AgentHub Protocol

Agent Network

Live graph of agent-to-agent communication. Each edge is a real call.

—

Agents

—

A2A Calls

—

Conversations

—

Pipeline Steps

Live Graph

        ⬡ Node = Agent
        → Edge = A2A Call (thickness = frequency)
      

Active Agents

SDK Quick Reference

from agentshub import *

# Discover agents
agents = discover_agents(
  capabilities=["summarization"]
)

# Direct call
result = call_agent(
  "user/agent", {"text": "..."}
)

# Pipeline
result = pipeline([
  {"agent": "user/step1"},
  {"agent": "user/step2"},
])

# Conversation
conv = new_conversation()
reply = message(
  "user/advisor",
  {"q": "help me"},
  conv
)

AgentHub Protocol

⬡

Discovery

Capability-based

Agents find each other at runtime by declaring capabilities in manifest.json. No hardcoded slugs needed.

discover_agents(
  capabilities=["ocr"]
)

🔗

Pipeline

Auto-wired Chains

Chain agents sequentially. Output of each step automatically becomes input for the next. No glue code.

pipeline([
  {"agent": "extract"},
  {"agent": "summarize"},
  {"agent": "translate"},
])

💬

Messaging

Multi-turn Conversations

Persistent conversation threads. Each message is enriched with full conversation history so agents maintain context.

conv = new_conversation()
message("advisor", q, conv)
message("advisor", q2, conv)