Literature: Agent Development Kit (ADK) Documentation
Source Metadata
- File:
00_Raw/adk-documentation.md - Topic: Framework for building, evaluating, and deploying multi-agent systems.
- Languages Supported: Go, python, Java, TypeScript.
High-Level Summary
The ADK is a code-first toolkit designed for sophisticated AI agent development. It moves beyond simple chat interfaces into structured, multi-agent orchestration and lifecycle management (Build, Evaluate, Deploy).
Key Primitives Identified
- Agent: Fundamental worker unit. Includes
LlmAgent(reasoning) and Workflow Agents (SequentialAgent,ParallelAgent,LoopAgent). - Tool: External capabilities (APIs, Search, Code Execution).
- Session & State: Context handling for single conversations.
- Memory: Long-term recall across multiple sessions.
- Artifacts: Binary/file data management.
- Runner: The execution engine orchestrating the flow.
Architectural Themes
- Multi-Agent Design: Hierarchical and specialized agents.
- Flexible Orchestration: Combining deterministic workflows with dynamic LLM routing.
- Developer Tooling: CLI and Dev UI for debugging and visualization.
- Streaming: Native support for bidirectional text/audio (e.g., Gemini Live API).
Next Steps for Synthesis
- Extract detailed specifications for workflow-agents.
- Map the adk-go-implementation (Go seems to be a primary focus in the docs).
- Detail the adk-artifact-service and adk-evaluation-framework.