Parenting kids 👶 and agents 🤖 @tuikagents that reduce costs by up to 10x.
Visit my newsletter →
Joined September 2014
- Tweets2K
- Following606
- Followers1K
- Likes1.8K
Last year, I was building a RAG system for a client, and we ran into a real engineering challenge.
We were extracting text from PDFs, but we also needed to extract images so that, during retrieval, we could return the relevant information along with its corresponding images.
The PDFs contained tutorials, so linking the right images to the right pieces of text was crucial.
I had to come up with a fairly complex approach to extract and associate those images with the relevant content.
Fast-forward to 2026, and now we have Optical "Everything" Recognition (OXR) by @somarkai.
This would have made things so much easier back then.
Add real document parsing to your AI agents.
Character recognition is good.
But "everything" recognition is better.
That's what OXR by @somarkai does.
It extracts text, images, LaTeX, and more from your PDF files.
All data becomes agent-friendly.
It's also fully open-source.
I'm tired.
I'm tired of reading AI-generated docs.
I'm tired of seeing messages from bots.
I'm tired of seeing blogs full of AI-slop articles.
There are so many things you can use AI for. Why writing?
Yes, use it for brainstorming.
Yes, use it for research.
Yes, use it to draft.
But please do not use it to replace your SOUL.md.
That's the reason people read your stuff in the first place.
hermes plugins install advanced-delegation
One agent. Three subagents. Three LLMs.
@NousResearch just introduced Advanced Delegation in the Hermes Agent harness.
You can spawn subagents, and they will run with specific models or profiles.
Route by model:
If you have a skill for writing, another for browser automation, and another for coding. They can all run with their specific model through a subagent.
Route by profile:
You can call other profiles to execute tasks as subagents with their own skills and expertise.
Creating a new Hermes profile just because you want to use a different model for specific skills?
Forget that.
You can associate each skill with a subagent, which has its own LLM.
(Oh, and Grok Bot can't do this)
LinkedIn is a gossip social media now.
If you don't say "My mom", "My wife" or "My daughter" your post is dead.
Which one is your favourite?
You’re using Claude for coding.
What about the other 99 use cases?
Claude, Codex, Cursor, and Grok can’t ship tailored AI agents for every use case.
They also come at a cost.
Open-source harnesses, on the other hand, are free and offer unlimited customization.
Here are 5 worth exploring, along with their strengths, weaknesses, and use cases 👇
Grok Bot offers ease of use, but it is limited.
1. You won't be able to connect your bots to other gateways like Slack, Telegram, and WhatsApp.
2. You're not in control of the LLMs, the agents + the tools.
3. You're giving all your personal and professional data to xAI.
If that's ok for you, stick with Grok Bot, indeed.
But if you need custom AI agents, open-source solutions are the way, and Hermes is probably the easiest of them.
Every day, a new cool plugin 😅
This is great for my accounting.
I am very happy to present `hermes record`, a Hermes Labs plugin built by Nous Research.
it records what you do in your browser, creates a skill and Hermes can then repeat the task with whatever details you give it.
in this video, I record my pizza order, then use /learn to make a skill out of that recording, then ask my Hermes Agent to place a different order on its own :)
'hermes plugins install workflow-record'
AGI.
If you still don't know what the right open-source harness for you is, you can find out more about the top 5 OSS harnesses in this article 👇
xitter.cf/dadhalfdev/status/2113…
5. Control the Desktop from your phone
@NirvanaJie30 create a plugin for the DeepSeek Harness.
You can find the repo here:
github.com/Nirvana-Jie/dsh-l…
4. Pi in Rust
@lassejlv migrated the full Pi Agent into Rust programming language.
The result is a coding agent with less than 90 MB of memory usage and a much faster startup.