adaptable-agents 0.1.3

Python client for Adaptable Agents API - wrap your LLM clients with automatic context generation

Adaptable Agents Python Client

A Python client library that wraps your LLM clients (OpenAI, Anthropic, etc.) with automatic context generation from the Adaptable Agents API. This package seamlessly integrates with your existing LLM workflows to enhance responses with relevant past experiences through intelligent memory strategies.

Features

• Automatic Context Integration: Automatically fetches relevant past experiences and appends them to your prompts
• Memory Storage: Optionally stores input/output pairs as memories for future learning
• Multiple Learning Strategies:
  • DC (Dynamic Cheatsheet): Intelligent extraction of relevant past experiences as contextual cheatsheets
  • AMEM (Agentic Memory): Evolving memory system with semantic search and memory evolution
• OpenAI Support: Drop-in replacement for OpenAI client with context integration
• Anthropic Support: Drop-in replacement for Anthropic client with context integration
• Easy Configuration: Simple API key-based setup
• Memory Scope Control: Organize memories hierarchically with custom scope paths

Installation

pip install adaptable-agents

Or install from source:

git clone https://github.com/your-org/adaptable-agents-python-package.git
cd adaptable-agents-python-package
pip install -e .

Development Setup with uv

This project uses uv for fast virtual environment management. To set up a development environment:

# Install uv (if not already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Create and activate virtual environment
uv venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install the package in editable mode with dev dependencies
uv pip install -e ".[dev]"

The .python-version file specifies Python 3.11, which uv will automatically use.

Quick Start

OpenAI Integration

from adaptable_agents import AdaptableOpenAIClient

# Initialize the client with your API keys
client = AdaptableOpenAIClient(
    adaptable_api_key="your-adaptable-agents-api-key",
    openai_api_key="your-openai-api-key",
    memory_scope_path="customer-support/billing",  # Optional: organize by scope
    api_base_url="http://localhost:8000",  # Optional: defaults to localhost
    strategy="vd"  # "vd" for DC (Dynamic Cheatsheet) or "kg" for AMEM (Agentic Memory)
)

# Use it just like the regular OpenAI client
response = client.chat.completions.create(
    model="gpt-4",
    messages=[
        {"role": "user", "content": "User asked about billing issues with subscription renewal"}
    ]
)

print(response.choices[0].message.content)

Anthropic Integration

from adaptable_agents import AdaptableAnthropicClient

# Initialize the client
client = AdaptableAnthropicClient(
    adaptable_api_key="your-adaptable-agents-api-key",
    anthropic_api_key="your-anthropic-api-key",
    memory_scope_path="engineering/frontend",
    api_base_url="http://localhost:8000",
    strategy="kg"  # Use AMEM strategy for evolving memories
)

# Use it just like the regular Anthropic client
response = client.messages.create(
    model="claude-3-opus-20240229",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "How do I implement authentication?"}
    ]
)

print(response.content[0].text)

How It Works

1. Context Retrieval: Before making an LLM call, the client automatically fetches relevant context from the Adaptable Agents API based on your input and selected strategy
2. Prompt Enhancement: The context is prepended to your user prompt to provide information from past successful interactions
3. LLM Generation: The enhanced prompt is sent to your LLM provider (OpenAI, Anthropic, etc.)
4. Memory Storage (optional): After generation, the input/output pair is stored as a memory for future learning

Learning Strategies

DC (Dynamic Cheatsheet) - Strategy "vd"

The DC strategy extracts relevant past experiences and presents them as contextual cheatsheets. It's ideal for:

• Quick reference to successful patterns
• Static knowledge extraction
• Fast context retrieval with caching support

client = AdaptableOpenAIClient(
    adaptable_api_key="your-api-key",
    openai_api_key="your-openai-key",
    strategy="vd",  # DC strategy
    memory_scope_path="customer-support/billing"
)

AMEM (Agentic Memory) - Strategy "kg"

The AMEM strategy creates evolving, semantic memories that adapt based on new experiences. It's ideal for:

• Evolving knowledge bases
• Complex semantic understanding
• Long-term memory evolution

client = AdaptableOpenAIClient(
    adaptable_api_key="your-api-key",
    openai_api_key="your-openai-key",
    strategy="kg",  # AMEM strategy
    memory_scope_path="engineering/frontend"
)

Advanced Usage

Custom Context Configuration

from adaptable_agents import AdaptableOpenAIClient, ContextConfig

config = ContextConfig(
    similarity_threshold=0.85,  # Higher threshold = more similar results
    max_items=10  # Maximum number of past experiences to include
)

client = AdaptableOpenAIClient(
    adaptable_api_key="your-api-key",
    openai_api_key="your-openai-key",
    context_config=config,
    memory_scope_path="my-scope",
    strategy="vd"
)

Disable Automatic Memory Storage

client = AdaptableOpenAIClient(
    adaptable_api_key="your-api-key",
    openai_api_key="your-openai-key",
    auto_store_memories=False,  # Don't automatically store memories
    strategy="vd"
)

Manual Memory Storage

from adaptable_agents import AdaptableAgent

agent = AdaptableAgent(
    api_key="your-api-key",
    memory_scope_path="customer-support/billing",
    strategy="vd"
)

# Store a memory manually
agent.store_memory(
    input_text="User asked about refund policy",
    output_text="Explained 30-day refund policy and processed refund",
    summarize_input=False  # Optional: whether to summarize input
)

Using Existing Client Instances

from openai import OpenAI
from adaptable_agents import AdaptableOpenAIClient

# Use your existing OpenAI client
existing_client = OpenAI(api_key="your-key")

# Wrap it with adaptable agents
client = AdaptableOpenAIClient(
    adaptable_api_key="your-adaptable-key",
    openai_client=existing_client,
    memory_scope_path="my-scope",
    strategy="vd"
)

Enable/Disable Adaptable Agents

You can toggle the adaptable agents functionality on or off:

from adaptable_agents import AdaptableAnthropicClient

client = AdaptableAnthropicClient(
    adaptable_api_key="your-api-key",
    anthropic_api_key="your-anthropic-key",
    enable_adaptable_agents=True  # Set to False to bypass context fetching
)

# Temporarily disable
client.enable_adaptable_agents = False
response = client.messages.create(...)  # Direct call, no context

# Re-enable
client.enable_adaptable_agents = True
response = client.messages.create(...)  # With context

API Reference

AdaptableOpenAIClient

Wrapper for OpenAI client with context integration.

Parameters:

• adaptable_api_key (str): API key for Adaptable Agents API
• openai_api_key (str, optional): OpenAI API key (if not using existing client)
• api_base_url (str): Base URL of Adaptable Agents API (default: "http://localhost:8000")
• memory_scope_path (str): Memory scope path for organizing memories (default: "default")
• context_config (ContextConfig, optional): Configuration for context generation
• auto_store_memories (bool): Whether to automatically store memories (default: True)
• openai_client (OpenAI, optional): Pre-initialized OpenAI client
• summarize_input (bool, optional): Whether to summarize inputs before storage
• strategy (StrategyType): Strategy to use - "vd" for DC or "kg" for AMEM (default: "vd")

AdaptableAnthropicClient

Wrapper for Anthropic client with context integration.

Parameters:

• adaptable_api_key (str): API key for Adaptable Agents API
• anthropic_api_key (str, optional): Anthropic API key (if not using existing client)
• api_base_url (str): Base URL of Adaptable Agents API (default: "http://localhost:8000")
• memory_scope_path (str): Memory scope path for organizing memories (default: "default")
• context_config (ContextConfig, optional): Configuration for context generation
• auto_store_memories (bool): Whether to automatically store memories (default: True)
• anthropic_client (Anthropic, optional): Pre-initialized Anthropic client
• summarize_input (bool, optional): Whether to summarize inputs before storage
• enable_adaptable_agents (bool): Enable/disable adaptable agents functionality (default: True)
• strategy (StrategyType): Strategy to use - "vd" for DC or "kg" for AMEM (default: "vd")

AdaptableAgent

Core client for interacting with Adaptable Agents API.

Parameters:

• api_key (str): API key for Adaptable Agents API
• api_base_url (str): Base URL of Adaptable Agents API (default: "http://localhost:8000")
• memory_scope_path (str): Memory scope path (default: "default")
• context_config (ContextConfig, optional): Context configuration
• strategy (StrategyType): Strategy to use - "vd" for DC or "kg" for AMEM (default: "vd")

Methods:

• get_context(input_text: str, use_cache: bool = False) -> Optional[str]: Fetch context for the given input. use_cache only works for DC strategy ("vd").
• load_prior_knowledge(input_text: Optional[str] = None) -> Optional[str]: Convenience method to load context without storing memories
• store_memory(input_text: str, output_text: str, summarize_input: Optional[bool] = None) -> Optional[Dict]: Store a memory
• format_prompt_with_context(user_prompt: str, context: Optional[str]) -> str: Format prompt with context

ContextConfig

Configuration for context generation.

Parameters:

• similarity_threshold (float): Minimum similarity score for inclusion (default: 0.8)
• max_items (int): Maximum number of past experiences to include (default: 5)

Memory Scopes

Memory scopes allow you to organize memories hierarchically. Use different scopes for different contexts:

• customer-support/billing - Billing-related support memories
• engineering/frontend - Frontend engineering memories
• sales/enterprise - Enterprise sales memories
• custom/scope/path - Any custom hierarchical structure

Different memory scope paths ensure complete isolation between contexts at the database level, allowing you to organize memories by department, project, feature, or any custom hierarchy.

Requirements

• Python >= 3.8
• requests >= 2.31.0
• openai >= 1.0.0 (for OpenAI support)
• anthropic >= 0.18.0 (for Anthropic support)

Examples

See the examples/ directory for complete usage examples:

• examples/openai_example.py - OpenAI client usage
• examples/anthropic_example.py - Anthropic client usage

Testing

Run tests with pytest:

pytest

With coverage:

pytest --cov=adaptable_agents --cov-report=html

License

MIT License

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Support

For issues and questions, please open an issue on GitHub or contact [e-mail].

Related Projects

• Adaptable Agents API - The backend API service that powers this client