---
title: "AI Model Providers"
source: https://docs.autohand.ai/integrations/ai-model-providers
---

# AI Model Providers

Connect Autohand to your preferred AI models. Use cloud APIs for maximum capability or run models locally for privacy and offline access.

## Overview

Autohand separates the CLI interface from the AI model provider. This means you can switch between different models and providers without changing how you work. Whether you prefer cloud-hosted models or local inference, Autohand adapts to your needs.

You have two main options:

-   **Cloud providers** - Access powerful models via API with no hardware requirements
-   **Local providers** - Run models on your own machine for privacy and offline use

## Cloud providers

Cloud providers offer access to the most capable models without requiring powerful hardware. Your prompts are sent to remote servers for processing.

[

### OpenRouter

Recommended

Access 200+ models from Anthropic, OpenAI, Google, and more through a single API. Automatic fallbacks and cost optimization included.

-   Claude, GPT-4, Gemini, Llama, and more
-   Single API key for all providers
-   Automatic model fallback
-   Pay-per-use pricing

](/docs/integrations/openrouter.html)[

### OpenAI

Direct access to GPT-4o, GPT-4 Turbo, and o1 reasoning models. The industry standard for AI development with excellent documentation.

-   GPT-4o, GPT-4 Turbo, o1 reasoning
-   128K context window
-   Function calling support
-   Vision capabilities

](/docs/integrations/openai.html)[

### AWS Bedrock

Enterprise

Run Autohand through Bedrock Converse or Bedrock's OpenAI-compatible endpoints. Designed for AWS credential chains, profiles, private endpoints, and model access controls.

-   Converse, Chat Completions, and Responses modes
-   AWS credentials, SSO profiles, IAM roles, or Bedrock API keys
-   Inference profiles and ARN model IDs
-   Regional and private VPC endpoints

](/docs/integrations/aws-bedrock.html)[

### DeepSeek

Direct access to DeepSeek V4 models through an OpenAI-compatible API. A strong choice for coding, reasoning, and cost-sensitive engineering workflows.

-   DeepSeek V4 Flash and V4 Pro
-   OpenAI-compatible chat completions
-   Fast coding and reasoning workflows
-   Simple API-key setup

](/docs/integrations/deepseek.html)[

### xAI Grok

Access xAI's Grok models with real-time knowledge and advanced reasoning. Built for developers who want cutting-edge AI assistance.

-   Grok-2 and Grok-2 mini
-   Real-time knowledge from X data
-   128K context window
-   Vision capabilities

](/docs/integrations/xai.html)[

### Z.ai

Use GLM-family models through Z.ai's OpenAI-compatible API for high-throughput coding and agentic workflows.

-   GLM coding and reasoning models
-   OpenAI-compatible API surface
-   High-performance inference
-   API-key authentication

](/docs/integrations/zai.html)[

### Sakana AI

Use Fugu's multi-agent orchestration models through Sakana AI's OpenAI-compatible API for complex reasoning and long-horizon coding tasks.

-   Fugu and Fugu Ultra orchestration models
-   OpenAI-compatible API surface
-   1M token context window
-   API-key authentication

](/docs/integrations/sakana.html)[

### Cerebras

Fastest

Lightning-fast inference at over 2000 tokens per second. Powered by the world's largest AI chip for near-instant responses.

-   Llama 3.1 8B and 70B models
-   2000+ tokens per second
-   128K context window
-   Competitive pricing

](/docs/integrations/cerebras.html)[

### NVIDIA

Connect to NVIDIA-hosted models through NVIDIA's OpenAI-compatible inference endpoint, including DeepSeek, Mistral, and NVIDIA model families.

-   NVIDIA-hosted model catalog
-   OpenAI-compatible chat completions
-   GPU-backed inference infrastructure
-   Good fit for enterprise GPU stacks

](/docs/integrations/nvidia.html)[

### LLMGateway

Enterprise

Enterprise AI gateway with unified API access to multiple providers. Smart routing, caching, and observability for production applications.

-   Multi-provider access
-   Automatic failover
-   Response caching
-   Cost optimization

](/docs/integrations/llmgateway.html)[

### Azure AI Foundry

Enterprise

Enterprise-grade AI models with Azure security and compliance. Access OpenAI, Llama, and Mistral through Azure infrastructure.

-   GPT-4o, Llama, Mistral models
-   API Key, Entra ID, and Managed Identity auth
-   Regional data residency
-   SOC 2, HIPAA, GDPR compliance

](/docs/integrations/azure-foundry.html)[

### GCP Vertex AI

Enterprise

Google's Gemini models and open-source alternatives through Vertex AI. Enterprise infrastructure with Google Cloud security.

-   Gemini 1.5 Pro, 2.0 Flash
-   2 million token context
-   Model Garden open-source models
-   Google Cloud IAM

](/docs/integrations/gcp-vertex.html)

**Default choice:** OpenRouter is the default provider for Autohand CLI. It gives you immediate access to the best models with a single API key.

## Local providers

Local providers run AI models directly on your machine. Your code never leaves your computer, and you can work without an internet connection.

[

### Ollama

Easy setup

The simplest way to run local models. One-command installation with automatic GPU detection on macOS, Linux, and Windows.

-   Simple installation and model management
-   Cross-platform support
-   Automatic GPU acceleration
-   Large model library

](/docs/integrations/ollama.html)[

### MLX

Apple Silicon

Apple's machine learning framework optimized for M1, M2, M3, and M4 chips. Native Metal acceleration for exceptional performance.

-   Native Apple Silicon optimization
-   Unified memory efficiency
-   Metal GPU acceleration
-   macOS only

](/docs/integrations/mlx.html)[

### llama.cpp

Advanced

High-performance C++ inference engine. Maximum flexibility and control for power users who want to fine-tune every parameter.

-   Maximum performance potential
-   GGUF model format
-   Extensive configuration options
-   Cross-platform support

](/docs/integrations/llama-cpp.html)

## Choosing a provider

Select a provider based on your priorities:

| Priority | Recommended Provider |
|---|---|
| Best model quality | OpenRouter or OpenAI |
| Fastest inference | Cerebras (2000+ tokens/sec) |
| AWS enterprise deployment | AWS Bedrock |
| Enterprise compliance | AWS Bedrock, Azure AI Foundry, or GCP Vertex AI |
| Long context (2M tokens) | GCP Vertex AI (Gemini) |
| Multi-provider management | LLMGateway |
| Cost-sensitive coding models | DeepSeek or Z.ai |
| Multi-agent orchestration | Sakana AI |
| NVIDIA-hosted inference | NVIDIA |
| Real-time knowledge | xAI Grok |
| Complete privacy | Ollama or MLX |
| Offline work | Ollama, MLX, or llama.cpp |
| No API costs | Any local provider |
| Apple Silicon performance | MLX |

## Switching providers

Switch between providers at any time using the CLI or configuration file.

### Using CLI flags

``` bash
# Use OpenRouter (default)
autohand --provider openrouter --model nvidia/nemotron-3-super-120b-a12b:free

# Use AWS Bedrock
AWS_PROFILE=enterprise-prod autohand --provider bedrock --model us.anthropic.claude-3-5-sonnet-20241022-v2:0

# Use DeepSeek
autohand --provider deepseek --model deepseek-v4-flash

# Use Ollama
autohand --provider ollama --model codellama:13b

# Use MLX
autohand --provider mlx --model codellama-7b-instruct

# Use llama.cpp
autohand --provider llamacpp --model codellama-7b
```

### Using configuration

Set your default provider in `~/.autohand/config.json`:

``` json
{
  "provider": "openrouter",
  "model": "nvidia/nemotron-3-super-120b-a12b:free"
}
```

### During a session

Switch models without restarting:

``` bash
# Switch to a different model
/model ollama/codellama:13b

# Switch to a different provider
/model openrouter/nvidia/nemotron-3-super-120b-a12b:free

# Configure enterprise provider details interactively
/model bedrock
```

## Provider comparison

### Cloud providers

| Provider | Best for | Authentication | API style |
|---|---|---|---|
| OpenRouter | Broad model access | API key | OpenAI-compatible |
| OpenAI | Direct OpenAI models | API key or ChatGPT auth | OpenAI native |
| AWS Bedrock | AWS enterprise workloads | AWS credentials/profile/IAM or Bedrock API key | Converse and OpenAI-compatible |
| DeepSeek | Coding and reasoning value | API key | OpenAI-compatible |
| xAI Grok | Grok models | API key | OpenAI-compatible |
| Z.ai | GLM-family inference | API key | OpenAI-compatible |
| Sakana AI | Multi-agent orchestration | API key | OpenAI-compatible |
| Cerebras | Low-latency inference | API key | OpenAI-compatible |
| NVIDIA | GPU-hosted model catalog | API key | OpenAI-compatible |
| Azure AI Foundry | Azure enterprise deployments | API key, Entra ID, or Managed Identity | Azure OpenAI |
| GCP Vertex AI | Google Cloud and Gemini | Google Cloud access token | Vertex AI OpenAPI endpoint |

### Local providers

| Feature | Ollama | MLX | llama.cpp |
|---|---|---|---|
| Setup difficulty | Easy | Medium | Advanced |
| Model quality | Good | Good | Good |
| Offline capable | Yes | Yes | Yes |
| Cost | Free | Free | Free |
| macOS | Yes | Apple Silicon | Yes |
| Linux | Yes | No | Yes |
| Windows | Yes | No | Yes |