With the latest update to mistral.rs, you can now freely run 'Agent Skills'—advanced task execution capabilities—on your personal computer using open-source AI models, without external help.
Imagine this: You wake up in the morning and tell your personal AI assistant, “Organize today’s meeting notes and email them to me.” Until now, for your assistant to perform such a task, it absolutely required an AI model from a giant corporation running on a massive server in the cloud. However, a path has now opened for that assistant to live solely inside your laptop, working more freely and intelligently. This is because the recent update to an AI execution tool called ‘mistral.rs’ allows us to directly teach our computer’s AI ‘professional skills.’
Why It Matters
Until now, to have an artificial intelligence perform sophisticated tasks, you mostly had to rely on ‘Closed Models’ (AI whose internals cannot be inspected without corporate permission) provided by giants like OpenAI or Anthropic. This meant that the content of your work had to be sent to an external server, which was a major concern for users sensitive to security or personal data.
But with this update, we can now run ‘Agent Skills’—advanced task processing techniques—even on ‘Open Models’ (AI that anyone can modify and execute) installed directly on our devices [Source 1, Source 10]. This means an environment has been created where you can maintain strict security without sending data to external servers while building your own powerful AI agent [Source 9].
The Explainer
The concept of ‘Agent Skills’ might feel a bit difficult. Let’s use an easy analogy. Suppose we have hired a very smart new employee. This employee is inherently bright, but they know nothing about our company’s complex document processing methods or how to use specific software. Handing them an ‘operations manual’ is exactly what ‘equipping a skill’ is like.
In short, Agent Skills are ‘procedural knowledge’ that tells an AI exactly how to perform a specific task [Source 4]. The updated mistral.rs allows you to hand these files, containing these skills, to the AI like puzzle pieces; the AI reads them and immediately performs the task [Source 3]. Because it follows existing OpenAI standard technology, it is now much easier to utilize over 1.7 million agent skills already out in the world in a local environment [Source 6].
Where We Stand
The developer maintaining mistral.rs stated that with the v0.8.10 update, these skills, which were previously trapped in specific corporate models, can now be fully brought to personal local devices [Source 8, Source 13]. Users simply need to upload skills in a compressed file format or deliver them as a directory structure [Source 3]. We have reached a level where, through local open models like Gemma, you can operate your own professional AI assistant without passing through the servers of giant corporations [Source 9].
However, it is important to remember that processing speed or accuracy may vary depending on the performance of the local model and your computer’s hardware specifications. Compared to the massive computing power of cloud servers, hardware limitations still exist on individual devices.
What’s Next
In the future, creating ‘your own expert living on your computer’ will become much more mainstream. Not just developers, but general users will be able to optimize their own work by creating skill files for repetitive tasks they perform frequently and inputting them into the AI. Efficient skills created by someone else are already overflowing on platforms like GitHub and various skill marketplaces [Source 7]. Now, you just need to find and install the skills that suit your taste. Artificial intelligence technology is moving into smaller, more efficient personal devices.
MindTickleBytes’ AI Reporter Perspective
If AI technology has been concentrated in the data centers of giant corporations until now, we have entered an era where that capability can be freely extended on individual devices. When the sharing of tools and the open-source ecosystem combine, artificial intelligence will no longer be ‘someone else’s technology,’ but ‘my assistant.’
References
-
[ShowHN:RunAgentSkillswithmistral.rsv0.8.10… Hacker News](https://news.ycombinator.com/item?id=48581792) -
[Mistral.rsv0.8.10: запуск агентных скиллов через /v1/skills AiManual](https://ai-manual.ru/article/obnovlenie-mistralrs-v0810-kak-zapuskat-agentnyie-skillyi-cherez-v1skills/) -
[OpenAI-compatibleSkills mistral.rs](https://ericlbuehler.github.io/mistral.rs/guides/agents/skills/) - Discover and installskillsfor AIagents.
- GitHub - EricLBuehler/mistral.rs: Fast, flexible LLM inference · GitHub
-
[AgentSkillsMarketplace - Claude, Codex & ChatGPTSkills SkillsMP](https://skillsmp.com/) - DiscoverAgentSkills
-
[Nuxt HN Run Agent Skills with mistral.rs v0.8.10: /v1 …](https://hn.nuxt.dev/item/48581792) - Mistral.rs v0.8.10 Adds Local Agent Skills Support
- Show HN: Run Agent Skills with mistral.rs v0.8.10: /v1/skills …
-
[mistral.rs mistral.rs](https://ericlbuehler.github.io/mistral.rs/) - Show HN: Run Agent Skills with mistral.rs v0.8.10: /v1/skills …
- Show HN: Run Agent Skills with mistral.rs v0.8.10: /v1/skills …
- Addition of web search functionality
- Support for local execution of OpenAI-compatible agent skills
- Doubling of AI model compression
- AI's emotional expression capability
- Reusable capabilities that provide procedural knowledge needed by AI
- Algorithms for training AI models
- Because it costs more
- Because it enables personalized local AI without cloud models
- Because it makes games run faster