Integrating Fikra API with LangChain

This guide demonstrates how to use the Fikra API as a custom OpenAI-compatible endpoint within the LangChain framework. By configuring the base_url, you can seamlessly swap out the default OpenAI provider for Fikra's hosted models.

Prerequisites

1. Python 3.8+ installed.
2. LangChain and OpenAI packages installed: bash pip install langchain openai
3. A valid Fikra API Key.

Step 1: Install Dependencies

Ensure you have the latest versions of the required libraries:

pip install --upgrade langchain openai

Step 2: Configure the Custom Base URL

The key to using Fikra API with LangChain is overriding the base_url parameter in the ChatOpenAI class. The Fikra API base URL is https://api.fikraapi.co.ke/v1.

Python Implementation

Below is the exact code block to initialize the LLM with the Fikra API endpoint:

from langchain_openai import ChatOpenAI

# Initialize the LLM with Fikra API as the custom OpenAI endpoint
llm = ChatOpenAI(
    model="fikra-model-name",  # Replace with the specific Fikra model ID
    api_key="your-fikra-api-key",  # Replace with your actual Fikra API key
    base_url="https://api.fikraapi.co.ke/v1"  # Critical: Point to Fikra API
)

# Test the connection
response = llm.invoke("Hello, Fikra! How are you?")
print(response.content)

JavaScript Implementation

If you are using LangChain.js, the configuration is similar:

import { ChatOpenAI } from "@langchain/openai";

// Initialize the LLM with Fikra API as the custom OpenAI endpoint
const llm = new ChatOpenAI({
  model: "fikra-model-name", // Replace with the specific Fikra model ID
  apiKey: "your-fikra-api-key", // Replace with your actual Fikra API key
  configuration: {
    baseURL: "https://api.fikraapi.co.ke/v1" // Critical: Point to Fikra API
  }
});

// Test the connection
const response = await llm.invoke("Hello, Fikra! How are you?");
console.log(response.content);

Step 3: Verify the Integration

Run the script. If configured correctly, you should receive a response from the Fikra model. If you encounter a 401 Unauthorized error, double-check your API key. If you receive a 404 Not Found error, ensure the base_url is exactly https://api.fikraapi.co.ke/v1.

Troubleshooting

• Connection Errors: Ensure your network allows outbound HTTPS traffic to api.fikraapi.co.ke.
• Model Not Found: Verify that the model parameter matches an available model on the Fikra platform.
• Timeouts: For large prompts, consider increasing the timeout parameter in the ChatOpenAI constructor.

Conclusion

By simply changing the base_url to https://api.fikraapi.co.ke/v1, you can leverage the power of Fikra's models within any LangChain application designed for OpenAI. This approach allows for easy switching between providers without refactoring your core logic.