Skip to content

Agents Association Registry

Introduction

The Agents Association Registry is a dynamic agent registry that manages and tracks runtime subjects (agents, processors, components, etc.) in an intelligent system. Each subject is a versioned, semantically described runtime unit that can be discovered and invoked by the agent.

This registry supports:

  • Fetching and maintaining runtime metadata from a centralized REST API
  • Registering agents by name and subject ID
  • Querying and listing all available runtime subjects
  • Providing LLM-compatible searchable representations for intelligent planners

Each subject includes details like its version, tags, role, supported DSLs, and runtime traits, making it highly suitable for use in autonomous workflows and agent orchestration.

Usage

Initialize the Registry

from agent_sdk.known_subjects import KnownSubjects

known_subjects = KnownSubjects(base_url="http://runtime-db-service")

Add an Agent

# subject_id = "block.image_processor:1.0-stable"
known_subjects.add_agent("image_agent", subject_id)

Remove an Agent

known_subjects.remove_agent("image_agent")

List Registered Agents

agents = known_subjects.list_agents()
for name, subject in agents.items():
    print(f"{name}: {subject.subject_description}")

Query All Runtime Subjects (from backend)

all_subjects = known_subjects.list_runtime_subjects()

Run a Custom Query

query = {"subject_type": "block", "subject_search_tags": ["text", "nlp"]}
results = known_subjects.query_runtime_subjects(query)

Searchable Representation

The method get_searchable_representation() returns a list of structured records that describe each known agent. This is particularly useful for use in LLM-based planners and agent selectors.

searchable = known_subjects.get_searchable_representation()

Each entry includes:

Field Description
id Unique subject identifier (type.name:version-tag)
name Subject name
type Subject type (e.g., block, processor, etc.)
description Human-readable description of what the subject does
tags Semantic search tags
traits Execution/environmental traits (e.g., gpu, batching)
role Subject's system role (generator, router, evaluator, etc.)
schema DSL input schema (if defined)
agent_name Name used in the registry (user-defined alias)

This output can be directly used by an LLM planner to select appropriate agents for downstream tasks.

Example

# Initialize the registry
known_subjects = KnownSubjects("http://runtime-db-service")

# Register an agent by its unique subject_id
known_subjects.add_agent("text_summarizer", "block.summarizer:1.0-stable")

# View registered agents
print("Registered Agents:")
for name, agent in known_subjects.list_agents().items():
    print(f"{name} -> {agent.subject_description}")

# Get structured view for LLM integration
llm_input = known_subjects.get_searchable_representation()

for entry in llm_input:
    print(f"Agent {entry['agent_name']} - Role: {entry['role']} - Tags: {entry['tags']}")