Google A2A

Google A2A

Tell me about Google A2A, where are the docs?

Search for: Tell me about Google A2A, where are the docs?

What is json RPC? how does it work?

Search for: What is json RPC? how does it work?

What are SSE and Http Streaming and how do they work?

Search for: What are SSE and Http Streaming and how do they work?

Home page of a2a

Key concepts from a2a site

Agent card object structure

  1. Explains the key objects in the spec
  2. Interestingly Agent is not one of them
  3. Briefly introduces transport and messaging
  4. Introduces "contextid" that is a grouping of tasks
  1. Agent card: The description of end points
  2. Task: a stateful object that is part of a conversation that is exchanged between client and server
  3. Message: id, role, many parts
  4. Part: text, file etc.
  5. Artifact: output, parts

The specification URL

Here is a sample agent card link

  1. name
  2. capabilities: streaming, notifications, state history
  3. security: openid
  4. skill: optimize traffic routes
  5. skill: custom maps

On an agent card, in a2a, how come there are no input arguments, like to a tool?

Search for: On an agent card, in a2a, how come there are no input arguments, like to a tool?

  1. Has definitions for various objects
  2. Common flows (briefly)
  3. Obtaining an agent card
  4. Simple q/a
  5. A long running task SSE interaction
  6. When agent needs input
  7. Push notiification
  8. File exchange
  9. Structured data

is there something called an Agent Executor as a class in Google A2A?

Search for: is there something called an Agent Executor as a class in Google A2A?

Here is the basic tutorial of A2A

  1. Has a formal definition for AgentExecutor
  2. The basic concepts behind the A2A protocol.
  3. How to set up a Python environment for A2A development using the SDK.
  4. How Agent Skills and Agent Cards describe an agent.
  5. How an A2A server handles tasks.
  6. How to interact with an A2A server using a client.
  7. How streaming capabilities and multi-turn interactions work.
  8. How an LLM can be integrated into an A2A agent.

Does google have their own agentic framework like Crew or LangGraph?

Search for: Does google have their own agentic framework like Crew or LangGraph?

How does Google A2A differ from, complement, from other efforts in Calude or OpenAI?

Search for: How does Google A2A differ from, complement, from other efforts in Calude or OpenAI?

  1. High-level agent framework with first-class Agent/LlmAgent classes for defining models, prompt templates, and tool plugins
  2. Built-in orchestration primitives (e.g. SequentialAgent, ParallelAgent) and evaluation hooks for constructing multi-step pipelines
  3. Cloud-native deployment to Vertex AI Agent Engine, offering integrated observability, authentication, and VPC controls
  4. Optional A2A integration via AgentExecutor, so ADK-built agents can automatically speak the A2A protocol when needed
  1. Protocol-centric implementation of JSON-RPC methods (message/send, tasks/*), SSE streaming, and core objects (Message, Task) for agent-to-agent communication
  2. Interoperability focus?provides discovery, task orchestration, and secure messaging without any built-in model or prompt abstractions
  3. Lightweight and agnostic?you supply your own business logic (often by implementing AgentExecutor), and the SDK handles compliance with the A2A spec
  4. Separately packaged and versioned as a2a-sdk/python-a2a, so you can adopt just the protocol layer independently of ADK

Agent Development Kit (ADK) Python SDK link

Agent2Agent (A2A) Python SDK

  1. What on-prem deployment models (bare-metal, private cloud, Kubernetes) does A2A support?
  2. How does A2A handle air-gapped environments or strictly isolated networks?
  3. Which self-hosted LLM runtimes (e.g., Vertex AI on-prem, Gemini locally, third-party open models) are certified or recommended?
  4. Can A2A integrate with our existing PKI/LDAP/SAML authentication infrastructure?
  5. What logging and audit trails are available for message exchanges, task lifecycles, and context propagation?
  6. How granular are the A2A permission controls (per-agent, per-task, per-method)?
  7. What metrics, traces, and dashboards come out of the box?and can we hook into Prometheus/Grafana?
  8. Which language-specific A2A SDKs (Python, Java, Go, .NET) are the most mature for on-prem development?
  9. Do you provide on-prem-friendly starter templates (Helm charts, Dockerized examples) for both agents and clients?
  10. What major on-prem enhancements or performance optimizations are planned in the next 6?12 months?