endee-crewai 0.1.0

CrewAI tools with Endee vector database support

EndeeCrewAI Integration

High-Performance Vector Database Integration for CrewAI Agents

EndeeCrewAI provides a seamless integration between Endee — a high-performance vector database — and CrewAI. It allows CrewAI agents to store, retrieve, and manage vector-based knowledge efficiently with metadata support, enabling advanced search and contextual reasoning.

---

Features

• Vector-based memory for CrewAI: Use Endee as a backend for short-term and entity memory.
• High-performance search: Approximate Nearest Neighbor (ANN) searches for fast retrieval.
• Metadata & filtering support: Store rich metadata and filter queries.
• Embeddings integration: Supports any embedding provider (e.g., Google Gemini, OpenAI).

---

Installation

pip install endee-crewai

Ensure you also have the required dependencies for CrewAI and your chosen embedding provider:

pip install crewai crewai-tools google-genai

---

Environment Variables

Create a .env file to store API credentials:

ENDEE_API_TOKEN=your_endee_api_token
GOOGLE_API_KEY=your_google_api_key
OPENAI_API_KEY=your_openai_api_key
COHERE_API_KEY=your_cohere_api_key

Note: You can use any embedding provider (OpenAI, Google, Cohere, HuggingFace). Supply the API key for the provider you use and omit the rest.

However, ENDEE_API_TOKEN is required, as it is needed to access your Endee vector database.

---

Quick Start

Initialize Endee Vector Store

from endee_crewai import EndeeVectorStore

# Embedding function (e.g., using OpenAI or Google or Cohere)
embedder_config = {
    "provider": "cohere",
    "config": {"model_name": "small", "api_key": "<COHERE_API_KEY>"}
}

# Create Endee store
memory_store = EndeeVectorStore(
    type="my_index",
    api_token="<ENDEE_API_TOKEN>",
    embedder_config=embedder_config,
    space_type="cosine",
    crew=None,
)

# Reset index if needed
memory_store.reset()
time.sleep(2) # Wait for reset

---

Save Documents

documents = [
    ("Python is dynamically typed.", {"creator": "Guido van Rossum", "typing": "dynamic"}),
    ("Go is statically typed.", {"creator": "Robert Griesemer", "typing": "static"})
]

for text, meta in documents:
    memory_store.save(text, meta)

---

Query Vector Store

results = memory_store.search(
    query="Python typing discipline",
    limit=3,
    filter={"typing": {"$eq": "dynamic"}}
)

for r in results:
    print(f"ID: {r['id']}, Score: {r['score']}, Text: {r['context']}")

---

CrewAI Integration

from crewai.memory import ShortTermMemory, EntityMemory
from crewai import Agent, Crew, Task, Process, LLM

# Create CrewAI memory objects
short_memory = ShortTermMemory(storage=memory_store)
entity_memory = EntityMemory(storage=memory_store)

# Define LLM (Use any valid model)
llm = LLM(model="gemini-2.5-flash-lite", api_key="<GOOGLE_API_KEY>")

# Define an agent
agent = Agent(
    role="Programming Language Expert",
    goal="Answer questions using stored programming language knowledge.",
    backstory="Consult documents and provide concise answers.",
    llm=llm,
    memory=short_memory,
    verbose=True
)

# Define a task
task = Task(
    description="Answer questions about programming languages.",
    agent=agent,
    expected_output="Concise and accurate answers."
)

# Run Crew
crew = Crew(
    agents=[agent],
    tasks=[task],
    process=Process.sequential,
    memory=True,
    short_term_memory=short_memory,
    entity_memory=entity_memory,
    verbose=True
)

result = crew.kickoff()
print(result)

---

Helper Functions

You can define custom helpers for:

• Checking vector entries in Endee
• Testing knowledge retrieval
• Validating entity memory

Example:

def check_language_vectors(memory_store):
    queries = ["Python language features", "Go concurrency"]
    for q in queries:
        results = memory_store.search(query=q, limit=3)
        for r in results:
            print(r["context"])

---

Configuration Options

The EndeeVectorStore constructor accepts the following parameters:

• type: Name of the Endee index.
• api_token: Your Endee API token (required if endee_index is not provided).
• embedder_config: Configuration for the embedding provider (OpenAI, Google, Cohere, HuggingFace).
• space_type: Distance metric for vector search. One of "cosine", "l2", or "ip" (default: "cosine").
• allow_reset: Whether to allow resetting (deleting) the index (default: True).
• crew: Optional CrewAI object for integration.
• text_key: Key to store text in metadata (default: "value").
• endee_index: Optional existing Endee index object.

---