Skip to main content

Agent-to-Agent (A2A) Protocol

CodeMie supports the Agent-to-Agent (A2A) Protocol, an open-standard protocol designed to enable seamless communication and collaboration between AI agents built with different frameworks and by different vendors. A2A provides a common language that breaks down silos and fosters interoperability across the AI ecosystem.

With A2A integration, you can connect CodeMie to external AI agents—regardless of their framework or origin—allowing them to communicate effortlessly and work together on complex tasks.

Key Features​

  • Framework-Agnostic: Connect agents built with LangGraph, LangChain, CrewAI, or any framework supporting A2A
  • Vendor-Independent: Integrate agents from different providers and platforms
  • Remote Assistants: Add external agents as assistants within CodeMie
  • Standardized Communication: JSON-RPC 2.0 protocol for reliable agent interactions
  • Multi-turn Conversations: Support for conversational flows with state management
  • Real-time Streaming: Get status updates and results as they're processed
  • Secure Integration: Optional authentication and authorization support

How It Works​

A2A enables standardized interaction between CodeMie and external agents through a client-server model:

Prerequisites​

To integrate an external agent with CodeMie via A2A:

  • Agent URL: The external agent must expose an A2A-compatible endpoint
  • Agent Card: The agent should provide a card describing its capabilities
  • Network Access: CodeMie must be able to reach the agent's URL
Authentication

Most A2A agents are publicly accessible and don't require authentication. If your agent requires authorization, see Authentication section below.

Create Remote Assistant​

  1. Navigate to Assistants and click the Create Remote Assistant button in the top-right corner:

    Assistants list with Create Remote Assistant button

  2. On the Create Remote Assistant page, configure the basic settings:

    • Project: Select the project where the assistant will be created.
    • Share with project: Toggle on to make the assistant available to project team members.
    • Integration (Optional): Leave empty for public agents. See Authentication section if your agent requires credentials.

    Create Remote Assistant form

  3. Enter the Assistant URL and click Fetch:

    The assistant URL should point to your A2A agent's endpoint. For example:

    • https://example.com/agent
    • https://example.ngrok-free.dev

    After clicking Fetch, CodeMie will send a request to retrieve the agent card.

  4. Review the fetched agent card details:

    Agent card details

    The agent card contains:

    • Assistant name: Pre-populated from the agent. You can customize it if desired.
    • Description: Agent's purpose and capabilities
    • Skills: Available tools and functions
    • Other metadata: Version, supported modes, etc.
    tip

    The assistant card provides transparency about what the remote agent can do. Review it carefully to understand the agent's capabilities before saving.

  5. Click Save to create the remote assistant.

Work with Remote Assistants​

Remote assistants are marked with an A2A label to differentiate them from regular CodeMie assistants.

Remote assistant with A2A label

Start a Conversation​

Click on the remote assistant card to open the chat interface. The conversation works just like with regular CodeMie assistants:

Chat with streaming status updates

  • Type your questions or requests
  • Receive real-time status updates during processing
  • View results directly in the chat
  • Continue multi-turn conversations with context

View Agent Card​

Click the assistant icon in the chat to view the agent card and detailed capabilities at any time.

Edit Remote Assistant​

Remote assistants support limited editing to maintain integrity:

  • Editable: Assistant name, project assignment, sharing settings
  • Read-only: All agent-specific fields (description, tools, capabilities)

This ensures that the remote agent's configuration remains consistent with its actual implementation.

Delete Remote Assistant​

Remote assistants can be deleted like regular assistants. This only removes the reference in CodeMie—the external agent continues to run independently.

Authentication (Optional)​

info

Most A2A agents are publicly accessible and don't require authentication. Only configure this if your external agent requires credentials.

If your external agent requires authentication, CodeMie supports multiple authentication types:

Configure A2A Integration​

  1. Navigate to Integrations → User or Project → + Create.

  2. Select basic parameters:

    • Project: Select your CodeMie project name.
    • Global Integration: Toggle on to use across multiple projects.
    • Credential Type: Select A2A.
    • Alias: Enter integration name (e.g., "my-agent-auth").

  3. Choose Authentication Type and configure credentials:

Use for agents deployed on AWS that require AWS Signature Version 4 authentication.

AWS Signature authentication

Required fields:

  • AWS Region: AWS region where your agent is deployed (e.g., us-east-1)
  • AWS Access Key: Your AWS access key ID
  • AWS Secret Key: Your AWS secret access key
  • AWS Session Token (optional): Temporary session token for role-based access
  • AWS Service Name (optional): AWS service name for signature (e.g., execute-api)

Using the Integration​

After creating the integration:

  1. Click Save.
  2. When creating a remote assistant, select this integration in the Integration (Optional) dropdown.
  3. CodeMie will automatically include the authentication credentials when communicating with the agent.

Example: Currency Conversion Agent​

A sample currency conversion agent built with LangGraph demonstrates A2A capabilities:

Repository: ai-run-demo-agents

Sample Agent: /python/agents/currency_converter

Agent Capabilities​

  • Real-time currency conversion using Frankfurter API
  • Multi-turn conversations with follow-up questions
  • Streaming status updates during processing
  • Conversational memory across interactions

Example Queries​

"How much is 100 USD in EUR?"
"What's the exchange rate for USD to JPY?"
"Convert 50 EUR to GBP"

Try It Yourself​

  1. Clone the ai-run-demo-agents repository:

    git clone https://github.com/epam-gen-ai-run/ai-run-demo-agents.git
    cd ai-run-demo-agents/python/agents/currency_converter

  2. Set up your environment variables:

    echo "CHAT_MODEL_PROVIDER=azure" > .env
    echo "AZURE_OPENAI_API_KEY=your_api_key_here" >> .env
    echo "AZURE_OPENAI_ENDPOINT=your_endpoint_url" >> .env
    echo "AZURE_OPENAI_API_VERSION=2024-12-01-preview" >> .env

  3. Run the agent with ngrok to expose it publicly:

    uv run . --ngrok_enabled

    The agent will start and ngrok will provide a public URL:

    Agent running with ngrok

  4. Copy the ngrok URL from the terminal output (e.g., https://easily-trasnocheoic-externally.ngrok-free.dev)

  5. In CodeMie, create a remote assistant using this URL as described in the Create Remote Assistant section

Developer Community

We encourage developers to create their own A2A-compatible agents and share them with the community. Check the ai-run-demo-agents repository for examples and contribution guidelines.

Important Notes​

Agent Compatibility​

  • External agents must implement the A2A protocol specification
  • Supported communication: JSON-RPC 2.0 over HTTP/HTTPS
  • Required endpoints: agent card fetch, task send, task subscribe (for streaming)

Network Requirements​

  • CodeMie must have network access to the agent URL
  • For local development, use tools like ngrok to expose agents
  • Production agents should use HTTPS with valid SSL certificates

Security Considerations​

Agent Trust

Only add agents from trusted sources. Remote agents have access to conversation context and can process sensitive information shared in chats.

  • Use A2A integration with authentication for production agents
  • Review agent cards to understand what data the agent accesses
  • Consider using project-level assistants to limit scope

Timeout Configuration​

Administrators can configure A2A timeout settings in the platform configuration:

  • A2A_AGENT_CARD_FETCH_TIMEOUT: Maximum seconds to fetch agent capability cards (default: 30s)
  • A2A_AGENT_REQUEST_TIMEOUT: Maximum seconds to wait for agent responses (default: 30s)

See API Configuration for details.

Learn More​