Add LLM Embeddings to a Knowledge Base on a Budget

Overview

Add inexpensive vector embeddings to your knowledge base using the Fikra API. The service is OpenAI‑compatible, so you can use the standard openai Python SDK with a custom base URL. You pay per‑token in USD, and the cheapest model (deepseek-v4-flash-0731) costs $0.09 USD for input tokens.

Prerequisites

• Python 3.8+ and pip
• openai package (pip install openai)
• A Fikra API key stored in the environment variable FIKRA_API_KEY
• A list of text chunks you want to embed (e.g., documents, FAQ entries)

Step-by-step

1. Set up the SDK ```python import os, openai

openai.api_key = os.getenv("FIKRA_API_KEY") openai.base_url = "https://api.fikraapi.co.ke/v1" 2. **Prepare your texts**python texts = [ "What is the return policy?", "How do I reset my password?", "Pricing details for premium plans", ] 3. **Request embeddings**python response = openai.embeddings.create( model="deepseek-v4-flash-0731", input=texts, ) embeddings = [e.embedding for e in response.data] `` 4. **Store vectors** – insert eachembedding` into your vector store (e.g., Pinecone, Qdrant, or a simple SQLite + FAISS index). The exact storage code depends on your chosen backend, but the output format is a list of floats ready for indexing.

Verify it works

# Print the dimension of the first vector to confirm a successful call
print(len(embeddings[0]))

If you see a non‑zero length (typically 768‑1536 depending on the model), the embeddings are ready for use.

Costs

• Model: deepseek-v4-flash-0731
• Pricing: $0.09 USD per 1 M input tokens (embeddings only consume input tokens)
• Rate limits: 30 requests/min for unverified accounts, 100 requests/min for trusted accounts.
• You are billed in USD (or KES) via the token‑credit wallet; top‑ups are available through M‑Pesa, card, or USDC on Base.

Common errors

• 401 Unauthorized – API key missing, malformed, or inactive. Verify FIKRA_API_KEY is set correctly.
• 400 Bad Request – Model name not supported or request payload malformed. Ensure you use a model from the pricing list.
• 429 Too Many Requests – Exceeded the rate limit (30/100 req/min). Add a short sleep or batch calls.
• 503 Service Unavailable – Upstream is busy. Retry with exponential back‑off.