Kagenti ADK
March 31, 2026 · View on GitHub
The developer toolkit for getting agents into production.
Documentation · Discord · Contributing
Kagenti ADK (Agent Development Kit) takes agents built with any framework or custom code and turns them into A2A-compliant services.
Warning
This project is under active development and not yet ready for production use.
What's in the kit
| Component | Description |
|---|---|
| CLI | Scaffold projects, run agents locally, and deploy |
| Python SDK | Wrap your agent with A2A, inject runtime services via dependency injection |
| TypeScript Client SDK | Build applications that talk to your agents |
| Server | Self-hostable runtime with everything below built in |
Runtime services
| Service | What it does |
|---|---|
| LLM proxy | Single API for 15+ providers — OpenAI, Anthropic, watsonx.ai, Bedrock, Ollama |
| MCP connectors | Connect agents to external tools via Model Context Protocol |
| PostgreSQL | Agent state, conversation history, and configuration |
| Vector search | pgvector for embeddings and similarity search |
| File storage | S3-compatible upload/download via SeaweedFS |
| Document extraction | Text extraction from PDFs, CSVs, and more via Docling |
| Authentication | Identity and access management via Keycloak |
| Observability | LLM tracing and agent debugging via Phoenix |
| Web UI | Built-in chat interface for testing your agents |
Quick start
sh -c "$(curl -LsSf https://raw.githubusercontent.com/kagenti/adk/main/install.sh)"
Get involved
We'd love to hear from you — whether you have questions, feedback, or want to contribute.
| Discord | Join the community |
| kagenti-maintainers@googlegroups.com | |
| Contributing | Read the guide |
| Issues | Report a bug or request a feature |