Skip to content

deploy-an-agent.md

Latest commit

 

History

History
162 lines (125 loc) · 4.63 KB

deploy-an-agent.md

File metadata and controls

162 lines (125 loc) · 4.63 KB

Deploy a LangGraph Agent

Overview

This guide explains how to deploy a LangGraph agent. There are two options:

Option 1: LangGraph Cloud

The simplest and fastest way to deploy a LangGraph agent is through LangGraph Cloud deployment.

You can learn more in LangGraph official documentation:

To deploy your agent, simply take these steps:

  1. Keep your API keys in a safe space and delete the .env file. Then push local changes to GitHub.

    Important: Never commit your API keys to production.

  2. In LangSmith, click Deployments in the left menu, then click New Deployment.

  3. Connect your GitHub account and select the repository with the project.

  4. Enter the deployment name.

  5. Add your OPENAI_API_KEY as an environment variable.

  6. Click Submit at the top and wait for your deployment.

  7. If everything is fine, you'll see the Currently deployed status. In the right panel, under API URL, you'll also find your agent's URL.

  8. Now you can interact with your agent:

    • Use you agent's API URL as the base URL for the calls.
    • The assistant will have the same ID as locally: fe096781-5601-53d2-b2f6-0d3403f7e9ca.
    • All production API calls require the x-api-key header for authorization. In this header, pass your LangSmith API key.

Example Call 1

Once your agent is deployed, you can test it by running Create Run, Wait for Output.

POST AGENT_API_URL:2024/runs/wait
Headers: Content-Type: application/json, x-api-key: LANGSMITH_API_KEY
Body:

{
  "assistant_id": "fe096781-5601-53d2-b2f6-0d3403f7e9ca",
  "input": {
    "messages": [
      {
        "role": "user",
        "content": "What can you do?"
      }
    ]
  }
}
curl AGENT_API_URL/runs/wait \
 --request POST \
 --header 'Content-Type: application/json' \
 --header 'x-api-key: LANGSMITH_API_KEY' \
 --data '{
   "assistant_id": "fe096781-5601-53d2-b2f6-0d3403f7e9ca",
   "input": {
     "messages": [
       {
         "role": "user",
         "content": "What can you do?"
       }
     ]
   }
 }'

Example Call 2 (Python Agents Only)

If you Agent is in Python, you can also try A2A Post.

POST AGENT_API_URL/a2a/fe096781-5601-53d2-b2f6-0d3403f7e9ca
Headers: Accept: application/json, Content-Type: application/json, x-api-key: LANGSMITH_API_KEY
Body:

{
  "jsonrpc": "2.0",
  "id": "",
  "method": "message/send",
  "params": {
    "message": {
      "role": "user",
      "parts": [
        {
          "kind": "text",
          "text": "What can you do?"
        }
      ],
      "messageId": ""
    },
    "thread": {
      "threadId": ""
    }
  }
}
curl AGENT_API_URL/a2a/fe096781-5601-53d2-b2f6-0d3403f7e9ca \
  --request POST \
  --header 'Accept: application/json' \
  --header 'Content-Type: application/json' \
  --header 'x-api-key: LANGSMITH_API_KEY' \
  --data '{
    "jsonrpc": "2.0",
    "id": "",
    "method": "message/send",
    "params": {
      "message": {
        "role": "user",
        "parts": [
          {
            "kind": "text",
            "text": "What can you do?"
          }
        ],
        "messageId": ""
      },
      "thread": {
        "threadId": ""
      }
    }
  }'

Option 2: Self-Hosting

You can also self-host the LangGraph runtime using any hosting stack. While this setup doesn't include LangSmith, the LangGraph Server API remains fully accessible.

For a simple example walkthrough, check out this video tutorial: Deploy ANY Langgraph AI Agent in Minutes!

Here are the main steps to take:

  1. Create a Dockerfile.

  2. Keep your API keys in a safe space and delete the .env file. Then push local changes to GitHub.

    Important: Never commit your API keys to production.

  3. Get environment variables for your Postgres database and key-value storage.

  4. Set up a hosting:

    • Add environment variables
    • Set up a key-value storage
    • Connect to your GitHub project

  5. Deploy. Your agent and LangGraph API endpoints will be available on a public URL.