AgentOS configuration

AgentOSConfig controls how a running OS presents its components and data domains to the control plane. These examples define the same concepts in Python and YAML, then read GET /config to prove what AgentOS rendered.

Files

File	What it teaches
`config_basics.py`	Define available models, per-agent manifest metadata, quick prompts, and named session/memory domains in Python.
`yaml_config.py`	Load those fields from `config.yaml` and verify the rendered configuration.
`config.yaml`	Declarative AgentOSConfig values whose IDs match the Python runtime objects.

Prerequisites

Configuration inspection needs no external credentials. Set OPENAI_API_KEY only when you also want to run the configured agent.

Run the Python configuration

Terminal 1:

bash

.venvs/demo/bin/python cookbook/05_agent_os/08_os_config/config_basics.py

Terminal 2:

bash

.venvs/demo/bin/python cookbook/05_agent_os/08_os_config/config_basics.py --demo

The manifest key is the explicit agent ID operations-agent. Labels appear on the component card and quick_prompts appear in chat. available_models controls the model choices presented by the UI; it does not replace the model configured on the Agent itself.

The session and memory entries name how the same real SQLite database appears in the control plane. An explicit domain entry takes precedence for its db_id; AgentOS auto-discovers and appends only databases that do not already have an entry.

Run the YAML configuration

Terminal 1:

bash

.venvs/demo/bin/python cookbook/05_agent_os/08_os_config/yaml_config.py

Terminal 2:

bash

.venvs/demo/bin/python cookbook/05_agent_os/08_os_config/yaml_config.py --demo

YAML is useful when operators should change UI metadata without editing Python. Python is preferable when configuration is assembled conditionally or shared with typed application constants. In both cases, Python still creates the runtime agents and databases: YAML configures their presentation, so its manifest keys and db_id values must match those objects. Passing a YAML path as config= selects the YAML values; passing an AgentOSConfig object selects the in-code values.

This example intentionally registers no messaging interfaces. Interfaces are runtime objects passed to AgentOS(interfaces=[...]), not phantom YAML declarations.

Input schemas and the chat form

An Agent, Team, or Workflow can define an input_schema. The chat UI discovers components through /config, fetches the selected component's detail response, and uses its JSON schema to render a structured form. See 09_serving_workflows/with_input_schema.py for a workflow and a live GET /workflows/{id} check.