Skip to content

README.md

Latest commit

 

History

History
64 lines (45 loc) · 2.57 KB

README.md

File metadata and controls

64 lines (45 loc) · 2.57 KB

Google ADK (Agent Development Kit) Examples

These examples demonstrate the same patterns from PicoAgents implemented using Google's Agent Development Kit (ADK). ADK is Google's framework for building AI agents, optimized for Gemini models.

Setup

# Install Google ADK
pip install google-adk

# Set your Google API key
export GOOGLE_API_KEY="your-api-key"

# Or use Application Default Credentials
gcloud auth application-default login

Examples

Agents

Example PicoAgents Equivalent Description
agents/basic_agent.py agents/basic-agent.py Agent with weather and calculator tools
agents/memory.py agents/memory.py Session state for memory management
agents/structured_output.py agents/structured-output.py Pydantic model responses

Workflows

Example PicoAgents Equivalent Description
workflows/sequential.py workflows/sequential.py Sequential agent pipeline

Orchestration

Example PicoAgents Equivalent Description
orchestration/loop_agent.py orchestration/round-robin.py Poet and critic collaboration
orchestration/parallel.py (new pattern) Parallel agent execution

Key Differences from PicoAgents

  1. Google-optimized: Default model is Gemini, tight integration with Google Cloud
  2. ToolContext: Tools receive ToolContext with state, user_id, session_id
  3. Output key pattern: Agents store output to session state via output_key
  4. State interpolation: Instructions can reference state variables: {generated_code}
  5. Native workflow agents: SequentialAgent, ParallelAgent, LoopAgent are first-class

Running Examples

# From the examples/frameworks/google-adk directory
python agents/basic_agent.py
python workflows/sequential.py
python orchestration/loop_agent.py

Model Configuration

All examples use gemini-flash-latest as the default model. You can also use:

  • gemini-flash-latest - Fast, efficient (default)
  • gemini-1.5-pro - More capable for complex tasks
  • OpenAI/Anthropic models via OpenAiLlm or AnthropicLlm