# Registry (/agent-os/studio/registry) **The Registry manages non-serializable components (tools, models, databases, schemas, functions, etc.) that Studio depends on.** ## Component Types [#component-types] * **Tools**: `Toolkit` instances, `Function` objects, or plain callables. * **Models**: model provider instances (OpenAI, Anthropic, etc.). * **Databases**: `BaseDb` instances for storage. * **Vector DBs**: `VectorDb` instances for knowledge bases. * **Schemas**: Pydantic `BaseModel` subclasses for structured I/O. * **Functions**: Python callables used as workflow evaluators, selectors, or executors. * **Knowledge**: `Knowledge` instances for RAG. * **Memory Managers**: `MemoryManager` instances for managing user memories. * **Session Summary Managers**: `SessionSummaryManager` instances for generating session summaries. * **Teams**: `Team` instances to reuse as members in teams and workflows. * **Agents**: `Agent` instances to reuse as members in teams and workflows. * **Workflows**: `Workflow` instances for rehydration and Studio. * **Learning Machines**: Named `LearningMachine` instances shared by stored components. Open Studio Registry from the Studio navigation ## Run the Example [#run-the-example] ```bash uv pip install -U "agno[os]" openai anthropic ddgs pgvector "psycopg[binary]" export OPENAI_API_KEY="your_openai_api_key" ``` Set `ANTHROPIC_API_KEY` when running a component that selects Claude. The registered knowledge and vector stores are configured here; populate them before expecting search results. Run PostgreSQL at `postgresql+psycopg://ai:ai@localhost:5532/ai`, or update the example URL. Enable the pgvector extension for the vector stores. See [PostgreSQL setup](/database/providers/postgres/overview). Save the code as `registry_app.py` and run `python registry_app.py`. Example of registry configuration: ```python title="registry_app.py" from agno.db.postgres import PostgresDb from agno.knowledge.knowledge import Knowledge from agno.learn.machine import LearningMachine from agno.memory import MemoryManager from agno.models.anthropic import Claude from agno.models.openai import OpenAIChat, OpenAIResponses from agno.os import AgentOS from agno.registry import Registry from agno.session import SessionSummaryManager from agno.tools.calculator import CalculatorTools from agno.tools.websearch import WebSearchTools from agno.vectordb.pgvector import PgVector from agno.workflow import StepInput, Workflow from pydantic import BaseModel DB_URL = "postgresql+psycopg://ai:ai@localhost:5532/ai" class InputSchema(BaseModel): input: str description: str def custom_evaluator(step_input: StepInput) -> bool: return "urgent" in (step_input.input or "").lower() db = PostgresDb(db_url=DB_URL, id="postgres_db") user_memory_manager = MemoryManager( model=Claude(id="claude-sonnet-4-5"), db=db, additional_instructions=""" IMPORTANT: Don't store any memories about the user's name. Just say "The User" instead of referencing the user's name. """, ) concise_summary_manager = SessionSummaryManager( model=OpenAIResponses(id="gpt-5-mini"), session_summary_prompt=( "Summarize the conversation in 3-5 bullet points focused on decisions, " "open questions, and any follow-ups required." ), last_n_runs=10, ) agent_knowledge = Knowledge( name="Agent Knowledge", description="Example knowledge base for agents", vector_db=PgVector(table_name="agent_knowledge_documents", db_url=DB_URL), contents_db=db, ) shared_learning = LearningMachine( name="Shared Learning", db=db, user_memory=True, ) triage_workflow = Workflow( id="triage-workflow", name="Triage Workflow", steps=[], ) registry = Registry( name="My Registry", tools=[CalculatorTools(), WebSearchTools()], models=[OpenAIChat(id="gpt-5-mini"), Claude(id="claude-sonnet-4-5")], dbs=[db], vector_dbs=[PgVector(db_url=DB_URL, table_name="embeddings")], schemas=[InputSchema], functions=[custom_evaluator], memory_managers=[user_memory_manager], session_summary_managers=[concise_summary_manager], knowledge=[agent_knowledge], learning=[shared_learning], workflows=[triage_workflow], ) agent_os = AgentOS(id="my-app", registry=registry, db=db) app = agent_os.get_app() if __name__ == "__main__": agent_os.serve(app="registry_app:app", port=7777) ``` ## Automatic discovery and available tools [#automatic-discovery-and-available-tools] AgentOS also discovers dependencies from its registered agents, teams, and workflows and adds them to the registry for reconstruction of stored components. Discovery does not automatically make a component's private tools available for building new components. Tools explicitly passed to `Registry(tools=[...])` are available for Studio composition. For `StudioTools`, `allowed_tools` can permit discovered tool names; `denied_tools` takes precedence over both explicit registration and the allowlist. This controls tool selection when building components, rather than replacing runtime authentication. ## Registry API [#registry-api] The registry exposes a [`GET /registry`](/reference-api/schema/registry/list-registry) endpoint through AgentOS with filtering and pagination. ### Query Parameters [#query-parameters] | Parameter | Type | Default | Description | | --------------- | -------- | ------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `resource_type` | `string` | `None` | Filter by type: `tool`, `model`, `db`, `vector_db`, `schema`, `function`, `agent`, `team`, `workflow`, `knowledge`, `memory_manager`, `session_summary_manager`, `learning` | | `name` | `string` | `None` | Partial name match (case-insensitive) | | `page` | `int` | `1` | Page number | | `limit` | `int` | `20` | Items per page (1-100) | ### Response Metadata [#response-metadata] Each component in the response includes type-specific metadata: | Component Type | Metadata Fields | | ----------------------- | ------------------------------------------------------------------------------------------------------------------------------------------ | | Tool | `class_path`, `parameters`, `signature`, toolkit `functions` | | Model | `provider`, `model_id` | | Database | `db_id` | | Vector DB | `collection`, `table_name` | | Schema | JSON schema definition | | Function | `signature`, `parameters` | | Knowledge | `class_path`, `vector_db_class`, `contents_db_class`, `max_results`, `num_readers` | | Memory Manager | `class_path`, `model_class`, `model_id`, `db_class`, memory flags (`add_memories`, `update_memories`, `delete_memories`, `clear_memories`) | | Session Summary Manager | `class_path`, `model_class`, `model_id`, `last_n_runs`, `conversation_limit` | | Team | `id`, `class_path` | | Agent | `id`, `class_path` | | Workflow | `id`, `class_path` | | Learning Machine | `class_path`, `namespace`, `stores`, `model_id`, `db`, `knowledge`, optional `custom_stores` | ## Developer Resources [#developer-resources] * [AgentOS reference](/reference/agent-os/agent-os) * [Registry API reference](/reference-api/schema/registry/list-registry) * [Studio overview](/agent-os/studio/introduction) * [Studio workflows](/agent-os/studio/workflows)