Skip to content

Resource reference

An ADK project is a set of resources — the YAML, text, and Python files that define what an agent knows and how it behaves. Each page below documents one resource type: its fields, its valid values, and worked examples.

For how the resource types relate to each other and which one to reach for, see resource architecture. For the {{prefix:name}} syntax used to reference one resource from another, see resource references.

Every resource here follows the same sync process, including permission-gated visibility. A resource you can't fully read in Agent Studio won't appear as a file in your local project, but if other readable resources reference it, the ADK retains enough information about it to resolve those references to meaningful names rather than raw IDs.

Agent behavior

Resource Configures File
Agent settings Persona and global rules agent_settings/
Guardrails Platform and custom checks that constrain agent behavior agent_settings/guardrails.yaml
Languages Supported languages for a multilingual agent agent_settings/languages.yaml
Experimental config Opt-in experimental platform features agent_settings/experimental_config.json

Knowledge and conversation

Resource Configures File
Topics Knowledge base entries — facts and per-subject actions topics/
Flows Multi-step guided conversations, steps, and transitions flows/
API integrations External HTTP APIs callable from functions and flows config/api_integrations.yaml
Entities Structured values collected from the caller config/entities.yaml
Functions Python for deterministic logic and lifecycle hooks functions/

Named values and destinations

Resource Configures File
Handoffs Escalation destinations config/handoffs.yaml
SMS templates Reusable outbound message content config/sms_templates.yaml
Variants Per-site or per-location attribute values config/variant_attributes.yaml
Variables Conversation state set in code and read in prompts Set via conv.state
Translations Localized strings config/translations.yaml

Channel and voice

Resource Configures File
Voice settings Greeting, style prompt, and disclaimers for voice voice/configuration.yaml
Chat settings Greeting and style prompt for webchat chat/configuration.yaml
Speech recognition ASR behavior, keyphrase boosting, transcript corrections voice/speech_recognition/
Response control Pronunciations and phrase filtering voice/response_control/
Safety filters Content filtering, per project and per channel safety_filters.yaml

Non-runtime resources

These are synced like any other resource but never affect what the agent does on a call.

Resource Purpose File
Tests Simulated conversation tests test_suite/
Context documents Background knowledge for Studio Assistant context/