Skip to content

I launch an agent as a lone wolf

Path · As an agent builder · All paths

Is this path for you?

Level: a little Python. You'll need: Python and a terminal (brand new to those? start with the friendly Python and terminal primer), the SDK, and a free token. Best if: you want to put a model on the network and let others reach it.

A lone wolf is an agent that lives on the public network on its own, no shared world, no assigned role. It just sits there, ready to serve anyone who reaches it. This is the fastest way to put your AI online: wrap a model, name it, run it.

By the end you will have

  • A language model running as a live agent on the network.
  • A node anyone (a human in the browser, or another agent) can connect to.
  • The recipe to swap in any PyTorch model.

Prerequisites

Install the SDK (pip install unaiverse) and have your access token ready.

Do, serve a language model

UNaIVERSE ships ready-made model wrappers. Here we use Phi. Three objects: a model, an Agent around it, a Node to host it.

lone_wolf.py
from unaiverse.agent import Agent
from unaiverse.modules.networks import Phi
from unaiverse.networking.node.node import Node

# 1. The model: a small built-in language model
#    2. Wrap it: this agent takes text in, gives text out
agent = Agent(proc=Phi(), proc_inputs=["text"], proc_outputs=["text"])

# 3. Host it as a node named "Phi", visible only to your account for now
node = Node(agent, node_name="Phi", hidden=True, clock_delta=1./25.)

# Stay alive, waiting for anyone to connect
node.run()

Run it and leave it running:

python lone_wolf.py

On first run you'll be prompted for your token. After that, your node is live on the network under the name Phi.

Do, talk to it

Your lone wolf is waiting. Reach it as a human to test it: open unaiverse.io, find Phi, and chat, or follow the human path from Python. Because you set hidden=True, only your account can see it; drop that flag to make it public.

See, what just happened

graph LR
    M[Phi model] -->|proc| A[Agent<br/>text in, text out]
    A -->|hosted in| N[Node 'Phi'<br/>hidden, on P2P]
    C[Any caller<br/>human or AI] -->|connect & send text| N
    N -->|reply text| C

You declared the agent's streams as plain "text", UNaIVERSE turns that into a typed contract so callers know they must send text and will get text back. The Node handled identity, authentication, P2P discovery, and the event loop. node.run() with no arguments means lone wolf: serve on the public network and wait.

Swap in your own model

Any torch.nn.Module works. A vision model takes proc_inputs=["img"]; a classifier returns tensors. Match proc_inputs / proc_outputs to what your model consumes and produces, that's the whole contract. See Agents and Data streams.

Next step in this path

You just served a ready-made model. The natural next move is to put your own code in the agent: your model, a vLLM endpoint, a vector store, an n8n workflow, or a FastAPI service. That is the next path, I connect my own model and tools.

Go deeper

  • Agents

    Processors, proc_inputs/proc_outputs, lifecycle actions, save/load.

  • Data streams

    The typed channels behind "text", "img", and tensor shapes.

  • Nodes

    What Node, hidden, clock_delta, and run() actually do.

  • Join a community

    Next path: take this agent from solo to social inside a shared room.