LangGraph

NiadraMiddleware wraps every model call of the agent built by langchain.agents.create_agent: the pack goes into the system message, the turns are recorded and the history tools come along. Python only; in JavaScript, LangGraph.js uses withNiadraContext() from LangChain.

Install

pip install 'niadra[langgraph]'   # langgraph 1.2, langchain 1.4 and langchain-core 1.6 or newer, below 2

The five primitives

Primitive	How the adapter wires it
Context	The pack is appended to the system message, which keeps your instructions first, and the `turn_block` goes after the messages. Nothing is written to the graph’s state: the checkpointer never stores a pack
Turns	The customer’s messages are recorded before the call, keyed by their position in the conversation (the history the checkpointer replays is never recorded twice); the model’s answer, with its `usage_metadata`, after it
Tools	The middleware brings the history tools (and the agent memory tools) as its own `tools`, so `create_agent` offers them without you listing them
Verification	`conversation.verify()` before invoking the agent
Handoff	`transferred_to_agent()` and `transferred_to_human()` record the transfer; call them where the graph hands the conversation over

For the older langgraph.prebuilt.create_react_agent, pass pre_model_hook=pre_model_hook(conversation): it gives the model the same messages through llm_input_messages, again without touching state; record the turns with the NiadraCallbackHandler from niadra.integrations.langchain.

Minimal example

"""A LangGraph agent (langchain.agents.create_agent) with the customer's memory as middleware."""

import asyncio

from langchain.agents import create_agent

from niadra import AsyncNiadra, phone
from niadra.integrations.langgraph import NiadraMiddleware

niadra = AsyncNiadra(channel="chat")


async def main() -> None:
    async with niadra.conversation("thread-81", subject=phone("+5511912345678")) as conversation:
        agent = create_agent(
            "openai:gpt-4.1",
            system_prompt="You are Acme's agent.",
            middleware=[NiadraMiddleware(conversation)],
        )
        result = await agent.ainvoke(
            {"messages": [{"role": "user", "content": "Where is my replacement lid?"}]}
        )
        print(result["messages"][-1].content)
    await niadra.close()


asyncio.run(main())

The same code is in examples/langgraph_agent.py.

Agent memory

NiadraMiddleware(conversation, agent_memory=True) puts the agent’s own notes right before the customer’s context and adds search_agent_memory (and remember, with write) to the tools. See Agent memory.

Turn records

With the turns feature on in the space (it ships on), the NiadraMiddleware with turns=True records each agent run as a turn record: before_agent opens the turn, wrap_tool_call records each tool call by the provider’s id and after_agent closes it with the model calls and their tokens. In LangGraph.js, the NiadraCallbackHandler of @niadra/sdk/langchain with turns: true does the same. In a replay, the graph’s tools answer from the record in Python, and in JavaScript the ones passed through recordTools(). A turn never waits for Niadra and never repeats the conversation’s text.

Limits

No BaseStore of its own: Niadra is not the graph’s state store.
No separate node for LangGraph.js: withNiadraContext() in @niadra/sdk/langchain already gives the model node its context without writing it into the graph’s checkpointed state.
Nothing here fails the agent: with Niadra slow or down, the model call goes on without the pack.
Tested against langgraph 1.2 and langchain 1.4 with a fake chat model and Niadra on the emulator.

Get started

Voice

Channels

Agent frameworks

Model SDKs

Automation platforms

MCP clients

Install

The five primitives

Minimal example

Agent memory

Turn records

Limits

Next steps

LangChain

Internal agents