litestar-ai-serving

What problem does it solve?

This Skill solves the recurring challenge of building consistent, production-ready HTTP-facing AI agent endpoints with Litestar and Google ADK, eliminating guesswork around dependency injection, session persistence, error handling, and security guardrails for model workflows.

Core Features & Use Cases

• Canonical ADK Integration Patterns: Pre-vetted wiring for Google ADK LlmAgent, Runner, and SQLSpecSessionService via Dishka dependency injection, with explicit guidance on scope and runtime introspection constraints.
• Production Guardrails: Built-in patterns for handling missing Vertex AI credentials, request timeouts, unbounded model call blocking, and authorization enforcement for agent endpoints.
• Reference Implementation Templates: Ready-to-use patterns for non-streaming chat handlers, persona-augmented system prompts, tool-use-first agent workflows, and multi-turn session memory backed by SQLSpec.
• Use Case Example: If you are building a customer support chat endpoint that uses Vertex AI, maintains conversation history across HTTP requests, and enforces auth on agent tools, this Skill provides the exact patterns to implement it correctly the first time.

Quick Start

Use the litestar-ai-serving skill to build a production-ready HTTP chat endpoint that integrates Google ADK, maintains multi-turn session state, and enforces security guardrails for AI model workflows.