How vishal-bulbule/adk-python works
ADK is a Python framework for defining, orchestrating, evaluating, and deploying AI agents. The primary workflow starts with a developer defining agents and optional workflows, then running them locally or through the web interface; agents use models and tools, while session/event state supports ongoing interactions. The README documents additional branches for evaluation, agent-to-agent tasks, configuration, and deployment. Source excerpts are limited, so unverified runtime connections are omitted rather than inferred.
Main entry points: workflow Workflow runtime · llm_agent.py LLM agent · session.py Session history · base_agent.py Agent definitions
Components 17
Agent authoring
- Agent definitions
src/google/adk/agents/base_agent.py - Agent Config
src/google/adk/agents/agent_config.py
Workflow runtime
- Workflow runtime
src/google/adk/workflow - LLM agent
src/google/adk/agents/llm_agent.py - Model requests
src/google/adk/models/llm_request.py - Model responses
src/google/adk/models/llm_response.py
Tools and delegation
- Tool execution
src/google/adk/tools/base_tool.py - Agent task API
src/google/adk/a2a
Interaction state
- Session history
src/google/adk/sessions/session.py - Session services
src/google/adk/sessions/base_session_service.py - Interaction events
src/google/adk/events/event.py - Scoped state
src/google/adk/sessions/state.py - Artifact services
src/google/adk/artifacts/base_artifact_service.py
Development operations
- CLI and web
src/google/adk/cli/cli.py - Agent evaluation
src/google/adk/evaluation/agent_evaluator.py - Deployment targets
src/google/adk/cli/deployers
Other components
- Developer
Connections 16
- Developer → Agent definitions: defines agents
README.md - Developer → Workflow runtime: composes workflow
README.md - Agent definitions → Workflow runtime: orchestrates agents
README.md - Developer → Agent Config: configures agents
README.md - Developer → CLI and web: runs locally
README.md - CLI and web → Deployment targets: deploys agents
README.md - Developer → Agent evaluation: evaluates agents
README.md - LLM agent → Model requests: requests model
- Model requests → Model responses: receives response
- LLM agent → Tool execution: invokes tools
README.md - Workflow runtime → Agent task API: supports task agents
README.md - Session services → Session history: stores sessions
src/google/adk/sessions/base_session_service.py - Session history → Interaction events: records events
src/google/adk/sessions/session.py - Session history → Scoped state: holds state
src/google/adk/sessions/session.py - Workflow runtime → Session services: uses session service
- LLM agent → Artifact services: uses artifacts