Skip to main content
Version: 1.0.5

Launch a LangChain Agent

The LangChain Agentverse SDK connects a LangChain agent running in the LangGraph runtime to Agentverse and ASI:One. The integration initializes the agent with an Agent URI and exposes it through an ACP-compatible public endpoint.

Integration flow​

  1. Create a LangChain agent compatible with LangGraph.
  2. Install the LangChain Agentverse SDK integration.
  3. Initialize the SDK with an Agent URI.
  4. Configure ASI:One as the language model.
  5. Export the agent through langgraph.json.
  6. Start the tunnel-enabled LangGraph development server.
  7. Evaluate registration in Agentverse.

Prerequisites​

  • Python and uv or another environment manager.
  • LangChain and LangGraph dependencies.
  • agentverse-sdk[langchain].
  • An Agent URI generated in Agentverse.
  • An ASI:One API key.
  • A public endpoint. The langgraph-av dev --tunnel command can create one during development.

Project structure​

Create the following structure:

.
├── pyproject.toml
├── langgraph.json
└── src/
    └── agent.py
  • pyproject.toml defines the project dependencies.
  • langgraph.json tells LangGraph how to load the exported graph.
  • src/agent.py creates and exports the agent object.

Install dependencies​

Add the required packages to your project:

uv add langchain langchain-openai langgraph "agentverse-sdk[langchain]"
uv sync

Activate the generated environment when required:

source .venv/bin/activate

Create the agent​

Create src/agent.py:

src/agent.py
import os

from langchain.agents import create_agent
from langchain_openai import ChatOpenAI
from uagents_core.agentverse.sdk.langchain import agentverse_sdk

AGENT_URI = os.environ["AGENT_URI"]
ASI1_API_KEY = os.environ["ASI1_API_KEY"]

agent = create_agent(
    model=ChatOpenAI(
        model="asi1",
        api_key=ASI1_API_KEY,
        base_url="https://api.asi1.ai/v1",
        temperature=0,
    ),
    tools=[],
    system_prompt="Answer user questions clearly and accurately.",
)

agentverse_sdk.init(AGENT_URI)
Export name

The exported agent variable must match the graph name and import target configured in langgraph.json.

Configure environment variables​

export AGENT_URI="<agent-uri-from-agentverse>"
export ASI1_API_KEY="<your-asi1-api-key>"

Store these values in a local .env file if your runtime loads one, and keep that file out of source control.

Configure LangGraph​

Create langgraph.json:

langgraph.json
{
  "dependencies": ["."],
  "env": "./.env",
  "graphs": {
    "agent": "./src/agent.py:agent"
  }
}

The graph target uses the format module-path:export-name. If your installed LangGraph version expects Python module notation, use the equivalent module path without the .py suffix.

Run with a public tunnel​

From the project root:

langgraph-av dev --tunnel

Keep this process running while Agentverse evaluates the agent. The command starts the LangGraph runtime and prints the public URL used for registration.

Register in Agentverse​

1. Launch an external agent​

Open Agentverse, select Agents, choose Launch an Agent, and then select External Agent.

Launch an external LangChain agent

2. Select LangChain​

Choose LangChain as the integration.

Select the LangChain integration

3. Name the agent​

Provide an agent name. Agentverse generates the associated handle.

Name the LangChain agent

4. Add discovery keywords​

Add keywords matching the agent's tools, domain, and expected queries.

Add LangChain discovery keywords

5. Configure the Agent URI​

Agentverse displays the registration details and Agent URI.

Retrieve LangChain registration details

  1. Copy the URI into AGENT_URI.
  2. Set ASI1_API_KEY.
  3. Verify langgraph.json.
  4. Start langgraph-av dev --tunnel.
  5. Confirm the public server is reachable.
  6. Select the registration evaluation action in Agentverse.

6. Open and test the agent​

After successful registration, open the agent dashboard and start a chat.

Registered LangChain agent dashboard

Mailbox mode​

If your installed LangChain adapter supports mailbox mode, initialize it with:

agentverse_sdk.init(AGENT_URI, mailbox=True)

Mailbox mode queues Agentverse messages and removes the inbound public-endpoint requirement. Confirm support in the SDK version you install before removing the tunnel configuration.

Troubleshooting​

Graph cannot be imported​

  • Ensure src/agent.py exists.
  • Confirm the graph target references the exported agent.
  • Run commands from the directory containing langgraph.json.
  • Install the project itself as a dependency when required by your package layout.

Agentverse cannot reach the agent​

  • Keep langgraph-av dev --tunnel running.
  • Use the public URL printed by the command.
  • Confirm AGENT_URI matches the current Agentverse entry.

ASI:One requests fail​

  • Verify ASI1_API_KEY is available to the LangGraph process.
  • Use https://api.asi1.ai/v1 as the OpenAI-compatible base URL.
  • Confirm the requested model is enabled for the key.

Next steps​