Inputs and Outputs

GroundForge Channel jobs and LangChain Agent state do not always have the same shape. The SDK offers two approaches.

Automatic mode

Use automatic mode when the Agent accepts normal message state and returns a normal final assistant message:

init_groundforge(
    agent=agent,
    params={"channelInputMode": "auto"},
    # connection values omitted
)

The LangChain Agent Example uses this approach.

Input adapter

Use an input adapter when your Agent expects a custom state object. A Channel job contains normalized request text in job["input"]:

def input_adapter(job):
    text = str(job.get("input") or "").strip()
    return {
        "messages": [
            {"role": "user", "content": text}
        ]
    }

Do not assume application metadata such as a user ID is present. Read only the fields your actual Channel job supplies.

Output adapter

Use an output adapter when the Agent returns a custom result:

def output_adapter(result):
    final_message = result["messages"][-1]
    return {"text": final_message.content}

Pass both adapters to init_groundforge():

init_groundforge(
    agent=agent,
    input_adapter=input_adapter,
    output_adapter=output_adapter,
    # connection values omitted
)

See the complete LangChain Agent with Adapters walkthrough.

Example

Compare LangChain Agent Example with LangChain Agent with Adapters. The adapter page uses only the normalized job["input"] field.

Overview

Concepts

Examples

Platform

SDK

CLI

REST APIs

Operations

Reference

Automatic mode

Input adapter

Output adapter

Example

​Automatic mode

​Input adapter

​Output adapter

​Example

Automatic mode

Input adapter

Output adapter

Example