> ## Documentation Index
> Fetch the complete documentation index at: https://docs.niadra.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Langflow

> Three Langflow components in one file: the customer's context for the prompt, the recorded reply and the history search as a tool.

One Python file, `integrations-extras/langflow/niadra_component.py` in the Python SDK repository, brings three components:

* **Niadra Context** records the customer's message and outputs the context for the prompt: the agent's own notes (with **Agent memory** on), the customer's pack and what changed on other channels. Connect it to the system prompt of your agent or prompt template, after your own instructions.
* **Niadra Record Reply** records the agent's reply and passes it on unchanged.
* **Niadra Search History** searches the customer's whole history. In tool mode the agent calls it with a query (and optionally a time phrase); the customer is the one the flow configured, never a value the model chooses.

## Install

Copy the file into your Langflow components folder (or paste it into a custom component) and install the SDK where Langflow runs:

```sh theme={null}
pip install niadra
```

## The five primitives

| Primitive    | How the components wire it                                                            |
| ------------ | ------------------------------------------------------------------------------------- |
| Context      | Niadra Context, connected to the system prompt after your instructions                |
| Turns        | Niadra Context records the customer's message; Niadra Record Reply, the agent's reply |
| Tools        | Niadra Search History in tool mode, with the kit's own description                    |
| Verification | The verification level field, on the components                                       |
| Handoff      | Outside the components; use the SDK or the API                                        |

## Limits

* The three take the same connection fields: the source key, the channel, the customer's handle and the conversation id, usually mapped from the flow's inputs.
* When Niadra is slow or down, the context comes out empty and the search says the history is unavailable; the flow never stops.
* The components use the synchronous SDK and cover context, turns and search; the timeline, opening an item and handoffs are left to the SDK.
* Tested in `integrations-extras/langflow/test_niadra_component.py`, against the emulator.

## Next steps

<CardGroup cols={2}>
  <Card title="Flowise" href="/en/integrations/flowise">
    the Flowise memory and tool nodes.
  </Card>

  <Card title="Python SDK" href="/en/sdk/python">
    the library the components use.
  </Card>
</CardGroup>