Quick Start

Installation

Get started by installing the Deeptrain SDK via pip. Deeptrain requires Python 3.8 or higher.

pip install deeptrain

Quick Start: Connect Your First Data Source

In this 5-minute guide, we will initialize the Deeptrain client and process a video URL to extract searchable data for your AI agent.

1. Initialize the Client

First, import Deeptrain and configure your environment. Deeptrain is model-agnostic, so you can specify your preferred LLM provider.

from deeptrain import Deeptrain

# Initialize the connector
dt = Deeptrain(api_key="YOUR_DEEPTRAIN_API_KEY")

2. Connect a Multi-modal Source

Deeptrain handles the heavy lifting of converting non-textual data into a format your LLM can understand. Let's process a video from YouTube using the Transcribe API.

# Process a video for your AI agent's knowledge base
video_data = dt.video.transcribe(
    source="https://www.youtube.com/watch?v=example",
    mode="high_precision"
)

print(f"Transcription complete: {video_data.text[:100]}...")

3. Query via Localized Embeddings

Once the data is processed, Deeptrain stores it in a localized embedding database. You can now query this data directly without worrying about the LLM's context window limits.

# Retrieve relevant context based on a natural language query
context = dt.embeddings.query(
    query="What are the key takeaways from the video?",
    top_k=3
)

# Pass this context directly to your LLM
response = dt.chat.complete(
    model="gpt-4", 
    prompt=f"Based on this context: {context}, answer the user query."
)

Supported Data Types

You can use the same pattern to integrate various data formats:

| Module | Description | Example Usage | | :--- | :--- | :--- | | dt.text | Manage live data streams and embeddings. | dt.text.sync(url="https://docs.example.com") | | dt.vision | Process flowcharts, graphs, and images. | dt.vision.analyze(image_path="./flowchart.png") | | dt.audio | Convert audio files into actionable training data. | dt.audio.process(file_path="./interview.mp3") | | dt.video | Multi-dimensional processing for Vimeo/YouTube. | dt.video.transcribe(source="vimeo_url") |

Next Steps

• Explore Models: Check the Model Support Guide to see how to configure any of the 200+ supported private and open-source models.
• Advanced Vision: Learn how to turn non-vision models into vision-enabled agents in the Computer Vision Tutorial.
• API Reference: View the full API Documentation for detailed parameter definitions.