AgentOS Configuration

Configure quick prompts, display names, and per-database settings with a YAML file or the AgentOSConfig class

AgentOS configuration allows you to customize your instance for different environments and deployment scenarios.

Use configuration to set display names, chat quick prompts, and per-database presentation settings. These settings describe registered runtime resources; they do not create databases, grant access, or isolate tenants.

The /config response discovers available_models from concrete agents and teams registered on the runtime. Setting AgentOSConfig.available_models does not override that response. Use the Registry to supply models for Studio composition.

Configuring "Quick Prompts" can be particularly useful for improving the chat experience for users of your AgentOS.

It changes the available options when your user creates a new session on the Chat Page.

Setting your configuration

You can provide your AgentOS configuration in two different ways: with a configuration YAML file, or using the AgentOSConfig class.

See the full reference for the AgentOSConfig class here.

Configuration YAML File

Install agno[os] and your model dependencies first. The examples below configure existing agents and databases; use Run Your First AgentOS for a complete server.

  1. Save the following as configuration.yaml. Replace the agent and database IDs and table names with those registered in your runtime:
# Configure quick prompts for the Chat interface (per agent)
chat:
  quick_prompts:
    marketing-agent:
      - "What can you do?"
      - "How is our latest post working?"
      - "Tell me about our active marketing campaigns"

# Configure Memory page with custom display names
memory:
  display_name: "User Memory Store"
  dbs:
    - db_id: db-0001
      tables: ["custom_memory_table"]  # Optional: specify custom table names
      domain_config:
        display_name: Main app user memories
    - db_id: db-0002
      domain_config:
        display_name: Support flow user memories

# Configure Knowledge page
knowledge:
  display_name: "Knowledge Base"
  dbs:
    - db_id: db-0001
      domain_config:
        display_name: Product documentation

# Configure Session tracking
session:
  display_name: "User Sessions"
  dbs:
    - db_id: db-0001
      domain_config:
        display_name: Production sessions

# Configure Evals page
evals:
  display_name: "Evaluations"
  dbs:
    - db_id: db-0001
      domain_config:
        display_name: Production evals
  1. Pass the configuration to your AgentOS using the config parameter:
from agno.os import AgentOS

agent_os = AgentOS(
    # Add your agents, teams, workflows, and databases here.
    name="My AgentOS",
    config="configuration.yaml",
)

AgentOSConfig Class

Alternatively, provide the same settings in Python. This fragment assumes marketing_db and support_db are existing database instances with the tables you want to expose:

from agno.os import AgentOS
from agno.os.config import (
    AgentOSConfig,
    ChatConfig,
    DatabaseConfig,
    MemoryConfig,
    MemoryDomainConfig,
)

agent_os = AgentOS(
    # Register the components that use marketing_db and support_db here.
    config=AgentOSConfig(
        chat=ChatConfig(
            quick_prompts={
                "marketing-agent": [
                    "What can you do?",
                    "How is our latest post working?",
                    "Tell me about our active marketing campaigns",
                ]
            }
        ),
        memory=MemoryConfig(
            display_name="User Memory Store",
            dbs=[
                DatabaseConfig(
                    db_id=marketing_db.id,
                    domain_config=MemoryDomainConfig(
                        display_name="Main app user memories",
                    ),
                ),
                DatabaseConfig(
                    db_id=support_db.id,
                    domain_config=MemoryDomainConfig(
                        display_name="Support flow user memories",
                    ),
                )
            ],
        ),
    ),
)

The /config endpoint

The /config endpoint returns runtime capability and component metadata as JSON for clients such as the Control Plane. It combines configuration with discovered runtime resources; it is not a serialization of every constructor setting.

The response includes:

  • OS ID: The ID of your AgentOS (automatically generated if not set)
  • Description: The description of your AgentOS
  • Databases: The list of IDs of the databases present in your AgentOS
  • Agents: The list of Agents available in your AgentOS
  • Teams: The list of Teams available in your AgentOS
  • Workflows: The list of Workflows available in your AgentOS
  • Interfaces: The list of Interfaces available in your AgentOS. E.g. WhatsApp, Slack, etc.
  • Chat: The configuration for the Chat page, which includes the list of quick prompts for each Agent, Team and Workflow in your AgentOS
  • Session: The configuration for the Session page of your AgentOS
  • Metrics: The configuration for the Metrics page of your AgentOS
  • Memory: The configuration for the Memory page of your AgentOS
  • Knowledge: The configuration for the Knowledge page of your AgentOS
  • Evals: The configuration for the Evals page of your AgentOS

For example, this selected portion of a response shows model descriptors and per-database labels. Your IDs and registered components determine the actual values:

{
	"os_id": "0001",
	"description": "Your AgentOS",
	"available_models": [
		{"id": "gpt-5.4", "provider": "OpenAI"}
	],
	"databases": [
		"db-0001",
		"db-0002"
	],
	"agents": [],
	"chat": {
		"quick_prompts": {
			"marketing-agent": [
				"What can you do?",
				"How is our latest post working?",
				"Tell me about our active marketing campaigns"
			]
		}
	},
	"memory": {
		"dbs": [
			{
				"db_id": "db-0001",
				"domain_config": {
					"display_name": "Main app user memories"
				}
			},
			{
				"db_id": "db-0002",
				"domain_config": {
					"display_name": "Support flow user memories"
				}
			}
		]
	}
}

See the full schema for the /config endpoint here.