Getting Started

This guide covers the installation and core lifecycle of the CiteKit SDK.

Prerequisites

1. FFmpeg (Required for Video/Audio): CiteKit uses local FFmpeg to extract high-fidelity clips. Ensure it is installed and available in your system PATH.

   • Windows: Download build and add the bin folder to your Path.
   • macOS: brew install ffmpeg
   • Linux: sudo apt install ffmpeg

   TIP

   To verify, run ffmpeg -version in your terminal. If you see version info, you are ready.

2. Mapper API Key (Gemini by default): Required only for the Ingestion (Mapping) phase. Resolution is 100% local. If you prefer local models, use a custom mapper instead (see Custom Mappers).

   • Get a key from Google AI Studio if you use the default Gemini mapper.
   • Set it as an environment variable: GEMINI_API_KEY (default mapper only).

Installation

Python

pip install citekit

Node.js

npm install citekit

IMPORTANT

JavaScript Peer Dependencies: CiteKit keeps its core lightweight by making modality-specific tools optional. Install based on your needs:

• Documents: npm install pdf-lib
• Video/Audio: npm install fluent-ffmpeg (also requires FFmpeg system install)
• Images: npm install sharp

See Requirements for details.

---

The 3-Step Lifecycle

CiteKit follows a simple Ingest → Map → Resolve flow.

1. Ingest (CLI)

The fastest way to get started is using the CLI to map your documents. The API key must be in your environment (default Gemini mapper) unless you supply a custom mapper in code.

# Python
python -m citekit.cli ingest lecture.mp4 --type video

# JavaScript
npx citekit ingest lecture.mp4 --type video

This generates a JSON map in the .resource_maps/ directory.

2. Connect (SDK)

Load the map into your application logic.

Python

from citekit import CiteKitClient

client = CiteKitClient() # Automatically reads GEMINI_API_KEY from env (Gemini default)
resource_map = client.get_map("lecture")
print(f"Nodes found: {len(resource_map.nodes)}")

Node.js

import { CiteKitClient } from 'citekit';

const client = new CiteKitClient();
const resourceMap = client.getMap('lecture');

3. Resolve (Extract)

Extract a specific node (a scene, a page, or a function) to disk.

# Returns an object with 'output_path' to the clipped file
evidence = client.resolve("lecture", "intro_scene")
print(f"Evidence saved to: {evidence.output_path}")

---

Next Steps

• Deep Dive: Learn how CiteKit handles Coordinate Systems with industrial precision.
• CLI Reference: View every command in the CLI docs.
• MCP Integration: Connect CiteKit directly to Claude or Cursor in the MCP guide.

IMPORTANT

Environment Context: By default, CiteKit creates .resource_maps/ and .citekit_output/ in your current working directory. You can change these by passing storage_dir and output_dir to the constructor.