crewAI/docs/concepts/llms.mdx

---
title: LLMs
description: Learn how to configure and optimize LLMs for your CrewAI projects.
icon: microchip-ai
---

# Large Language Models (LLMs) in CrewAI

Large Language Models (LLMs) are the backbone of intelligent agents in the CrewAI framework. This guide will help you understand, configure, and optimize LLM usage for your CrewAI projects.

## Key Concepts

- **LLM**: Large Language Model, the AI powering agent intelligence
- **Agent**: A CrewAI entity that uses an LLM to perform tasks
- **Provider**: A service that offers LLM capabilities (e.g., OpenAI, Anthropic, Ollama, [more providers](https://docs.litellm.ai/docs/providers))

## Configuring LLMs for Agents

CrewAI offers flexible options for setting up LLMs:

### 1. Default Configuration

By default, CrewAI uses the `gpt-4o-mini` model. It uses environment variables if no LLM is specified:
- `OPENAI_MODEL_NAME` (defaults to "gpt-4o-mini" if not set)
- `OPENAI_API_BASE`
- `OPENAI_API_KEY`

### 2. String Identifier

```python Code
agent = Agent(llm="gpt-4o", ...)
```

### 3. LLM Instance

List of [more providers](https://docs.litellm.ai/docs/providers).

```python Code
from crewai import LLM

llm = LLM(model="gpt-4", temperature=0.7)
agent = Agent(llm=llm, ...)
```

### 4. Custom LLM Objects

Pass a custom LLM implementation or object from another library.

## Connecting to OpenAI-Compatible LLMs

You can connect to OpenAI-compatible LLMs using either environment variables or by setting specific attributes on the LLM class:

1. Using environment variables:

```python Code
import os

os.environ["OPENAI_API_KEY"] = "your-api-key"
os.environ["OPENAI_API_BASE"] = "https://api.your-provider.com/v1"
```

2. Using LLM class attributes:

```python Code
from crewai import LLM

llm = LLM(
    model="custom-model-name",
    api_key="your-api-key",
    base_url="https://api.your-provider.com/v1"
)
agent = Agent(llm=llm, ...)
```

## LLM Configuration Options

When configuring an LLM for your agent, you have access to a wide range of parameters:

| Parameter         | Type            | Description                                                                                      |
|:------------------|:---------------:|:-------------------------------------------------------------------------------------------------|
| **model**            | `str`             | Name of the model to use (e.g., "gpt-4", "gpt-3.5-turbo", "ollama/llama3.1"). For more options, visit the providers documentation. |
| **timeout**          | `float, int`      | Maximum time (in seconds) to wait for a response.                                                |
| **temperature**      | `float`           | Controls randomness in output (0.0 to 1.0).                                                      |
| **top_p**            | `float`           | Controls diversity of output (0.0 to 1.0).                                                       |
| **n**                | `int`             | Number of completions to generate.                                                               |
| **stop**             | `str, List[str]`  | Sequence(s) where generation should stop.                                                        |
| **max_tokens**       | `int`             | Maximum number of tokens to generate.                                                            |
| **presence_penalty** | `float`           | Penalizes new tokens based on their presence in prior text.                                      |
| **frequency_penalty**| `float`           | Penalizes new tokens based on their frequency in prior text.                                     |
| **logit_bias**       | `Dict[int, float]`| Modifies likelihood of specified tokens appearing.                                              |
| **response_format**  | `Dict[str, Any]`  | Specifies the format of the response (e.g., JSON object).                                        |
| **seed**             | `int`             | Sets a random seed for deterministic results.                                                    |
| **logprobs**         | `bool`            | Returns log probabilities of output tokens if enabled.                                           |
| **top_logprobs**     | `int`             | Number of most likely tokens for which to return log probabilities.                              |
| **base_url**         | `str`             | The base URL for the API endpoint.                                                               |
| **api_version**      | `str`             | Version of the API to use.                                                                       |
| **api_key**          | `str`             | Your API key for authentication.                                                                 |


## OpenAI Example Configuration

```python Code
from crewai import LLM

llm = LLM(
    model="gpt-4",
    temperature=0.8,
    max_tokens=150,
    top_p=0.9,
    frequency_penalty=0.1,
    presence_penalty=0.1,
    stop=["END"],
    seed=42,
    base_url="https://api.openai.com/v1",
    api_key="your-api-key-here"
)
agent = Agent(llm=llm, ...)
```

## Cerebras Example Configuration

```python Code
from crewai import LLM

llm = LLM(
    model="cerebras/llama-3.1-70b",
    base_url="https://api.cerebras.ai/v1",
    api_key="your-api-key-here"
)
agent = Agent(llm=llm, ...) 
```

## Using Ollama (Local LLMs)

CrewAI supports using Ollama for running open-source models locally:

1. Install Ollama: [ollama.ai](https://ollama.ai/)
2. Run a model: `ollama run llama2`
3. Configure agent:

```python Code
from crewai import LLM

agent = Agent(
    llm=LLM(model="ollama/llama3.1", base_url="http://localhost:11434"),
    ...
)
```

## Changing the Base API URL

You can change the base API URL for any LLM provider by setting the `base_url` parameter:

```python Code
from crewai import LLM

llm = LLM(
    model="custom-model-name",
    base_url="https://api.your-provider.com/v1",
    api_key="your-api-key"
)
agent = Agent(llm=llm, ...)
```

This is particularly useful when working with OpenAI-compatible APIs or when you need to specify a different endpoint for your chosen provider.

## Best Practices

1. **Choose the right model**: Balance capability and cost.
2. **Optimize prompts**: Clear, concise instructions improve output.
3. **Manage tokens**: Monitor and limit token usage for efficiency.
4. **Use appropriate temperature**: Lower for factual tasks, higher for creative ones.
5. **Implement error handling**: Gracefully manage API errors and rate limits.

## Troubleshooting

- **API Errors**: Check your API key, network connection, and rate limits.
- **Unexpected Outputs**: Refine your prompts and adjust temperature or top_p.
- **Performance Issues**: Consider using a more powerful model or optimizing your queries.
- **Timeout Errors**: Increase the `timeout` parameter or optimize your input.