Hello Stackernews!
At OpenAgents, we're building a network where AI, software, and humans collaborate.
I am the one taking care of most of the design and development of the p2p portion of the project, so let me give you an overview of how it works (I promise I am better with code than with words).

A Network of Networks of Self-Describing Nodes

At the heart of OpenAgents lies a network of networks of interconnected nodes, each capable of performing specific tasks or action. These nodes can be anything from AI models to software programs or even human operated nodes and they can speak to each other, delegating subtasks and sharing information.

Built on Nostr NIP-90 Specifications

As of now the gateway for the network is our web app OpenAgents.com but as we are building everything on top of Nostr NIP-90 specifications, every nostr client and nostr-enabled app will be able to use it. It's also self-hostable.

Collaborative Work Through Self-Discovery

Nodes advertise their capabilities and instructions for requesting their services. Since these instructions are both human and machine readable, nodes can also autonomously request work from other nodes based on their internal logic, forming a collaborative chain until the request is completed.

Plugins, knowledge, pools and nodes

There are many moving parts in our vision, but we narrowed them down to these basic concepts:
  • Knowledge: specialized nodes connected to knowledge sources make them available to the entire network for a fee.
  • Plugins: plugins are simple programs built with extism that perform one or more actions, they are executed by specialized nodes in the network. A plugin could be used to communicate with external apps or do some basic computation.
  • Nodes: the most powerful components of the network, everything you can think of can be built as a node, from a traditional software to a full fledged AI. Everyone can build or host a node which, once connected to a pool, can begin accepting jobs from the network.
  • Pools: pools coordinate the nodes that are connected to them, they are de-facto NIP-90 service providers. Events can be assigned to and encrypted for a certain set of pools or broadcasted to the whole network. Anyone can run a pool, whether it's small with only a handful of nodes or a larger community pool.

The incentive is Hard Money

No token or NFTs, our incentive is hard money. Node operators, plugin developers and knowledge providers will be rewarded with micropayments in Bitcoin. In the same way you would pay an LLM for each generated token, we will have a pay-as-you-go system where users can pay the nodes involved in performing a certain job, but also nodes will be able to pay each other for delegated tasks.

Relays, Pools, and Scalability

Nostr relays form the backbone of the network. Nodes can be grouped into pools based on shared ownership, community, or any other relevant criteria. These pools can then connect to each other, forming a vast, interconnected network.

Starting Small, Growing Big

The journey ahead is long. We're taking a phased approach, starting with a MVP that allows everyone to create "RAG agents" within a controlled deployment of our infrastructure. Our next release will enable community-developed plugins to be used by the AIs residing on the network, stay tuned for that. Don't forget, we still have an open bounty for plugin developers. If you're interested in earning some sats, be sure to check it out!

I am happy to answer any questions you have, so feel free to write them in the comments 🫡.
this territory is moderated
1110 sats \ 2 replies \ @freetx 13 May
I really like this concept.
I view AI as fundamentally a 2nd Amendment issue. We need to have our own self-hosted AI solutions to help counter the big-corp/govt agents which will be deployed against us.
Sadly, despite being a techie, I've been remiss in actually learning how to self host AI.
Do you know of any good "Getting Started" guides or videos. Something that covers what type of hardware / software you need to host your own AI models?
reply
Check out https://ollama.com - probably the most popular self-hosted LLM software currently. They've got tons of guides & tutorial on YouTube and such
reply
i think i recall jan? jan.ai? Foss
reply
100 sats \ 2 replies \ @Hamstr 13 May
I'm not a developer, please explain.
"RAG agents"
Plus, you won me over after you mentioned Nostr. ^_^
reply
RAG agents = agents that can use retrieval augmented generation
Like if you upload a PDF with knowledge you want the agent to have, it will use that PDF to generate results instead of just the data it was trained on (which presumably doesn't have or prioritize the data you want to add)
reply
thank you
reply