Personas and system instructions

The SDK allows you to customize your agent’s identity, behavioral constraints, and instructions using templated or custom system instructions.

Templated system instructions

Templated system instructions are recommended for most applications. This approach preserves default SDK scaffolding and environment context while overriding the persona identity and appending custom structured sections.

For example, you can template system instructions to define an agent that focuses on reviewing a pull request for code quality:

from google.antigravity import Agent, LocalAgentConfig, types

identity = "You are an expert Code Quality Reviewer."
review_criteria = types.SystemInstructionSection(
    title="review_criteria",
    content="- Focus on readability and maintainability.",
)

templated_si = types.TemplatedSystemInstructions(
    identity=identity,
    sections=[review_criteria],
)

config = LocalAgentConfig(system_instructions=templated_si)

async with Agent(config) as agent:
    response = await agent.chat("Review the latest pull request.")
    print(await response.text())

Custom system instructions

For complete control, custom system instructions bypass all default SDK scaffolding and environment context, allowing you to pass raw system prompts directly.

For example, you can customize the system instructions to set an identity of a “Custom Code Reviewer”:

from google.antigravity import Agent, LocalAgentConfig, types

custom_si = types.CustomSystemInstructions(
    text="<identity>Custom Code Reviewer</identity>"
)

config = LocalAgentConfig(system_instructions=custom_si)

async with Agent(config) as agent:
    response = await agent.chat("Audit this file.")
    print(await response.text())

Sample code

For full working code examples, see the GitHub repository:

• persona_config.py