Gzuuus

Yes, what I mean is the where and how of the implementation. I'll share my thoughts on how I was planning to do it and how it could work for this use case. My perspective shifted when I created the transparent OAI proxy https://github.com/gzuuus/transparent-oai-api . With that proxy, I could connect to PPQ, intercept all requests and responses, mainly for debugging purposes. However, that concept could also be used to integrate a cashu wallet into the proxy, enabling automatic payments when necessary. A local piece of software that expose the same API of the provider, acting as a smart proxy to manage payments and top-ups using NWC. This way, the user centralizes their connection with providers and interacts with them using the localhost endpoint.
I have more ideas about this, but simply the system is already pluggable to Goose. It would be as easy as configuring the endpoint in its configuration. Its also more portable. You can also write it in any language

2025-05-31 12:18:27 UTC

- reply

There is a party in cyberspace

2025-05-31 12:17:47 UTC

- reply

Thats imminent. I want to also structure the packages better to use it as a library. The commons pkg is about as that but could be even more useful.

2025-05-31 10:17:12 UTC

GM nostr! We are celebrating today 🧡

2025-05-31 08:44:28 UTC

- reply

Hay works good right? 👀🔥🔥
How was the experience of implementing it?

2025-05-30 08:13:20 UTC

In the era of AI, memory is becoming a commodity.
Every provider you use now creates memories to offer a more personalized experience tailored to you. This makes users' lives easier and adds value to their products. It enables personal assistants/agents to become more effective by getting to know you, reducing the amount of information you have to provide each time you use them, and even predicting your wants and needs.
However, the problem is that users may become locked in with a vendor that captures their memories, increasing their exit costs because they would have to recreate all those useful memories by starting a new "relationship" from scratch with a different provider.
Once again, we see the same mentality of capture and lock that has caused the web to evolve in such a negative way. Fortunately, Nostr exists as an outlet for the healthier use of this technology. It has the potential to put users in control of their data and memories in a decentralized and pluggable manner
#ai

2025-05-29 07:08:21 UTC

Gf*M 🌞🔥

2025-05-27 14:18:53 UTC

- reply

Why? Is an android limitation or something?

2025-05-27 14:10:37 UTC

- reply

You mean from keeping the subs open?

2025-05-27 10:53:56 UTC

- reply

Was looking forward for these sub features 🔥🔥

2025-05-24 15:56:29 UTC

- reply

Portability and anonymity I think are the most obvious improvements

2025-05-22 06:13:26 UTC

Woke up, pizza dreams, pizza breakfast, pizza day 🍕 GM

2025-05-20 14:20:41 UTC

I've been testing Zen Browser since this morning. It's a beautiful browser that's more performant and less resource-hungry than Degoogled Chromium. It's slightly less private than Librewolf, though. It reveals 18.12 bits of identifying information, whereas Librewolf reveals 14.95 bits, according to https://coveryourtracks.eff.org/

2025-05-19 20:32:54 UTC

Stay focused, don't get distracted. There are loads of people/entities competing for your attention. Don't give it away for free; use it instead. Stay focused 🧘

2025-05-15 08:16:33 UTC

- reply

I mean to eat, I saw them flying around.

2025-05-14 17:56:44 UTC

- reply

And poncha 🧡
#nevent1q…9j7f

2025-05-10 11:40:42 UTC

- reply

I think coinos is down 🙏

2025-05-09 09:01:28 UTC

GmGm

2025-05-08 08:30:15 UTC

GM🌞 https://image.nostr.build/cb7b43e7be949ce8b3ab32f441ce1aae9337d0e6d2e63a3b131892ba4b647466.jpg

2025-05-07 14:27:23 UTC

I had to touch a windows laptop today. I still feeling dirty

2025-05-04 09:29:17 UTC

- reply

5300 is a DVM kind, when you interact with an algo feed DVM you follow the same pattern of request response than with other dvms, if it behaves like a DVM is a DVM xD and yes you can solve the same use case in different ways

2025-05-03 21:21:59 UTC

- reply

I cannot fully follow you sorry 🙏

2025-05-03 17:51:02 UTC

Relays, bots, and DVMs are not the same.
These terms are being mixed up recently, leading to clear confusion about what each of them represents. I’m going to share my perspective on what they are, and I hope this can clarify the vision for others or, at least, provide a good opportunity to better define them.
A relay is "essentially" a dumb piece of software that just receives and sends notes upon request. I say "essentially" because there are some advanced and sophisticated relays that offer additional services like translation, spam filtering, etc. That’s fine, but it is still fundamentally a relay.
A bot can be anything, it is an NPUB that can create notes, whether they are posts, responses, reactions, or any other type. Its operation is based on any conditions or instructions that the creator of the bot wants to implement. For example, it could listen for notes mentioning the NPUB of the bot, notes with a certain hashtag, or even other conditions like a weather station publishing weather reports and alerts. Bots are extremely flexible, and can be as simple or complex as you want.
A DVM in the other hand provides specific functionality, whether it’s a translation service, algo feeds, public key lookups, image generation, or anything else. DVMs are also quite flexible, as you can abstract any type of computation behind them, but the key idea is that they use a standard way to access their services.
Putting this all together, you can have an image generation bot that listens for notes mentioning its NPUB or notes with the hashtag #cats (or whatever you choose) to respond to those notes with an image. The bot will be able to use a DVM to fulfill its functionality, and both the bot and DVMs will utilize relays for communication and coordination.
Bots can be complex systems that integrate multiple capabilities, while DVMs are more atomic component, with a single responsibility and a standard interface. One bot could use multiple dvms to handle its functionality.

2025-05-03 17:11:33 UTC

- reply

I would have to study a bit more to give you an answer to that XD

2025-05-03 12:44:50 UTC

- reply

Not really afaik, but shouldn't be pretty difficult

2025-05-02 14:42:17 UTC

- reply

I agree, a DVM is not suitable for everything, or perhaps more accurately, a DVM is not the best solution for every scenario. From my point of view, it primarily describes the pattern of interaction between the client and the service. While this pattern can address many use cases, it is not a silver bullet and can introduce unnecessary round trips, latency, and so on, in certain situations, as you pointed out. The cases you are describing do not necessarily require a DVM, but they could benefit from one if they aim to achieve something more advanced, such as searching across multiple relays with specific condition that cannot be achieved by just nostr filters, or discover content from different relays with advanced parameters. I believe the goal of DVMs is to abstract and provide functional components that users and clients can utilize.

2025-05-02 14:30:50 UTC

- reply

A bot can have any kind of arbitrary behavior, such as listening for and responding to kind 1 notes, or listening for and responding to any other type of event in they way they want. DVMS describes and tries to standardize how this interaction is done by defining requests, responses, and job feedback. Of course, you could do the same thing and call it a bot, but that is not the point, bots can use dvms to solve some of their features.
The differences between an http enpoint and a dvm is nostr

2025-05-02 13:59:42 UTC

- reply

DVMs describes a pattern for clients to interact with some computation or functionality, and bots are some sort of systems that have a defined behaviour and do stuff

2025-05-02 13:44:38 UTC

Due to the recent and ongoing discussions about DVMs and the @nprofile…dlnm intention to reset it, I wrote a proposal rewriting the nip, This is a starting point and, hopefully, it will make sense to others. If you are interested in DVM stuff, please read it and provide feedback https://github.com/nostr-protocol/nips/pull/1904

2025-05-02 13:39:25 UTC

- reply

Totally, what about you?

2025-05-02 09:14:35 UTC

- reply

Loving LNVPS 🧡
#nevent1q…duak

2025-05-02 08:53:56 UTC

- reply

Yes, I was thinking about that thoughtfully when drafting dvmcp spec and my article about dvms and the reflection pattern.

2025-05-02 08:52:32 UTC

- reply

Yea, there are some valid points there

2025-05-01 17:02:23 UTC

- reply

I would like to hear constructive feedback from you. That afirmation is for my recent article about dvms? I agree there is loads of room for improvement

2025-05-01 15:33:52 UTC

- reply

U're a troll, sir

2025-05-01 13:58:17 UTC

- reply

Totally, I moved to ephemeral events for dvmcp

2025-04-30 12:50:57 UTC

- reply

CC @nprofile…vu4p

2025-04-30 08:24:46 UTC

Bitcoin is for enemies, that's right, but it's not a question of making it easier for them. GM

2025-04-29 22:24:30 UTC

- reply

Fue bonito, además luna nueva 🌚

2025-04-29 22:12:57 UTC

- reply

You don't have to support everything specially for specific client APIs

2025-04-29 22:09:34 UTC

Follow packs yeah! At least! people have been using them in a primitive way since nip51 was a mess, it was time they got more attention 💪

2025-04-29 14:14:28 UTC

I am currently testing Qwen3 (1B and 4B) locally with Ollama. It is the small model I was looking for to call MCP tools. While it still a bit dumb in its responses, it successfully calls the tools perfectly. No other model of this size can call tools as effectively.

2025-04-29 10:05:27 UTC

- reply

2025-04-29 10:05:04 UTC

- reply

I think this is quite possible, I already do this in dvmcp and dvmcp.fun as you have defined data types and everything is a compliant json schema where the inputs are defined, you can easily build a form with some defined input methods for each data type, string is just a text input, boolean is just a checkbox, etc. There are already libs that can help you with this and do json schema to form

2025-04-28 09:41:26 UTC

I will start implementing the new DVMCP spec this week, probably starting tomorrow or wed. Please, if you are interested in dvmcp and want to collaborate and review the new spec, read this article that highlights the changes and points you to the pr where im working on it https://dvmcp.fun/blog/evolving-dvmcp

2025-04-28 07:33:40 UTC

GMGMGoodMonday 🌞

2025-04-26 17:41:15 UTC

- reply

Hmm... I think the most logical flow would be per kind, reusing the req/res kinds of each job type, so in this way you will request the input schema from a dvm using its 5xxx req kind and it will respond in its 6xxx, the request can contain a standardized method to request the input schema so clients will know how to request this schema on any dvm. This is also more private than creating a new kind just for req res input schema, since this will look like a normal operation.
And yes, the dvm can choose not to announce itself publicly, so no nip89, this is something I am also introducing for the revised dvmcp spec.
wdyt? Someone should draft this to potentially upgrade nip90

2025-04-26 10:57:10 UTC

- reply

BTW, I am not a mathematician XD

2025-04-26 09:53:45 UTC

- reply

Please @nprofile…xnfk @nprofile…6rpa @nprofile…7n8d can you validate this? 🙏

2025-04-26 09:52:44 UTC

- reply

Thanks to @nprofile…4umc as we discovered this together

2025-04-26 09:52:03 UTC

The GFY framework
Abstract:
In this note, we explore a whimsical yet insightful relationship between two vectors representing opposing life forces: Good For You (GFY1) and Go Fuck Yourself (GFY2). We formalize the idea that the more something is good for you, the less go fuck yourself is, and vice versa.
Introduction
Life often presents us with choices and experiences that can be broadly categorized as either beneficial or detrimental. To capture this duality mathematically, we define two vectors in an abstract vector space:
GFY1: The Good For You vector, representing positive influences, benefits, and well-being.
GFY2: The Go Fuck Yourself vector, representing negative influences, detriments, and frustration.
Defining the Relationship
Intuitively, these two vectors are inversely related, an increase in one corresponds to a decrease in the other. We express this relationship as:
GFY2 = -GFY1
And consequently:
GFY1 + GFY2 = 0
These equations state that the Go Fuck Yourself vector is the additive inverse of the Good For You vector, and their sum is always zero. This is a profound insight for your life.
Interpretation
When GFY1 points strongly in a positive direction (high benefit), GFY2 points equally strongly in the negative direction (low detriment). Conversely, if GFY2 increases (more frustration), GFY1 decreases (less benefit). The sum of these opposing forces is always zero: GFY1 + GFY2 = 0, which symbolizes the balance (or tension) between good and bad.
This apparently simple equation offers a powerful framework for evaluating life choices. When faced with a decision, we cannot truly assess its GFY1 value (how beneficial it is) without simultaneously contemplating its GFY2 value (how frustrating or detrimental it is), as they inherently influence and define each other. To understand how good something is for you requires understanding how much "go fuck yourself" is contained within it, and vice versa. This interdependent relationship helps us make more holistic assessments, allowing us to prioritize choices that maximize GFY1 while naturally minimizing GFY2 in our lives.
Extensions and Applications
This model can be applied to real-life decision-making, where each choice or event can be represented as a vector in the GFY space. By consciously measuring the magnitude and direction of these vectors, we can make more informed decisions. The GFY framework provides a simple yet powerful tool for navigating complex situations, helping us distinguish between what truly serves our interests and what merely frustrates our progress.
Practical applications might include evaluating job opportunities, relationships, or daily habits. By assessing whether something increases your GFY1 or GFY2 vector, you gain clarity about its true value in your life. This mindful approach to decision-making can lead to greater overall satisfaction and reduced stress.
The next time you face a dilemma, just remember: The more GFY, the less GFY, and vice versa. This perspective can guide you toward healthier choices and a more balanced, fulfilling life.

2025-04-26 09:13:26 UTC

- reply

I think I'll write a note expanding on this, it might be something important
#nevent1q…hmrk

2025-04-26 08:37:15 UTC

- reply

It's a mathematical expression I'm working on, indeed is more like GFY₂=-GFY₁ where GFY₁ is the *Good For You* vector, and GFY₂ is the *Go F*** Yourself* and GFY₁+GFY₂ = 0

2025-04-26 07:58:16 UTC

- reply

And then enjoy the rest of the weekend feeling like an oss paladin ❤️‍🔥
#nevent1q…k92m

2025-04-26 07:56:56 UTC

- reply

GM its a good day to read specs and provide feedback 🤙
#nevent1q…uyrg

2025-04-26 07:29:21 UTC

- reply

That's why #soveng @nprofile…3ra5
#nevent1q…e30c

2025-04-26 07:04:03 UTC

- reply

Same 🫂

2025-04-25 12:21:25 UTC

- reply

If you are interested in dvmcp read this, we are reviewing the spec and looking for feedback
#nevent1q…k7kk

2025-04-25 06:24:34 UTC

- reply

Fried brains with beans

2025-04-25 06:18:30 UTC

- reply

I'm not a 0 I'm a 1

2025-04-25 06:13:13 UTC

- reply

Aaa I wish I could assist, sounds super fun

2025-04-24 15:35:21 UTC

- reply

Just wrote this article based in what I'm learning from building dvmcp and the ongoing discussion about dvms
#naddr1qq…fgn4

2025-04-24 07:20:08 UTC

- reply

Wow, the David Lynch tribute page is lovely ❤️‍🔥

2025-04-24 06:30:18 UTC

- reply

Ask if you have any doubt 👌

2025-04-23 21:40:45 UTC

- reply

Yes, for this you can use the nostr beating heart mcp, it has a rag db behind it and can also search for code snippets. You can use dvmcp to use my instance without having to run it locally, it has all nostrbook.dev and nips repo embedded. It will provide all the context to cline about nostr protocol related stuff

2025-04-23 21:36:07 UTC

- reply

Looking forward to catch up also with the new url sharing kinds for nostree 🫡

2025-04-23 21:30:03 UTC

- reply

Yes, I would recommend visiting https://modelcontextprotocol.com to address some initial questions. After that, you can choose one of the sdks. If you prefer a different programming language that isn’t available, you can check gh for libraries in other languages. You can also create your own implementation based on the base protocol... if you want xD Once you pick your poison, writing a server is quite straightforward, for example, I wrote this some time ago in ts https://github.com/gzuuus/mcp-echo/blob/master/src/mcp-server.ts You can also choose from one of the mcp servers available on mcp.so or other repositories. Then if you want to test it out, use the inspector https://modelcontextprotocol.io/docs/tools/inspector which will allow you to call tools and debug the server. After that, you can integrate it with an llm if you wish. Additionally, there is a workshop that utilizes also the dvmcp packages https://dvmcp.fun/blog/dvmcp-workshop

2025-04-23 13:41:58 UTC

- reply

Have you guys seen the movie Snowpiercer?
#nevent1q…9p6l

2025-04-23 11:53:37 UTC

- reply

Yes 👌

2025-04-23 11:45:08 UTC

- reply

I like goose

2025-04-23 10:54:24 UTC

Showtime!🔥
I've been busy working on the @npub1dvm…3jdm stuff, focusing on the spec and packages, and I forgot to show some of the servers I'm already running that are free and open to everyone, these are like apps you can plug into your llm or agents to give them new capabilities. Of course, there's no guarantee of uptime, but they've been running well for a few weeks now with no noticeable downtime. I wanted to put together some useful tools that the majority of people might find useful.
- DuckDuckGo Search
https://dvmcp.fun/dvm/dvm-announcement-duckduckgo
his is an MCP server that you can use to perform searches with DuckDuckGo and retrieve the results. This is useful for improving your search experience by collecting real-time data, performing research, and anything that involves fetching fresh data from the Internet and making it available to your llm.
- Nostr Beating Heart
https://dvmcp.fun/dvm/dvm-announcement-nostr-beating-heart
This MCP server is something I'm experimenting with. Behind it is a RAG DB that contains all the knowledge about the Nostr protocol, from the NIPs repo to all the content from nostrbook.dev (cc: @npub1q3s…d26p). The experimental part involves providing pluggable RAG DBs into your LLMs, allowing you to have these kind of buckets of context to retrieve accurate and concise data from a knowledge base, which is especially good for avoiding bloating the context window of your LLM. It has been working pretty well since nostrbook.dev was indexed. It also includes a tool to search code snippets, making this MCP server quite effective for quickly retrieving data about the protocol, NIPs, kinds, code examples etc.
- Nostr Daily News
https://dvmcp.fun/dvm/dvm-announcement-nostr-daily-news
This is another experimental MCP server where you can find tools to retrieve data from configured sources, including Nostr and RSS. The cool part is that it is collaborative, meaning you can add some RSS sources to Nostr relays, could be like a colaborative bookmark of news sources. It also has some preconfigured relays to fetch news from Nostr, Stacker News, Hacker News, and custom sources like The Rage, TechCrunch, Hugging Face Blog, and more. You can simply prompt your LLM with "give me an overview of the news from all your sources," and it will gather the latest posts from each source, providing you with a comprehensive overview, pretty cool for a morning news digest.
- Nostr MCP Server
https://dvmcp.fun/dvm/dvm-announcement-nostr-mcp-server
This is the MCP server from @npub1s9e…lxzl. It has some interesting tools for using Nostr, including getting profiles, retrieving the latest notes from someone, zaps, long-form content, searching the NIPs repo, and even sending anonymous zaps, very powerful.
---
These are the tools I'm running, but there are also other interesting tools that others are running, which you can plug directly into your LLM or simply visit dvmcp.fun to use them without needing an LLM. If you want to use this with your LLM, I recommend using Goose, but you can use any other client with MCP capabilities. Once you have it set up, using these tools is as easy as configuring the commands or settings you can find in the 'Install' tab of each tool on dvmcp.fun. For example, by setting this command as the MCP server:
`npx @dvmcp/discovery --server #naddr1qv…f7mt`, you'll be plugging into the Nostr Beating Heart. You don't have to run the whole RAG system locally; you can leverage the instance that I'm running on a cheap $2 VPS from @npub1lnv…7vjj
Hope you find this useful! Let's keep building! If you have any doubts, do not hesitate to ask :)

2025-04-22 14:47:51 UTC

- reply

Love

2025-04-22 14:47:30 UTC

- reply

Just saw it today, thinking in spin it up as dvmcp. Also https://dvmcp.fun/dvm/dvm-announcement-nostr-beating-heart goes in the same direction

2025-04-22 08:13:36 UTC

- reply

Fuk, this one looks even worse🙃
#nevent1q…muh3

2025-04-21 15:51:52 UTC

- reply

Tool calling are just structured outputs, but seems that small models sometimes struggle to follow the structure, or get confused. I think this will be solved in the tiny models of the future because standards and over feeding

2025-04-21 15:33:23 UTC

- reply

Men that's hot ❤️‍🔥

2025-04-21 14:59:23 UTC

- reply

You mean the new qat models? I'll be running and testing 12b asap. It's the biggest one my laptop can afford, no racks for this pleb yet 🙃

2025-04-19 14:17:45 UTC

- reply

For the curious cats, this was at the beginning #internetisforcats #catsareforinternet
https://video.nostr.build/a94f00782dad853b6501eca62fcc25b2faae2c2973361d8479d25b0533ee807e.mp4
#nevent1q…mphx

2025-04-19 08:23:36 UTC

GM!🌞🐈 new tiny creature have appeared, just born ~5 w ago
https://image.nostr.build/e31bc17d1fd8dd134ac2b494f72b7229417a94326fed04c05e59913c328ed6c9.jpg

2025-04-16 07:36:29 UTC

GM 🌞

2025-04-11 11:03:27 UTC

- reply

Goose + dvmcp discovery

2025-04-10 09:10:51 UTC

We are so early for some things and so late for others

2025-04-02 16:44:04 UTC

Random Access Melancholy

2025-04-01 21:21:46 UTC

- reply

I agree, I'm just saying that things are changing, and now it's like you can have some sort of chef next to you teaching you how to do the soufflé, I've never said that you shouldn't touch any code 🤷 anyways just opinions about what I can observe around

2025-04-01 12:02:26 UTC

- reply

But that's true, for that reason I think now it's more important to understand what your code is doing and how it plays in your system than how to write it

2025-04-01 09:54:15 UTC

- reply

Things are changing, learning the basics of code is important and should not be avoided, but as I said, for the times to come I think it's more important to understand how software works, and of course how to read code, than to master a language or understand low level stuff. The history of software itself shows this, we can just look at how languages evolved from low level to high level, it was always about making the process of building software easier.

2025-04-01 09:45:57 UTC

GM I'm going to leave nostr and bitcoin and get a fiat job, then get a mortgage and die happy as a normie, it was good to meet you all 🫡

2025-04-01 09:40:14 UTC

- reply

I've been thinking lately that learning how to code is not the right approach anymore, now when a friend asks me how to learn development and code I give more importance to learning software and system design, architecture principles and so on. Code is the mechanical part of software development, wdyt?

2025-03-30 08:33:10 UTC

GSM 🌞

2025-03-26 17:26:32 UTC

🧙‍♂️
https://image.nostr.build/a9c8c4db47c92c4d7790c041ec9c3baef4389f9d8799e7f8dc1d867ec57f33aa.jpg

2025-03-26 08:45:55 UTC

- reply

Grand Master 🙌

2025-03-26 08:45:26 UTC

Today I'll give a workshop about MCP servers, how to use them and deploy using dvmcp bridge. Wednesdays are for workshops 🔥 Then I'll publish the documents so you can follow the workshop too 👌
#sec04 #sovengs #wed4workshop

2025-03-26 08:40:24 UTC

GM 🌞

2025-03-25 09:00:59 UTC

Just encoded a bitcoin seed phrase in this emoji fifo 😅󠅒󠅑󠅓󠅟󠅞󠄐󠅒󠅑󠅓󠅟󠅞󠄐󠅒󠅑󠅓󠅟󠅞󠄐󠅒󠅑󠅓󠅟󠅞󠄐󠅒󠅑󠅓󠅟󠅞󠄐󠅒󠅑󠅓󠅟󠅞󠄐󠅒󠅑󠅓󠅟󠅞󠄐󠅒󠅑󠅓󠅟󠅞󠄐󠅒󠅑󠅓󠅟󠅞󠄐󠅒󠅑󠅓󠅟󠅞󠄐󠅒󠅑󠅓󠅟󠅞󠄐󠅒󠅑󠅓󠅟󠅞

2025-03-25 08:57:16 UTC

GM🌞

2025-03-23 23:17:35 UTC

- reply

Just deployed https://dvmcp.fun/dvm/cb94717693d4845bbfaee30e8adc7fb2055d3babe7d84c9c83a59b1162ea22b4