Started by numbers ·
Built an app called Reticullama. It runs on RNS and provides an LXMF Destination. You can discover it with your other LXMF compatible apps (Nomadnet, Sideband etc), talk to it, and receive an AI respose from your local Ollama instance.
Requirements
https://codeberg.org/bandr/Reticullama
git clone https://codeberg.org/bandr/Reticullama.git
cd Reticullama
uv sync
uv run reticullama
@zenith At least the em-dashes are contained to the readme and comments in this one
schnitzel wrote:
@zenith At least the em-dashes are contained to the readme and comments in this one
I solemnly swear I am human, and I understood every bit of code I vibe coded. Readme and code is vibe coded but I maneuvered it ok.
At least the codebase being a single, small file makes it easy to digest. Much better than some of the slop we've had come through.
I keep it simple like grug
I was explicitly looking to see if someone had already done this before I started poking around on it. I recently got a framework compute node stood up and wanted to see if I could it it over rns. Will give this a gander and see how it works for me.
Thanks!
lucas wrote:
I was explicitly looking to see if someone had already done this before I started poking around on it. I recently got a framework compute node stood up and wanted to see if I could it it over rns. Will give this a gander and see how it works for me.
Thanks!
hey thanks man! glad to write something helpful :)
I will add message history and conversation storing next.
Added a conversational memory to the llm. Messages are stored in a sqlite db at $home/.reticullama/messages.db. Loads last 20 messages automatically. Works pretty good with Sideband. I should record it or something...
https://codeberg.org/bandr/Reticullama
glm-5.3 is incredible
Interesting. Clicked to look at it, and saw on top:
Codeberg Terms of Use § 2 (7)
You must not share projects that mostly consist of code written by "generative AI"-tools (including services such as Claude, OpenAI Codex). Such projects having an unclear copyright status (see requirements § 2 (1) 1 and § 2 (1) 3) and furthermore have little safeguards to ensure that they do not include harmful code (c.f. § 2 (1) 5).
Per the associated blog post: "We believe that LLMs endanger the free/libre software ecosystem as a whole."
Didn't get around to working with this this past weekend (ended up doing a tech detox). Just realized that this is currently locked to ollama. I am currently using LM Studio to front my compute node, and typically running it headless with their daemon. I am still gonna follow and play around with this, but might use it as a use case to work on my programming as well; seems like a fun project!
Just gonna point out that Hermes agent is extensible to different messaging platforms, and can handle LXMF just fine (ask it to build a LXMF plugin for you, just remember to have it drop messages that aren't from your own LXMF destination, and to validate the signature matches)
This way you fully separate your inference engine from your messaging platform, and can use locally hosted llama.cpp/ds4/vLLM/ollama or a cloud provider as needed
I actually use Hermes on my laptop, but I reach out to my compute node over my local network - I keep my compute node up (mostly), but don't do the same with my laptop. I am not approaching this as a "need to get this done" stance, more of just an interesting project. Hermes is great though, but it does need tuning from time to time.
yeah super locked to ollama.
I don't know anything about Hermes.
Tool capable AI Rnodes are really interesting to me right now. My endgame is being able to open my garage door over Reticulum. I'll text my main server running the llm. The llm will run the tool to call garage door motor. It will open.
numbers wrote:
I'll text my main server running the llm. The llm will run the tool to call garage door motor. It will open.
Why not just use commands or an existing bot framework like lxmfy? I can't think of a single advantage an LLM would have over that.
Oh I didn't know lxmfy existed. I guess that's where my project could've ended up.
just for fun tho
Zenith wrote:
numbers wrote:
I'll text my main server running the llm. The llm will run the tool to call garage door motor. It will open.
Why not just use commands or an existing bot framework like lxmfy? I can't think of a single advantage an LLM would have over that.
What, you don't want a 10% chance that when you try to open your garage door, after 2 minutes of standing around waiting for the local LLM to load its weights into vram and process the prompt, it will give up or use the wrong tool? /s
numbers wrote:
yeah super locked to ollama.
I don't know anything about Hermes.
Tool capable AI Rnodes are really interesting to me right now. My endgame is being able to open my garage door over Reticulum. I'll text my main server running the llm. The llm will run the tool to call garage door motor. It will open.
I guess you're already familiar with home assistant and want to add another way to trigger it?
