LiteLLM Agent Gateway: one gateway for every A2A agent

Оригинальный автор：LiteLLM AI Gateway

Первоисточник：https://www.youtube.com/embed/0gU9V947NGY

Bring agents from any platform into the LiteLLM gateway and manage them in one place. This walkthrough registers a Pydantic AI agent over the A2A protocol, activates it with a virtual key, adds a second agent from config.yaml, calls the agents through the gateway, and finds the calls in the logs. What's covered - Adding an agent with A2A Standard, or from a platform such as LangGraph, Azure AI Foundry, Bedrock AgentCore or Vertex AI - Agent URL, autofilled agent card fields, and Discover - The A2A protocol version, and how LiteLLM converts the upstream response to it - Skills, cost per query or per token, and making an agent public to the organization - Allowed models, sub-agents and MCP servers per agent - Governance: session budgets, agent rate limits, session rate limits and guardrails - Why a new agent shows Needs Setup, and assigning a virtual key to make it active - Declaring agents in config.yaml - Testing from the terminal - Agent calls and usage in the logs Chapters 0:00 Intro 0:19 Add a new agent over A2A 0:37 Register a Pydantic AI agent 1:05 Agent URL and autofilled card 1:16 Protocol version 1:34 Skills, cost and public agents 1:56 Discover 2:06 Allowed models, sub-agents and MCP servers 2:33 Budgets, rate limits and guardrails 2:47 Assign a team and create the agent 2:59 Needs Setup: assign a virtual key 3:28 Add agents from config.yaml 4:13 Test from the terminal 4:32 Send a message and check the logs 4:46 Recap Links LiteLLM Agent Gateway docs: https://docs.litellm.ai/docs/a2a LiteLLM on GitHub: https://github.com/BerriAI/litellm