How bhancockio/deploy-adk-agent-engine works
ADK Short Bot takes a message, uses an instruction-driven Gemini agent to shorten it while preserving meaning, and returns original/new character counts plus the revised message. The agent exposes a character-counting tool. The README documents both local session-based use and remote Agent Engine deployment, including session management and deployment cleanup. Deployment source supports these operations; the exact internal wiring of the ADK runtime is not shown in the sampled code.
Main entry points: agent.py Shortening agent · local.py Local ADK app · remote.py Remote operations · cleanup.py Deployment cleanup
Components 12
Shortening agent
- Shortening agent
adk_short_bot/agent.py - Shortening instructions
adk_short_bot/prompt.py - Character counter
adk_short_bot/tools/character_counter.py
Local sessions
- Local session runner
deployment/local.py - Local ADK app
deployment/local.py
Remote deployment
- Remote operations
deployment/remote.py - Deployment cleanup
deployment/cleanup.py
Google Cloud runtime
- Agent Engine
- Vertex AI
- Gemini model
- Session state
Other components
- Message sender
Connections 16
- Message sender → Local session runner: starts local test
- Local session runner → Local ADK app: creates app
- Local ADK app → Shortening agent: hosts agent
- Local ADK app → Agent Engine: creates, queries sessions
- Message sender → Remote operations: requests operation
- Remote operations → Shortening agent: packages for deployment
- Remote operations → Agent Engine: manages deployment and sessions
- Deployment cleanup → Agent Engine: lists and deletes deployments
- Agent Engine → Session state: stores session state
- Agent Engine → Shortening agent: runs deployed agent
- Shortening agent → Shortening instructions: uses instructions
- Shortening agent → Character counter: calls tool
- Shortening agent → Gemini model: requests shortening
- Gemini model → Vertex AI: runs on
- Local ADK app → Message sender: prints response events
- Remote operations → Message sender: prints results