Skip to content

README.md

Latest commit

 

History

History
135 lines (106 loc) · 3.4 KB

README.md

File metadata and controls

135 lines (106 loc) · 3.4 KB

Provider Examples

This folder contains examples for each supported LLM provider.

🎯 Quick Start

Choose your provider and run the corresponding example:

OpenAI (GPT models)

export OPENAI_API_KEY='sk-...'
python examples/05_providers/openai_example.py

Models: gpt-5, gpt-4.1, gpt-4.1-mini

Claude (Anthropic)

export ANTHROPIC_API_KEY='sk-ant-...'
python examples/05_providers/claude_example.py

Models: claude-4-5-sonnet, claude-4-5-haiku, claude-4-opus

Google Gemini

export GOOGLE_API_KEY='...'
python examples/05_providers/gemini_example.py

Models: gemini-2.5-flash-lite, gemini-2.5-pro, gemini-2.5-flash

Ollama (Local)

# No API key needed! Run locally
ollama pull qwen3:1.7b
python examples/05_providers/ollama_example.py

Models: llama3.2, qwen3, mistral, phi3, deepseek-r1, and 100+ more

🔑 API Key Setup

Linux/Mac

export OPENAI_API_KEY='your-key-here'
export ANTHROPIC_API_KEY='your-key-here'
export GOOGLE_API_KEY='your-key-here'

Windows (PowerShell)

$env:OPENAI_API_KEY='your-key-here'
$env:ANTHROPIC_API_KEY='your-key-here'
$env:GOOGLE_API_KEY='your-key-here'

Windows (CMD)

set OPENAI_API_KEY=your-key-here
set ANTHROPIC_API_KEY=your-key-here
set GOOGLE_API_KEY=your-key-here

📊 Provider Comparison

Provider Best For Cost Speed Quality
OpenAI General purpose, production apps $$$ Fast Excellent
Claude Long context, detailed analysis $$$ Medium Excellent
Gemini Multimodal, fast prototyping $$ Very Fast Good
Ollama Privacy, offline, no cost FREE Variable Good

🎨 Common Usage Pattern

All providers share the same API:

from memlayer.wrappers.{provider} import {Provider}

# Initialize with memory
client = Provider(
    model="model-name",
    storage_path="./memories",
    user_id="user_id",
    operation_mode="local"  # or "online" or "lightweight"
)

# Chat with automatic memory
response = client.chat(messages=[
    {"role": "user", "content": "Remember my name is Alice"}
])

# Memory is automatically stored and retrieved!

🚀 Salience Modes

Each provider supports three memory modes:

1. Local Mode (Default)

  • Uses sentence-transformers locally
  • High accuracy semantic filtering
  • ~10s startup time, no API costs
  • Best for: High-volume usage, offline apps

2. Online Mode

  • Uses OpenAI embeddings API
  • Fast startup (~2s), small API cost
  • Requires OPENAI_API_KEY
  • Best for: Production apps, serverless

3. Lightweight Mode

  • Keyword-based filtering only
  • Instant startup, no embeddings
  • Best for: Prototyping, testing

Example:

client = Provider(
    model="...",
    operation_mode="online",  # Fast startup!
    storage_path="./memories"
)

💡 Tips

  1. Mix and match: Use different providers for different tasks
  2. Ollama for dev: Use Ollama locally, then switch to cloud provider for production
  3. Online mode for serverless: Use operation_mode="online" in AWS Lambda/Cloud Functions
  4. Share memories: All providers can read the same memory storage if using same storage_path

📚 More Examples

  • Basics: examples/01_basics/
  • Search Tiers: examples/02_search_tiers/
  • Advanced Features: examples/03_features/
  • Benchmarks: examples/04_benchmarks/