Skip to content

AI Providers

Complete setup guide for all supported AI providers.

Overview

The AI service supports multiple providers through either Pydantic AI or LangChain (see Engines), giving you flexibility to choose based on your needs: cost, speed, features, or specific model capabilities.

All Providers Work with Both Engines

Whether you chose Pydantic AI or LangChain as your engine, all 7 providers are fully supported with identical configuration.

Quick comparison:

Provider API Key Required Speed Best For
PUBLIC No Basic Instant testing, no setup
Ollama No (local) Varies Privacy, offline, no API costs
Google Gemini Yes (free tier) Good Development, prototyping
Groq Yes (free tier) Very fast Production, low cost
OpenAI Yes Good Production, familiar API
Anthropic Yes Good Claude models
Mistral Yes Good Open models
Cohere Yes (free tier) Good Command models

Recommendation

Start: PUBLIC (no API key, instant) → Develop: Google Gemini (free tier) → Production: Groq (very fast, low cost)

Configuration

All providers are configured through environment variables in your .env file:

# Core AI Service Settings
AI_ENABLED=true                    # Enable/disable service
AI_PROVIDER=public                 # Provider: openai, anthropic, google, groq, mistral, cohere, ollama, public
AI_MODEL=auto                      # Model name (varies by provider, "auto" for PUBLIC)
AI_TEMPERATURE=0.7                 # Response creativity (0.0-2.0)
AI_MAX_TOKENS=1000                 # Maximum response length
AI_TIMEOUT_SECONDS=30.0            # Request timeout

# Provider API Keys (only needed for non-PUBLIC providers)
OPENAI_API_KEY=sk-...              # OpenAI API key
ANTHROPIC_API_KEY=sk-ant-...       # Anthropic API key
GOOGLE_API_KEY=...                 # Google API key
GROQ_API_KEY=gsk_...               # Groq API key
MISTRAL_API_KEY=...                # Mistral API key
COHERE_API_KEY=...                 # Cohere API key

Provider Setup Guides

No setup required! Works out of the box with zero configuration.

Setup:

# Already configured by default - just start chatting
my-app ai chat "Hello! Can you help me?"

Best for: Instant testing, demos, getting started without any setup

Blazing fast inference with very generous free tier.

Setup:

  1. Sign up at console.groq.com
  2. Create an API key (free tier available)
  3. Configure your environment:
export AI_PROVIDER=groq
export GROQ_API_KEY=gsk_your_key_here
export AI_MODEL=llama-3.1-8b-instant  # Fastest model

Available Models: See Groq's model documentation for the latest available models.

Best for: Production use (very fast, extremely low cost)

Generous free tier with daily rate limits. Great for development.

Setup:

  1. Get free API key from aistudio.google.com
  2. Configure your environment:
export AI_PROVIDER=google
export GOOGLE_API_KEY=your_key_here
export AI_MODEL=gemini-2.0-flash-exp

Available Models: See Google's model documentation for the latest available models.

Best for: Development and prototyping within rate limits

Industry-standard GPT models. Most widely used and documented.

Setup:

  1. Get API key from platform.openai.com
  2. Add payment method (required for API access)
  3. Configure your environment:
export AI_PROVIDER=openai
export OPENAI_API_KEY=sk-your_key_here
export AI_MODEL=gpt-3.5-turbo

Available Models: See OpenAI's model documentation for the latest available models.

Best for: Production with familiar API, extensive ecosystem

Claude models from Anthropic. Known for safety and reasoning.

Setup:

  1. Get API key from console.anthropic.com
  2. Add payment method
  3. Configure your environment:
export AI_PROVIDER=anthropic
export ANTHROPIC_API_KEY=sk-ant-your_key_here
export AI_MODEL=claude-3-5-sonnet-20241022

Available Models: See Anthropic's model documentation for the latest available models.

Best for: High-quality responses, safety-critical applications

Open-source Mistral models with European data residency.

Setup:

  1. Get API key from console.mistral.ai
  2. Configure your environment:
export AI_PROVIDER=mistral
export MISTRAL_API_KEY=your_key_here
export AI_MODEL=mistral-small-latest

Available Models: See Mistral's model documentation for the latest available models.

Best for: Open-source preference, European compliance requirements

Command models optimized for enterprise and RAG use cases.

Setup:

  1. Get API key from dashboard.cohere.com
  2. Configure your environment:
export AI_PROVIDER=cohere
export COHERE_API_KEY=your_key_here
export AI_MODEL=command-r-plus

Available Models: See Cohere's model documentation for the latest available models.

Best for: Enterprise features, RAG applications

Ollama (Local)

Run models locally with zero API costs. Supports hundreds of open-source models.

Setup

  1. Install Ollama from ollama.ai
  2. Pull a model:
ollama pull llama3.1
  1. Configure your environment:
AI_PROVIDER=ollama
AI_MODEL=llama3.1

Ollama Deployment Modes

Configure at project generation:

# Host mode: Ollama runs on your machine (default)
aegis init my-app --services ai

# Docker mode: Ollama runs in a Docker container
# (select during aegis init interactive prompts)
Mode Description Use Case
host Ollama on localhost:11434 Development, GPU on host
docker Ollama in Docker container Portable, CI/CD
none No Ollama support Cloud-only providers

Switching Models via CLI

With the LLM Catalog, you can discover and switch Ollama models:

# Sync Ollama models into catalog
my-app llm sync --source ollama

# List available Ollama models
my-app llm list --vendor ollama

# Switch to an Ollama model
my-app llm use llama3.1

# Or use /model in interactive chat
my-app ai chat
> /model llama3.1

Cost Tracking with Ollama

Ollama models show $0.00 cost in usage stats - this is expected and normal since all processing happens locally. Illiana is aware of this and won't flag it as an issue.

Switching Providers

You can switch providers at any time by changing your .env file or using the CLI:

# Switch from PUBLIC to Groq
AI_PROVIDER=groq
GROQ_API_KEY=your-key-here
AI_MODEL=llama-3.1-8b-instant

Via .env

Edit .env and restart:

make stop && make serve

Via CLI (with LLM Catalog)

Update .env from the catalog (takes effect on next CLI invocation or server restart):

# Auto-detects provider from model name and updates .env
my-app llm use gpt-4o              # → sets AI_PROVIDER=openai
my-app llm use claude-sonnet-4-20250514  # → sets AI_PROVIDER=anthropic
my-app llm use llama3.1            # → sets AI_PROVIDER=ollama

Switch instantly within an interactive chat session (refreshes in-process, no restart needed):

my-app ai chat
> /model gpt-4o
✓ Switched to OpenAI/gpt-4o

Verify the switch:

my-app ai status
my-app llm current

Troubleshooting

"Missing API key" Error

❌ Missing API key for groq provider. Set GROQ_API_KEY environment variable.

Solution: Add the API key to your .env file:

GROQ_API_KEY=your-actual-key-here

"Provider not available" Error

Check configuration:

my-app ai status

Verify provider is installed:

my-app ai providers

Next Steps:

Was this page helpful?