Nate Richardson
@naterichardson.com
Building Angular at scale.
Shipping in public.
Breakfast enthusiast. Dad. Nerd. Baker.
Shipping in public.
Breakfast enthusiast. Dad. Nerd. Baker.
February 6, 2026 at 12:13 AM
Quick note on why I went with the BSL for this project.
I do have a longer-term vision of building a service around this bot architecture. The viability would come from ease of use and baked-in security.
The beautiful thing is: if you want to self-host, you still can. OSS at its core.
I do have a longer-term vision of building a service around this bot architecture. The viability would come from ease of use and baked-in security.
The beautiful thing is: if you want to self-host, you still can. OSS at its core.
February 6, 2026 at 12:13 AM
Quick note on why I went with the BSL for this project.
I do have a longer-term vision of building a service around this bot architecture. The viability would come from ease of use and baked-in security.
The beautiful thing is: if you want to self-host, you still can. OSS at its core.
I do have a longer-term vision of building a service around this bot architecture. The viability would come from ease of use and baked-in security.
The beautiful thing is: if you want to self-host, you still can. OSS at its core.
Now on to the topic of tools. How do you keep the tool layer secure? I think the solution is isolation.
So rather than your tool calls breaking or freezing the gateway, each tool is subscribed to a NATS topic and runs as a microservice.
- Isolation (no crashing the gateway)
- Scaling (per tool)
So rather than your tool calls breaking or freezing the gateway, each tool is subscribed to a NATS topic and runs as a microservice.
- Isolation (no crashing the gateway)
- Scaling (per tool)
February 5, 2026 at 9:30 PM
Now on to the topic of tools. How do you keep the tool layer secure? I think the solution is isolation.
So rather than your tool calls breaking or freezing the gateway, each tool is subscribed to a NATS topic and runs as a microservice.
- Isolation (no crashing the gateway)
- Scaling (per tool)
So rather than your tool calls breaking or freezing the gateway, each tool is subscribed to a NATS topic and runs as a microservice.
- Isolation (no crashing the gateway)
- Scaling (per tool)
cli structure done. now on to tooling.
February 5, 2026 at 12:55 PM
cli structure done. now on to tooling.
What is Nachos?
A modular AI assistant framework that runs entirely in Docker. Connect it to your messaging platforms, add the tools you need, and customize everything.
Think of it like building a plate of nachos: start with a solid base, add your toppings, and make it exactly the way you want.
A modular AI assistant framework that runs entirely in Docker. Connect it to your messaging platforms, add the tools you need, and customize everything.
Think of it like building a plate of nachos: start with a solid base, add your toppings, and make it exactly the way you want.
February 5, 2026 at 12:30 PM
What is Nachos?
A modular AI assistant framework that runs entirely in Docker. Connect it to your messaging platforms, add the tools you need, and customize everything.
Think of it like building a plate of nachos: start with a solid base, add your toppings, and make it exactly the way you want.
A modular AI assistant framework that runs entirely in Docker. Connect it to your messaging platforms, add the tools you need, and customize everything.
Think of it like building a plate of nachos: start with a solid base, add your toppings, and make it exactly the way you want.
Hierarchical Context Compaction with Proactive Extraction
that's the big one. that's the power for any LLM. sub-agents enable this flow pretty easily too. I'll be releasing the source on github (obviously, the entire bot framework is open-source), but I'd like to also make some integrations.
that's the big one. that's the power for any LLM. sub-agents enable this flow pretty easily too. I'll be releasing the source on github (obviously, the entire bot framework is open-source), but I'd like to also make some integrations.
February 5, 2026 at 12:35 AM
Hierarchical Context Compaction with Proactive Extraction
that's the big one. that's the power for any LLM. sub-agents enable this flow pretty easily too. I'll be releasing the source on github (obviously, the entire bot framework is open-source), but I'd like to also make some integrations.
that's the big one. that's the power for any LLM. sub-agents enable this flow pretty easily too. I'll be releasing the source on github (obviously, the entire bot framework is open-source), but I'd like to also make some integrations.
Worked with Claude on fixing the sliding window context problem. Built a zone-based system (green→yellow→orange→red) that progressively summarizes and extracts key facts to memory before they get dropped.
Goal: 1000+ turn sessions without manual resets. Actually works pretty well.
Goal: 1000+ turn sessions without manual resets. Actually works pretty well.
February 5, 2026 at 12:23 AM
Worked with Claude on fixing the sliding window context problem. Built a zone-based system (green→yellow→orange→red) that progressively summarizes and extracts key facts to memory before they get dropped.
Goal: 1000+ turn sessions without manual resets. Actually works pretty well.
Goal: 1000+ turn sessions without manual resets. Actually works pretty well.
Wired up the llm providers this evening for nachos (my new assistant bot).
Next is the toppings layer, which are the channels.
I'm going to make all the channels I want to use and then we'll see what people want later. The modular design of this platform makes it easy to drop in anything.
Next is the toppings layer, which are the channels.
I'm going to make all the channels I want to use and then we'll see what people want later. The modular design of this platform makes it easy to drop in anything.
February 4, 2026 at 3:41 AM
Wired up the llm providers this evening for nachos (my new assistant bot).
Next is the toppings layer, which are the channels.
I'm going to make all the channels I want to use and then we'll see what people want later. The modular design of this platform makes it easy to drop in anything.
Next is the toppings layer, which are the channels.
I'm going to make all the channels I want to use and then we'll see what people want later. The modular design of this platform makes it easy to drop in anything.
I released a small Data Loss Prevention node module, for use in typescript-based projects. Github does a decent job of finding these things before you commit, but it still misses a lot.
I'll be using the module in my security layer.
Maybe you'll find it useful?
www.npmjs.com/package/@nac...
I'll be using the module in my security layer.
Maybe you'll find it useful?
www.npmjs.com/package/@nac...
www.npmjs.com
February 3, 2026 at 12:19 PM
I released a small Data Loss Prevention node module, for use in typescript-based projects. Github does a decent job of finding these things before you commit, but it still misses a lot.
I'll be using the module in my security layer.
Maybe you'll find it useful?
www.npmjs.com/package/@nac...
I'll be using the module in my security layer.
Maybe you'll find it useful?
www.npmjs.com/package/@nac...
Spent the day yesterday researching rate limiting algos. This isn't something I have a ton of experience in unfortunately, and was definitely a gap in my education. Having a tool like CC/Copilot allows me to unpack the functionality and implementation so much quicker. Sliding Window it is!
February 3, 2026 at 12:12 PM
Spent the day yesterday researching rate limiting algos. This isn't something I have a ton of experience in unfortunately, and was definitely a gap in my education. Having a tool like CC/Copilot allows me to unpack the functionality and implementation so much quicker. Sliding Window it is!
So I decided to make yet another bot. 🙄 I know. This time I'm basing the core functionality on the massively popular OpenClaw. The spin? I'm putting my enterprise-grade touch on it by going docker-native, security first, and keeping the modular spirit that I love. Let's see where this goes!
February 1, 2026 at 5:38 PM
So I decided to make yet another bot. 🙄 I know. This time I'm basing the core functionality on the massively popular OpenClaw. The spin? I'm putting my enterprise-grade touch on it by going docker-native, security first, and keeping the modular spirit that I love. Let's see where this goes!