Nostr Web Client

I've setup ollama.jb55.com as my AI model server. my own private GPU-powered coding agent over wireguard that I can use anywhere. so cool.

I've also setup n8n.jb55.com as an AI workflow/CI server, going to build MCP/goose tools where my agents can trigger specific workflows for me.

Example 1: "hey goose, can you start a code review for prs #42 and #920?

Action:

1. n8n workflow that fetches the PR

2. passes the diffs to the local LLM for review

3. sends nostr node to my local relay so that I am notified.

Example 2: "hey goose, can you tell me if I have any important looking emails from my team"

Action: call `notmuch search query:work` and read my damus work emails, summarizing the most important ones for me

After this I am going to try to implement hotword detection so I can just say "hey goose" and then have it listen to commands so I don't have to type them. All with local LLMs!

super fun!

nostr:note1sr8fms9tdvasuc5x03p3fd9j9unpka46we327utdtqw67etjvzvsd9phxr

Reply to this note

Please Login to reply.

Discussion

a1 10mo ago

Dope

InBoston 10mo ago

Does goose work well with ollama model instances? How many tokens/s are you getting?

jb55 10mo ago

The tools feature only works with models that have tool support. gpt4o, sonnet, and the one i linked for a local model. Haven’t tested many. Generation is a bit slower for the local one but i generally run these as background tasks

guy 10mo ago

There's a special goose model for tools of deep seek r1 but it was 70b. Bit slow for my setup, but nice to have locally too.

Gísli Kristjánsson 10mo ago

I’ve a similar setup for accessing my local models from anywhere. Love the CI idea. Might have to steal it.

matevz 10mo ago

do you have ollama integrated into your IDE? I use Claude with my Intellij but I'm considering an offline alternatives...

jb55 10mo ago

I don’t use an ide, i just use vim

Séimí Mac Síomón 10mo ago

I use MSPaint

DarkySparkle 10mo ago

You should check out the Devvox Genie plugin and make your LLM pick. Local or cloud services, including api keys of paid services 👌

guy 10mo ago

Have you tried openwebui yet... It's pretty fun too. Has lots of plugins and expandable features as well.

jb55 10mo ago

Yeah i have that under ai.jb55.com and its hooked up to ollama

Justin Moon 10mo ago

Have you found any benefits to self-hosting the model other than privacy?

jb55 10mo ago

its a prerequisite for a lot of things I want to do, mainly around private RAG (emails, etc). I would obviously use a frontier model if I could, but I don't have the GPU power for that.

guy 10mo ago

I want to see screenpipe integration 😂 I would definitely only run something like that locally.... And still feel paranoid about it.

Justin Moon 10mo ago

Ok yea that sounds really enticing. Look forward to hearing about your progress …

wolfpy 10mo ago

Style points

772f9545... 10mo ago

Wow!

Joe Resident 10mo ago

Nice. And just think, all this infra will just magically get smarter every 2 months as models improve