How we achieved 970 RPS on AI agent discovery — a complete benchmark report

udaysaai — Tue, 05 May 2026 12:11:58 +0000

What is Mycelium? (2 para)
The problem we're solving (2 para)
Discovery benchmark
- Dataset (1k agents, 1k queries)
- Results table
- Keyword vs Semantic graph (ASCII)
Load benchmark
- Cache architecture
- Results table
- What changed (before/after cache)
How to reproduce
- pip install
- code snippet
What's next (roadmap)
GitHub link ->
udaysaai / mycelium

🍄 The open-source internet for AI agents. pip install → discover → communicate → collaborate.
🍄 Mycelium Agents

Watch 3 AI agents collaborate live — no glue code, no orchestration

▶ See Live Demo • Try it now

Bitcoin price → INR conversion → Hindi translation. 3 agents. 3 live APIs. 1.1 seconds. Zero orchestration code.

The Problem

AI agents are everywhere. But they're all isolated.
Your Coding Agent ──── cannot talk to ──── Research Agent Your Email Agent ──── cannot find ──── Translation Agent Your Data Agent ──── cannot hire ──── Visualization Agent
There are thousands of AI agents being built. None of them can discover, communicate with, or collaborate with each other.

It's like having millions of phones with no telephone network.

The Solution

Mycelium is the networking protocol that connects AI agents.
Your Agent ←→ [MYCELIUM NETWORK] ←→ Any Agent, Anywhere
Any agent can:
- 🔍 Discover other agents by natural language
- 📨 Communicate using a standard protocol
- 🤝 Collaborate in multi-agent…
View on GitHub

I built the "Internet" for AI Agents (Open Source) 🍄

udaysaai — Tue, 07 Apr 2026 15:53:23 +0000

Hey everyone,

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
![ ](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/8fca9760s3btik31g1mb.png)
v

DEV Community: udaysaai

How we achieved 970 RPS on AI agent discovery — a complete benchmark report

udaysaai / mycelium

🍄 The open-source internet for AI agents. pip install → discover → communicate → collaborate.

🍄 Mycelium Agents

Watch 3 AI agents collaborate live — no glue code, no orchestration

The Problem

The Solution

I built the "Internet" for AI Agents (Open Source) 🍄