CiteKitMultimodal Context SDK

Quick Install ​

pip install citekit

npm install citekit

# Installs CLI tool for Claude/Cline
pip install citekit

Quick Usage ​

CiteKit works the same way across all interfaces: Ingest to map, Resolve to extract.

from citekit import CiteKitClient

# 1. Ingest (Map)
client = CiteKitClient()
video_map = await client.ingest("lecture.mp4", "video")

# 2. Resolve (Extract)
# Extracts exact clip from 10s to 20s
evidence = client.resolve(video_map.resource_id, "intro_scene")
print(evidence.output_path)

import { CiteKitClient } from 'citekit';

// 1. Ingest (Map)
const client = new CiteKitClient();
const map = await client.ingest('lecture.mp4', 'video');

// 2. Resolve (Extract)
const evidence = await client.resolve(map.resource_id, 'intro_scene');
console.log(evidence.output_path);

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

# 2. Resolve
python -m citekit.cli resolve lecture intro_scene

{
  "mcpServers": {
    "citekit": {
      "command": "citekit",
      "args": ["serve"],
      "env": { "GEMINI_API_KEY": "..." }
    }
  }
}

Overview ​

CiteKit is the high-fidelity orchestration layer for Modern AI Architectures. It solves the Context Precision Problem in multimodal applications (Agentic RAG, Long-Context reasoning, and Multi-Agent flows).

Instead of "fuzzy" context loading, CiteKit employs a Semantic Indexing strategy:

1. Map: Files are analyzed to discover their "Structural DNA" (Topics, Scenes, Charts).
2. Orchestrate: Agents use the Map to navigate and choose relevant logical units.
3. Resolve: CiteKit extracts exactly those high-fidelity segments (Physical or Virtual).