Overview

Fikra API is an OpenAI-compatible AI inference API that lets you run chat and embedding models behind one OpenAI-style endpoint. For retrieval augmented generation (RAG), you can send documents to the embeddings endpoint and get vector embeddings back in OpenAI format. Because the API is OpenAI-compatible, it works with anything that accepts an OpenAI-compatible base URL.

Prerequisites

A Fikra API key, stored in the environment variable FIKRA_API_KEY
Python with the openai SDK installed
Documents as plain text or a list of text strings

Step-by-step

Install the OpenAI SDK

bash pip install openai

Create the client

The base URL is https://api.fikraapi.co.ke/v1.

```python from openai import OpenAI

client = OpenAI( base_url="https://api.fikraapi.co.ke/v1", api_key=os.environ["FIKRA_API_KEY"], ) ```

Embed your documents

Choose a model from the Fikra API model list. For example, you can use fikra-auto.

```python documents = [ "Fikra API supports OpenAI-style chat completions.", "Embeddings are returned in OpenAI format.", ]

response = client.embeddings.create( model="fikra-auto", input=documents, )

for i, data in enumerate(response.data): print(i, len(data.embedding)) ```

Store and use the vectors

Store the returned embedding vectors in your vector database, then use them for retrieval in your RAG pipeline.

Verify it works

A successful call returns an OpenAI-style embeddings object with a data list. Each item in that list contains an embedding array and an index matching the order of your input documents. If you see one embedding per input document, the request worked.

Costs

Fikra API uses a token-credit wallet. You can top up with M-Pesa, cards via Paystack, or x402 USDC on Base for agents. Model pricing ranges from $0.09 to $6.00 per 1M input tokens and $0.15 to $12.00 per 1M output tokens depending on the model you choose. For exact embedding pricing and the full model list, check the pricing page.

Common errors

401 Unauthorized — The API key is missing, invalid, or inactive. Check that FIKRA_API_KEY is set and that your key has not been revoked.
400 Bad Request — The requested model is not supported or the request payload is invalid. Confirm the model name matches one in the model list.
429 Too Many Requests — You have hit the rate limit: 30 requests per minute for unverified accounts and 100 requests per minute for trusted accounts. Reduce request frequency or apply for trusted status.
503 Service Unavailable — The upstream service is busy. Retry the request with backoff.

Embed Documents for Retrieval Augmented Generation

Fact block — plain text for agents