Provider Capabilities¶

AMP supports multiple AI providers for content generation.

LLM Providers¶

Comparison¶

Claude (Anthropic)¶

Models: - claude-3-opus — Highest quality - claude-3-sonnet — Balanced - claude-3-haiku — Fastest

Strengths: - Excellent writing quality - Strong brand voice consistency - Nuanced content generation - Long context handling

Best for: High-quality content, thought leadership

GPT-4 (OpenAI)¶

Models: - gpt-4-turbo — Latest, fastest - gpt-4 — Stable - gpt-3.5-turbo — Budget option

Strengths: - Versatile capabilities - Good at various formats - Strong reasoning

Best for: General content, diverse formats

Gemini (Google)¶

Models: - gemini-pro — Standard - gemini-pro-vision — With images

Strengths: - Large context window - Good multimodal capabilities - Cost-effective

Best for: Long-form content, visual analysis

Image Providers¶

Comparison¶

DALL-E 3 (OpenAI)¶

Sizes: - 1024x1024 - 1024x1792 - 1792x1024

Strengths: - High quality - Good text rendering - Natural compositions

Stability AI¶

Models: - SDXL 1.0 - SD 3.0

Strengths: - Style control - Faster generation - More customizable

Replicate¶

Models: - Various community models

Strengths: - Model variety - Custom fine-tunes - Cost-effective

Provider Selection¶

Automatic Selection¶

AMP automatically selects providers based on:

1. Tenant preference — Configured default
2. Capability match — Required features
3. Availability — Health status
4. Cost — Budget constraints

Manual Override¶

{
  "constraints": {
    "preferred_llm": "claude",
    "preferred_image": "openai"
  }
}

Fallback Chain¶

When primary fails:

Claude → OpenAI → Gemini → Error

Cost Comparison¶

LLM Costs (per 1K tokens)¶

Image Costs (per image)¶

Rate Limits¶

Self-Hosted Options¶

Ollama¶

OLLAMA_URL=http://localhost:11434
OLLAMA_MODEL=llama2

vLLM¶

VLLM_URL=http://localhost:8000
VLLM_MODEL=meta-llama/Llama-2-70b

Trade-offs: - Lower cost - Data privacy - Higher latency - Quality varies