Langcorn

LangCorn is an API server that enables you to serve LangChain models and pipelines with ease, leveraging the power of FastAPI for a robust and efficient experience.

Features

Easy deployment of LangChain models and pipelines
Ready to use auth functionality
High-performance FastAPI framework for serving requests
Scalable and robust solution for language processing applications
Supports custom pipelines and processing
Well-documented RESTful API endpoints
Asynchronous processing for faster response times

📦 Installation

To get started with LangCorn, simply install the package using pip:

pip install langcorn

⛓️ Quick Start

Example LLM chain ex1.py

import os

from langchain import LLMMathChain, OpenAI

os.environ["OPENAI_API_KEY"] = os.environ.get("OPENAI_API_KEY", "sk-********")

llm = OpenAI(temperature=0)
chain = LLMMathChain(llm=llm, verbose=True)

Run your LangCorn FastAPI server:

langcorn server examples.ex1:chain


[INFO] 2023-04-18 14:34:56.32 | api:create_service:75 | Creating service
[INFO] 2023-04-18 14:34:57.51 | api:create_service:85 | lang_app='examples.ex1:chain':LLMChain(['product'])
[INFO] 2023-04-18 14:34:57.51 | api:create_service:104 | Serving
[INFO] 2023-04-18 14:34:57.51 | api:create_service:106 | Endpoint: /docs
[INFO] 2023-04-18 14:34:57.51 | api:create_service:106 | Endpoint: /examples.ex1/run
INFO:     Started server process [27843]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://127.0.0.1:8718 (Press CTRL+C to quit)

or as an alternative

python -m langcorn server examples.ex1:chain

Run multiple chains

python -m langcorn server examples.ex1:chain examples.ex2:chain


[INFO] 2023-04-18 14:35:21.11 | api:create_service:75 | Creating service
[INFO] 2023-04-18 14:35:21.82 | api:create_service:85 | lang_app='examples.ex1:chain':LLMChain(['product'])
[INFO] 2023-04-18 14:35:21.82 | api:create_service:85 | lang_app='examples.ex2:chain':SimpleSequentialChain(['input'])
[INFO] 2023-04-18 14:35:21.82 | api:create_service:104 | Serving
[INFO] 2023-04-18 14:35:21.82 | api:create_service:106 | Endpoint: /docs
[INFO] 2023-04-18 14:35:21.82 | api:create_service:106 | Endpoint: /examples.ex1/run
[INFO] 2023-04-18 14:35:21.82 | api:create_service:106 | Endpoint: /examples.ex2/run
INFO:     Started server process [27863]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://127.0.0.1:8718 (Press CTRL+C to quit)

Import the necessary packages and create your FastAPI app:

from fastapi import FastAPI
from langcorn import create_service

app:FastAPI = create_service("examples.ex1:chain")

Multiple chains

from fastapi import FastAPI
from langcorn import create_service

app:FastAPI = create_service("examples.ex2:chain", "examples.ex1:chain")

from fastapi import FastAPI
from langcorn import create_service

app: FastAPI = create_service(
    "examples.ex1:chain",
    "examples.ex2:chain",
    "examples.ex3:chain",
    "examples.ex4:sequential_chain",
    "examples.ex5:conversation",
    "examples.ex6:conversation_with_summary",
    "examples.ex7_agent:agent",
)

Run your LangCorn FastAPI server:

uvicorn main:app --host 0.0.0.0 --port 8000

Now, your LangChain models and pipelines are accessible via the LangCorn API server.

Docs

Automatically served FastAPI doc Live example hosted on vercel.

Auth

It possible to add a static api token auth by specifying auth_token

python langcorn server examples.ex1:chain examples.ex2:chain --auth_token=api-secret-value

app:FastAPI = create_service("examples.ex1:chain", auth_token="api-secret-value")

Custom API KEYs

POST http://0.0.0.0:3000/examples.ex6/run
X-LLM-API-KEY: sk-******
Content-Type: application/json

Handling memory

{
  "history": "string",
  "input": "What is brain?",
  "memory": [
    {
      "type": "human",
      "data": {
        "content": "What is memory?",
        "additional_kwargs": {}
      }
    },
    {
      "type": "ai",
      "data": {
        "content": " Memory is the ability of the brain to store, retain, and recall information. It is the capacity to remember past experiences, facts, and events. It is also the ability to learn and remember new information.",
        "additional_kwargs": {}
      }
    }
  ]
}

Response:

{
  "output": " The brain is an organ in the human body that is responsible for controlling thought, memory, emotion, and behavior. It is composed of billions of neurons that communicate with each other through electrical and chemical signals. It is the most complex organ in the body and is responsible for all of our conscious and unconscious actions.",
  "error": "",
  "memory": [
    {
      "type": "human",
      "data": {
        "content": "What is memory?",
        "additional_kwargs": {}
      }
    },
    {
      "type": "ai",
      "data": {
        "content": " Memory is the ability of the brain to store, retain, and recall information. It is the capacity to remember past experiences, facts, and events. It is also the ability to learn and remember new information.",
        "additional_kwargs": {}
      }
    },
    {
      "type": "human",
      "data": {
        "content": "What is brain?",
        "additional_kwargs": {}
      }
    },
    {
      "type": "ai",
      "data": {
        "content": " The brain is an organ in the human body that is responsible for controlling thought, memory, emotion, and behavior. It is composed of billions of neurons that communicate with each other through electrical and chemical signals. It is the most complex organ in the body and is responsible for all of our conscious and unconscious actions.",
        "additional_kwargs": {}
      }
    }
  ]
}

LLM kwargs

To override the default LLM params per request

POST http://0.0.0.0:3000/examples.ex1/run
X-LLM-API-KEY: sk-******
X-LLM-TEMPERATURE: 0.7
X-MAX-TOKENS: 256
X-MODEL-NAME: gpt5
Content-Type: application/json

Custom run function

See ex12.py

chain = LLMChain(llm=llm, prompt=prompt, verbose=True)


# Run the chain only specifying the input variable.


def run(query: str) -> Joke:
    output = chain.run(query)
    return parser.parse(output)

app: FastAPI = create_service("examples.ex12:run")

Documentation

For more detailed information on how to use LangCorn, including advanced features and customization options, please refer to the official documentation.