Skip to content

langflow

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

langflow

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
.env.example
.env.example
 
 
README.md
README.md
 
 
moss_langflow.py
moss_langflow.py
 
 
pyproject.toml
pyproject.toml
 
 
test_integration.py
test_integration.py
 
 

README.md

Moss Langflow Cookbook

This cookbook provides drag-and-drop Moss retrieval components for Langflow.

Overview

Moss is a sub-10 ms semantic search runtime. These Langflow custom components let visual-builder users add Moss retrieval to their flows without writing code.

Two components are included:

  1. Moss Retriever — returns structured Data objects (text, score, metadata) for downstream pipeline use.
  2. Moss Search — returns formatted text ready for direct LLM prompt injection.

Installation

Option A: Paste into Langflow (quickest)

  1. Open Langflow and create a new flow.
  2. Add a Custom Component node to the canvas.
  3. Open the code editor and paste the contents of moss_langflow.py.
  4. The component will appear with configurable inputs.

Option B: Install as a package

pip install langflow-moss

Or install from source:

cd examples/cookbook/langflow
pip install -e .

The components will be available in Langflow's component sidebar after restart.

Setup

You need a Moss project. Sign up at moss.dev to get your credentials (free tier available).

Configure credentials either:

  • In the component UI: Fill in the Moss Project ID and Moss Project Key fields directly.
  • Via environment variables: Set MOSS_PROJECT_ID and MOSS_PROJECT_KEY before starting Langflow.
cp .env.example .env
# Edit .env with your credentials

Component Reference

Moss Retriever

Returns a list of Data objects, each containing:

Field Type Description
text str Document content
score float Relevance score (0–1)
id str Document ID
metadata dict Document metadata

Inputs:

Input Type Default Description
Moss Project ID Text $MOSS_PROJECT_ID Your Moss project ID
Moss Project Key Secret $MOSS_PROJECT_KEY Your Moss project key (masked in UI)
Index Name Text (required) Name of the Moss index to query
Search Query Text (required) The search query text
Top K Integer 5 Number of results to return
Alpha (Hybrid Search) Float 0.5 0.0 = keyword, 1.0 = semantic, 0.5 = balanced
Metadata Filter (JSON) Text (empty) Advanced: Moss filter as JSON

Moss Search

Returns a Message with formatted text like:

Result 1 (score: 0.923):
Refunds are processed within 3-5 business days.

Result 2 (score: 0.847):
You can track your order on the dashboard.

Same inputs as the Moss Retriever.

Example Flow

A typical RAG flow in Langflow:

[Chat Input] → [Moss Retriever] → [Prompt Builder] → [OpenAI LLM] → [Chat Output]
  1. Chat Input: User asks a question.
  2. Moss Retriever: Searches your knowledge base with sub-10 ms latency.
  3. Prompt Builder: Combines the retrieved documents with the user's question.
  4. OpenAI LLM: Generates an answer grounded in the retrieved context.
  5. Chat Output: Returns the answer to the user.

Alternatively, use the Moss Search component to get pre-formatted text and wire it directly into a prompt template.

Advanced: Metadata Filtering

Use the Metadata Filter (JSON) input to narrow results:

{"$eq": {"category": "faq"}}
{"$and": [{"$eq": {"department": "support"}}, {"$eq": {"language": "en"}}]}

See Moss metadata filtering docs for the full filter syntax.

Related