AI Provider Guide

July 22, 2026 · View on GitHub

Setting up and configuring AI providers in SAM

Overview

SAM connects to AI providers to power its conversation and tool capabilities. You can use cloud providers, local models, or a mix of both. This guide covers setup, configuration, and best practices for each provider.

Quick Comparison

Provider	Cost	Speed	Privacy	Best For
OpenAI	Pay-per-token	Fast	Cloud	General use, coding, creative writing
Anthropic	-	-	-	No direct provider. Use OpenRouter for Claude models.
GitHub Copilot	Subscription	Fast	Cloud	Developers with existing Copilot subscription
DeepSeek	Pay-per-token (low cost)	Fast	Cloud	Budget-friendly general use
Google Gemini	Pay-per-token	Fast	Cloud	Google ecosystem, multimodal
MiniMax	Pay-per-token	Fast	Cloud	Cost-effective, large context
OpenRouter	Varies by model	Varies	Cloud	Access to 100+ models
Ollama Cloud	Pay-per-token	Varies	Cloud	Cloud-hosted Ollama models
Z.AI (Chat)	Pay-per-token	Fast	Cloud	GLM models, Chinese/English bilingual
Z.AI (Coding)	Pay-per-token	Fast	Cloud	GLM coding models
MLX (Local)	Free	Varies	Full privacy	Offline use, sensitive data, Apple Silicon
llama.cpp (Local)	Free	Varies	Full privacy	Offline use, Intel or Apple Silicon
Remote llama.cpp	Free	Varies	Self-hosted	Self-hosted inference servers
Custom Endpoint	Varies	Varies	Self-hosted	Self-hosted servers, Ollama, LM Studio

Cloud Providers

OpenAI

What you get: GPT-4o, GPT-4, GPT-3.5 Turbo, o1, o3 reasoning models

Setup:

Create an account at platform.openai.com
Go to API Keys and create a new key
In SAM Settings > AI Providers, add OpenAI
Paste your API key
Select your default model

Model Recommendations:

GPT-4o - Best overall balance of speed, quality, and cost
GPT-3.5 Turbo - Fast and cheap for simple tasks
o1 / o3 - Reasoning models for complex logic and math (slower, more expensive)

Pricing: See openai.com/pricing for current rates.

GitHub Copilot

What you get: Access to GPT-4o, Claude 3.5, o1, and other models through your Copilot subscription.

Setup:

You need an active GitHub Copilot subscription (Individual, Business, or Enterprise)
In SAM Settings > AI Providers, add GitHub Copilot
Click "Sign in with GitHub"
SAM uses the GitHub device flow - you'll see a code to enter on github.com
After authorization, SAM automatically manages token refresh

How It Works:

SAM authenticates via GitHub's device flow (no manual API key needed)
Tokens are refreshed automatically in the background
Available models depend on your Copilot subscription tier

Advantages:

No separate API key or billing
Uses your existing Copilot subscription
Access to multiple model providers through one authentication

DeepSeek

What you get: DeepSeek Chat, DeepSeek Coder

Setup:

Create an account at platform.deepseek.com
Create an API key
In SAM Settings > AI Providers, add DeepSeek
Paste your API key

Model Recommendations:

DeepSeek Chat - General conversation and tasks
DeepSeek Coder - Optimized for coding tasks

Advantages:

Significantly lower cost than OpenAI or Anthropic
Good quality for the price

Google Gemini

What you get: Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 2.0 Flash

Setup:

Get an API key from aistudio.google.com
In SAM Settings > AI Providers, add Gemini
Paste your API key
Select your default model

Model Recommendations:

Gemini 2.5 Pro - Top-tier reasoning and long context
Gemini 2.5 Flash - Fast and cost-effective
Gemini 2.0 Flash - Budget-friendly for simple tasks

Advantages:

Large context windows (up to 1M tokens on some models)
Strong multimodal capabilities
Competitive pricing

MiniMax

What you get: MiniMax-M3, MiniMax-M2.7, MiniMax-M2.5, and high-speed variants

Setup:

Create an account at minimax.io
Create an API key
In SAM Settings > AI Providers, add MiniMax
Paste your API key

Model Recommendations:

MiniMax-M3 - Latest model, best quality
MiniMax-M3-highspeed - Faster variant with slightly lower quality
MiniMax-M2.7 - Previous generation, still capable

Advantages:

128K token context window
Competitive pricing
Good tool use capabilities

OpenRouter

What you get: Access to 100+ models from multiple providers through a single API.

Setup:

Create an account at openrouter.ai
Create an API key
In SAM Settings > AI Providers, add OpenRouter
Paste your API key
Browse available models

Advantages:

One API key for dozens of providers
Try different models without separate accounts
Automatic routing and load balancing
Pay-per-token across all models

Recommended Model for New Users:

minimax/minimax-m2.7 - Excellent all-around model with 1M token context, strong reasoning, and competitive pricing. A great starting point when using OpenRouter.

Ollama Cloud

What you get: Cloud-hosted Ollama models without managing your own server.

Setup:

Create an account at ollama.com
Create an API key
In SAM Settings > AI Providers, add Ollama Cloud
Paste your API key
Select a model from the available list

Advantages:

No local infrastructure needed
Access to popular Ollama models in the cloud
Pay-per-token pricing

Z.AI (Chat)

What you get: Z.AI's GLM models optimized for general conversation and reasoning.

Setup:

Create an account at z.ai
Create an API key
In SAM Settings > AI Providers, add Z.AI (Chat)
Paste your API key

Models: GLM-5.1, GLM-4.9

Features:

Strong bilingual capabilities (Chinese/English)
Chain-of-thought reasoning
Competitive pricing

Z.AI (Coding)

What you get: Z.AI's GLM models optimized for coding tasks.

Setup:

Create an account at z.ai
Create an API key
In SAM Settings > AI Providers, add Z.AI (Coding)
Paste your API key

Models: GLM-5.1, GLM-4.9

Features:

Specialized for code generation and analysis
Supports chain-of-thought reasoning for complex coding tasks
Large context windows

Local Models

MLX (Apple Silicon Only)

What you get: Run language models directly on your Mac using Apple's MLX framework with Metal GPU acceleration.

Requirements:

Apple Silicon Mac (M1, M2, M3, M4)
8GB+ unified memory (16GB+ recommended)
macOS 14.0+

Setup:

In SAM Settings > AI Providers, click Add Provider
Choose "Local MLX Model"
Browse available models
Click Download on your chosen model
Wait for the download to complete
The model is ready to use

RAM Requirements:

Model Size	Minimum RAM	Recommended RAM
1-3B parameters	4GB	8GB
7B parameters	8GB	16GB
13B parameters	16GB	32GB
30B+ parameters	32GB	64GB
70B parameters	64GB	96GB+

Performance Tips:

Larger models are more capable but slower
Unified memory means the GPU shares RAM with the system
Close other memory-intensive apps for best performance
First generation is slower (model loading), subsequent ones are faster

Advantages:

Complete privacy - nothing leaves your Mac
No internet connection needed after download
No per-token costs
Fast inference on Apple Silicon

llama.cpp (Any Mac)

What you get: Run GGUF-format models on any Mac, including Intel.

Requirements:

Any Mac (Apple Silicon or Intel)
8GB+ RAM
macOS 14.0+

Setup:

Download a GGUF model file (from Hugging Face or other sources)
In SAM Settings > AI Providers, add llama.cpp
Point to the model file location
Configure context size and other parameters

Advantages:

Works on Intel Macs (unlike MLX)
Supports GGUF quantized models for lower memory usage
Wide model compatibility

Limitations:

Generally slower than MLX on Apple Silicon
Manual model file management

Remote llama.cpp

What you get: Connect to a remote llama.cpp server (e.g., running on a GPU server) for local-model inference without local hardware requirements.

Setup:

Run a llama.cpp server on your remote machine:

./llama-server -m model.gguf --host 0.0.0.0 --port 8080

In SAM Settings > AI Providers, add Remote llama.cpp
Enter the server URL (e.g., http://192.168.1.100:8080)
Configure authentication if needed

Use Cases:

Offload inference to a dedicated GPU server
Run larger models than your local machine can handle
Share a model server across multiple machines

Custom Endpoints

OpenAI-Compatible Servers

SAM can connect to any server that implements the OpenAI chat completions API. This includes:

Ollama - http://localhost:11434/v1
LM Studio - http://localhost:1234/v1
text-generation-webui - http://localhost:5000/v1
vLLM - http://localhost:8000/v1
Any OpenAI-compatible API

Setup:

In SAM Settings > AI Providers, add a Custom provider
Enter the endpoint URL (e.g., http://localhost:11434/v1)
Enter an API key if required (some local servers don't need one)
Configure the model name
Test the connection

Managing Multiple Providers

Switching Between Providers

You can have multiple providers configured simultaneously and switch between them:

Use the model selector in the toolbar to pick a different model/provider
Switch mid-conversation - the history carries forward
Each conversation remembers which model was last used

Strategy Recommendations

Use Case	Recommended Approach
Daily use	Cloud provider (GPT-4o or Claude) for quality and speed
Sensitive content	Local model (MLX) for complete privacy
Budget-conscious	DeepSeek, MiniMax, or local models for routine tasks, GPT-4o for complex ones
Coding	Claude or GPT-4o for best tool use, DeepSeek Coder for budget
Long documents	Claude (200K context), Gemini (1M context), or local models with large context
Offline use	Local models (MLX or llama.cpp)
Experimentation	OpenRouter for access to many models

Troubleshooting

"Authentication failed"

Verify your API key is correct
For GitHub Copilot: try signing out and back in
Check that your account has billing configured (cloud providers)

"Model not found"

The model may have been renamed or deprecated
Refresh the model list in Settings
Check the provider's documentation for current model names

"Rate limited"

You've exceeded the provider's rate limits
Wait a moment and try again
Consider upgrading your plan or using a different provider

"Request too large"

Your conversation has exceeded the model's context window
Start a new conversation
Use a model with a larger context window
SAM's context management should handle this automatically, but very long conversations with many tool calls can hit limits

Local model loading fails

Ensure you have enough free RAM
Try a smaller model
Check that the model file isn't corrupted (re-download if needed)
For MLX: verify you're on Apple Silicon
For llama.cpp: verify the file is in GGUF format