handlebar-google-adk 0.1.0

Handlebar AI governance for Google-ADK agents

Handlebar for google-adk

Add Handlebar runtime checks, controls and governance into the Google ADK. This package provides a HandlebarRunner wrapper around the default Runner class, to provide agent governance in a single-line code change. The Handlebar runner collects audit logs of your agents' runs, tool and model use, and enforces rules around agent behaviour you've configured on the Handlebar platform.

Note: This package in early development and the interface is subject to change.

Features

Short-term roadmap:

• Rule engine for allow/block tools, based on:
  • user category
  • rule category
• Tool ordering + execution time checks
• custom checks for rules (numeric tracking; boolean evaluation)
• Audit telemetry + consumers
• Agent lockdown + human-in-the-loop actions

Roadmap

Handlebar is in early development. We have a lot of functionality planned, but need your feedback on what you need to help you build better agents.

• Please feel free to open an issue if you have any feedback or suggestions
• or join our Discord to talk to us directly

Getting started

Install the package in your codebase

pip install handlebar_google_adk

The package provides a runner, HandlebarRunner, which is a drop-in replacement for the default google-adk Runner class

- from google.adk.runners import Runner;
+ from handlebar_google_adk import HandlebarRunner;

- runner = Runner(
+ runner = HandlebarRunner(
    agent=agent,
    app_name=handlebar_app_name,
    session_service=session_service,
+   handlebar_api_key=handlebar_api_key,
)

On the Handlebar platform create an api key for your org and ensure you've set the environment variable in your codebase as HANDLEBAR_API_KEY=<your api key>

By default, the Handlebar runner enforces no governance rules, but does collect agent telemetry which can be exported to analyse your agent's behaviour. You can configure rules on tool use and behaviour for Handlebar to enforce at runtime on the platform

Audit logs

The Handlebar runner collects audit logs for key events in the lifetime of an agent run, such as tool usage and agent rule evaluations (see below for more on Handlebar rules). With your Handlebar API key configured, the platform will automatically ingest the logs created by the runner, which you can view on the dashboard.

Contributing

We welcome contributions from the community: bug reports, feedback, feature requests. Please refer to [CONTRIBUTING.md][root_contributing] for ways you can help, and guidelines.

About Handlebar

Find out more at https://gethandlebar.com