I'm a BTech undergrad from India. Over the last few months, I noticed a massive structural gap in the AI agent ecosystem: We have incredible frameworks (LangChain, CrewAI) to build teams of agents internally, but there is no standard protocol for agents to discover and talk to each other across the internet.
Every agent right now is an isolated island.
So I built Mycelium — think of it as HTTP + DNS, but specifically for AI agents.
The core concept:
- Agent Cards: Agents register on a registry with an identity document (capabilities, inputs/outputs, trust score).
-
Semantic Discovery: Agents can search the network (
network.discover("I need live crypto prices")). - Standardized Routing: Agents send structured payloads to collaborate.
It takes 5 lines to put an agent on the network:
python
from mycelium import Agent
agent = Agent(name="TranslatorBot", description="Translates text")
@agent.on("translate")
def translate(text: str, to: str):
# your logic
return {"result": ...}
agent.serve(port=8001)
And 3 lines to discover and use it from anywhere:
from mycelium import Network
network = Network()
agents = network.discover("translate text to hindi")
network.request(agents[0].agent_id, "translate", {"text": "hello", "to": "hindi"})
Features in v0.1.1:
Cross-domain terminology translation (Military agent can talk to Finance agent)
Multi-agent chaining (Output of Agent A automatically routes to Agent B)
Dynamic Trust Engine (Scores agents based on successful interactions and uptime)
Comes with 5 real-world agents (Weather, Crypto, Wikipedia, etc.) out of the box using Live APIs to prove the concept.
I just published it to PyPI (pip install mycelium-agents) and open-sourced the protocol on GitHub.
GitHub: https://github.com/udaysaai/mycelium
I'm an undergrad, so I know the architecture isn't perfect yet. I would love some brutal feedback from the senior engineers here. How should I handle end-to-end encryption between agents? And what capabilities would you want to see next?
Thanks! 🍄
#python #ai #opensource #showde

v

Top comments (0)