> For the complete documentation index, see [llms.txt](https://flashfi.gitbook.io/flashfi/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://flashfi.gitbook.io/flashfi/contribution-to-covalent-agent-kit.md).

# Contribution To Covalent Agent Kit

## Overview

This guide explains how I contributed to Covalent's AI Agent SDK by integrating **Ollama**, allowing the use of locally running **LLMs (Large Language Models)**. This integration enables developers to run AI agents without relying on paid API keys, making AI-driven automation **more accessible and cost-effective**. Here is the link to my [PR ](https://github.com/covalenthq/ai-agent-sdk/pull/46)

## Benefits of Ollama Integration

* **No API Costs**: Since Ollama runs locally, there is no need to pay for external LLM API services.
* **Increased Privacy**: Data remains on the local machine, enhancing security and confidentiality.
* **Customization & Control**: Developers can fine-tune models and experiment without restrictions imposed by third-party services.
* **Offline Capability**: AI agents can function even without an active internet connection.

## Technical Implementation

### 1. Defining the Ollama Configuration

I introduced a new provider type, `OLLAMA`, to the existing model configuration:

{% code overflow="wrap" %}

```typescript
type OllamaConfig = {
    provider: "OLLAMA";
    name: OllamaModel;
    toolChoice?: "auto" | "required";
    temperature?: number;
    apiKey?: string;
    baseURL?: string; // Option to override base URL
};
```

{% endcode %}

### 2. Handling Ollama API Calls

Since Ollama runs locally, API requests are sent to `http://localhost:11434/api/chat`. Here’s how I implemented the request logic:

{% code overflow="wrap" %}

```typescript
const response = await fetch(`${config.baseURL}/api/chat`, {
    method: "POST",
    headers: { "Content-Type": "application/json" },
    body: JSON.stringify({
        model: this.model.name,
        messages: messages.map((msg) => ({ role: msg.role, content: msg.content || "" })),
        stream: true,
    }),
});
```

{% endcode %}

### 3. Parsing Responses Efficiently

I structured the response parsing to extract relevant information:

{% code overflow="wrap" %}

```typescript
const formatResponse = (content: string): FormattedResponse => {
    const parts = content.split("</think>");
    return {
        thinking: parts[0]?.replace("<think>", "").trim() || "",
        response: parts[1]?.trim() || content.trim(),
    };
};
```

{% endcode %}

### Future Enhancements

* Enhancing logging for better debugging and transparency.
* Optimizing response parsing for performance improvements.
* Implementing function calls

By integrating **Ollama**, developers can now use AI agents **free of cost**, making AI automation truly accessible to everyone!