness is what makes local AI useful. We spent our time optimizing our harness for local models, and it's capable of doing complex agentic tasks.
Congrats on the launch! With the approval gate on every action, does that mean autonomous runs stop and wait a lot, or is it smart about only flagging the risky stuff like sending a message or running code?
Osaurus
@irahimiam It's about having control, and you will be asked each step, but you can always choose "Always Allow" for specific actions you feel comfortable with.
Downloading to check this out now! Just the branding of the dinosaurs is unique enough to pique my interest, would love to know where you got the idea to use them. Make the internet fun again!
Osaurus
@rick_mcclelland Let's make it fun!
My son was getting into dinosaurs (he's turned 3), and I wanted to build something fun with AI. I built Dinoki (pixel dinosaur desktop AI companion), and realized there's a growing need to run local models (for costs and privacy). Dinoki will be central to what we do, so when we built Osaurus, we decided to keep the same branding.
Local-first agents is the direction I keep hoping wins. The cryptographic identity part is interesting — most agent frameworks treat identity as an afterthought. Curious how heavy it gets memory-wise with a couple of agents running on an M-series Air?
Osaurus
@alex_tomilin How big is your Air? I would suggest having minimum RAM of 16GB, but 24GB is recommended. Depending on the model you choose but it could take up anywhere between 8GB ~ 12GB RAM during active usage.
the approval gate plus "always allow" per action is the right default, but the part I'm actually curious about is the open source contribution angle - if community members start shipping shareable agent skills/plugins that get access to iMessage or Contacts, is there any review step before those get trusted, or is it on the user to audit what a downloaded skill can touch before installing it
Osaurus
@galdayan We accept contributions but we make sure it's strict to our code standards. We have artifacts that allows agents to run through series of checks before making the PR. We set our standards high for developer contributions