---
title: "Model Selection Guide"
source: https://docs.autohand.ai/guides/model-selection
---

# Choose a model provider

Start with privacy, governance, and runtime constraints. Then open the canonical provider guide and select a model available to that account or local runtime.

Choose by constraint

## Start with the decision you already know

Pick the operating constraint first. Choose a specific model only after the provider connection works, because model IDs, access, pricing, and limits change independently.

01

### Start with the broadest hosted catalog

**OpenRouter.** One account gives you room to compare supported hosted model families before standardizing.

[OpenRouter](/integrations/openrouter.html)

02

### Keep existing cloud governance

**AWS, Azure, or Google Cloud.** Use the identity, regional, networking, and billing controls your organization already operates.

[AWS Bedrock](/integrations/aws-bedrock.html) · [Azure AI Foundry](/integrations/azure-foundry.html) · [GCP Vertex AI](/integrations/gcp-vertex.html)

03

### Use a direct model-provider account

**Choose the provider your team already trusts.** Direct integrations keep credentials, quotas, and support relationships with that provider.

[OpenAI](/integrations/openai.html) · [DeepSeek](/integrations/deepseek.html) · [xAI Grok](/integrations/xai.html) · [Z.ai](/integrations/zai.html) · [Sakana AI](/integrations/sakana.html) · [Cerebras](/integrations/cerebras.html) · [NVIDIA](/integrations/nvidia.html)

04

### Keep inference on hardware you control

**Ollama, MLX, or llama.cpp.** Local runtimes avoid a hosted model credential but make your team responsible for models and capacity.

[Ollama](/integrations/ollama.html) · [MLX](/integrations/mlx.html) · [llama.cpp](/integrations/llama-cpp.html)

Current integrations

## Compare provider paths

Every row opens the canonical setup guide. No model IDs or prices are duplicated here.

### Hosted providers

Use an account, cloud identity, or gateway credential.

[OpenRouter Trying and switching among hosted model families through one account. OPENROUTER\_API\_KEY →](/integrations/openrouter.html) [OpenAI Teams that want a direct OpenAI account and billing relationship. OPENAI\_API\_KEY →](/integrations/openai.html) [AWS Bedrock Organizations already governing model access through AWS. AWS credentials or API key →](/integrations/aws-bedrock.html) [Azure AI Foundry Organizations using Azure identity, networking, and regional controls. Azure endpoint + credentials →](/integrations/azure-foundry.html) [GCP Vertex AI Teams operating model access through a Google Cloud project. Google Cloud credentials →](/integrations/gcp-vertex.html) [DeepSeek Direct access to DeepSeek models through a compatible hosted API. DEEPSEEK\_API\_KEY →](/integrations/deepseek.html) [xAI Grok Direct use of Grok models from an xAI account. XAI\_API\_KEY →](/integrations/xai.html) [Z.ai Using GLM-family models through a compatible hosted endpoint. ZAI\_API\_KEY →](/integrations/zai.html) [Sakana AI Accounts using supported Sakana AI orchestration models. SAKANA\_API\_KEY →](/integrations/sakana.html) [Cerebras Hosted inference through a Cerebras account. CEREBRAS\_API\_KEY →](/integrations/cerebras.html) [NVIDIA Hosted model inference through NVIDIA services. NVIDIA\_API\_KEY →](/integrations/nvidia.html) [LLMGateway Centralized routing, policy, caching, and observability for teams. LLMGATEWAY\_API\_KEY →](/integrations/llmgateway.html)

### Local runtimes

Keep inference on hardware you operate.

[Ollama The quickest path to private local inference on common platforms. Ollama host →](/integrations/ollama.html) [MLX Private local inference optimized for Apple Silicon. Apple Silicon →](/integrations/mlx.html) [llama.cpp Portable, configurable GGUF inference on hardware you control. llama.cpp server →](/integrations/llama-cpp.html)

This directory is generated from the same provider manifest as the [integrations directory](/integrations/).

Change scope safely

## Switch models without rewriting setup

Session

### Try a model now

Use `/model` inside an interactive session, then select a model exposed by the active provider.

Command

### Override one run

Pass `--model <model-id>` when launching Autohand so the saved default remains unchanged.

Default

### Standardize future runs

Set the provider and model in Autohand configuration only after the smallest verification prompt succeeds.

Before standardizing

## Validate the choice against your workload

1.  01
    
    **Run a representative task**
    
    Use code from the repository and workflow you actually plan to support.
    
2.  02
    
    **Measure the constraints that matter**
    
    Record completion quality, latency, tool use, context handling, and cost or hardware load.
    
3.  03
    
    **Confirm production controls**
    
    Review credential handling, data terms, quotas, regional availability, and fallback behavior.
    

[Compare Autohand-hosted models →](/models/) [Review provider configuration patterns →](/integrations/ai-model-providers.html) [Plan model cost controls →](/guides/cost-optimization.html)