JavaScript/TypeScript Client API Reference

The CiteKitClient is the primary interface for CiteKit in Node.js and TypeScript environments. It provides async/await-based ingestion, map management, and resolution.

Constructor

import { CiteKitClient } from 'citekit';
import { OllamaMapper } from './ollama-mapper';

// Using default Gemini mapper
const client = new CiteKitClient({
    apiKey: "YOUR_GEMINI_API_KEY"
});

// Using custom mapper
const client = new CiteKitClient({
    mapper: new OllamaMapper("llama3")
});

// Full options
const client = new CiteKitClient({
    apiKey?: "YOUR_GEMINI_API_KEY",
    baseDir?: ".",
    storageDir?: ".resource_maps",
    outputDir?: ".citekit_output",
    model?: "gemini-2.0-flash",
    maxRetries?: 3,
    concurrencyLimit?: 5,
    mapper?: undefined
});

Constructor Options

interface CiteKitClientOptions {
    /** Gemini API Key. Falls back to GEMINI_API_KEY environment variable. */
    apiKey?: string;

    /** Gemini Model ID (default: "gemini-2.0-flash") */
    model?: string;

    /** Max retries for Gemini API calls (default: 3) */
    maxRetries?: number;

    /** Base directory for all CiteKit operations. Useful for serverless (e.g., /tmp). */
    baseDir?: string;

    /** Directory where resource maps are stored (relative to baseDir). Default: ".resource_maps" */
    storageDir?: string;

    /** Directory for resolved output files (relative to baseDir). Default: ".citekit_output" */
    outputDir?: string;

    /** Max concurrent ingestion calls (default: 5). Prevents rate-limiting. */
    concurrencyLimit?: number;

    /** Custom mapper implementation. If provided, overrides Gemini mapper. */
    mapper?: MapperProvider;
}

Option	Type	Default	Description
apiKey	string | undefined	GEMINI_API_KEY env	Gemini API Key (only used if mapper is not provided).
model	string	"gemini-2.0-flash"	Gemini model ID (only used if mapper is not provided).
maxRetries	number	3	Retry attempts for failed mapper calls (only used if mapper is not provided).
baseDir	string	"."	Root directory for all operations (useful in serverless environments).
storageDir	string	".resource_maps"	Relative path where resource maps are persisted as JSON.
outputDir	string	".citekit_output"	Relative path where resolved clips/extracts are written.
concurrencyLimit	number	5	Maximum number of parallel mapper calls (ingestion).
mapper	MapperProvider | undefined	undefined	Custom mapper instance. If provided, Gemini is not used.

Throws

• Error: If neither mapper nor apiKey (or GEMINI_API_KEY env) is provided when calling ingest().

---

Methods

async ingest(resourcePath, resourceType, options?)

Analyzes a file using the configured mapper and generates a ResourceMap. This is the primary entry point for structuring your resources.

async ingest(
    resourcePath: string,
    resourceType: 'video' | 'audio' | 'document' | 'image' | 'text',
    options?: { resourceId?: string }
): Promise<ResourceMap>

Parameters

Parameter	Type	Description
resourcePath	string	Absolute or relative path to the resource file.
resourceType	'video' | 'audio' | 'document' | 'image' | 'text'	The resource modality.
options?.resourceId	string | undefined	Optional custom ID. Defaults to the filename stem.

Returns

• Promise<ResourceMap>: The generated resource structure containing nodes, metadata, and location data.

Ingestion Workflow

The ingestion process is atomic and idempotent:

1. Path Validation: Checks that the file exists.
2. SHA-256 Hashing: Computes a content hash for deduplication.
3. Cache Lookup: Scans storageDir for an existing map with the same hash (skips LLM call if found).
4. Concurrency Gate: Waits for a semaphore slot (respects concurrencyLimit).
5. Mapper Generation: Calls the configured MapperProvider.generateMap().
6. JSON Extraction: Automatically extracts JSON from the LLM response.
7. Persistence: Saves the map as <resourceId>.json in storageDir.
8. Metadata Injection: Adds source_hash and source_size to the map.

Examples

Basic ingestion:

import { CiteKitClient } from 'citekit';

const client = new CiteKitClient({
    apiKey: process.env.GEMINI_API_KEY
});

const resourceMap = await client.ingest('lecture_01.mp4', 'video');
console.log(`Mapped '${resourceMap.resource_id}' with ${resourceMap.nodes.length} nodes`);

Explicit type and custom ID:

const resourceMap = await client.ingest(
    'src/main.ts',
    'text',
    { resourceId: 'codebase_v2' }
);
console.log(resourceMap.resource_id);  // "codebase_v2"

Using a custom mapper:

import { OllamaMapper } from './ollama-mapper';

const client = new CiteKitClient({
    mapper: new OllamaMapper('llama3')
});

// Ingest with local LLM (no API calls)
const resourceMap = await client.ingest('docs/README.md', 'text');
console.log(`Mapped locally with ${resourceMap.nodes.length} sections`);

Concurrent ingestion:

// Multiple ingests run in parallel (respecting concurrencyLimit)
const maps = await Promise.all([
    client.ingest('video1.mp4', 'video'),
    client.ingest('document1.pdf', 'document'),
    client.ingest('code.ts', 'text')
]);

console.log(`Ingested ${maps.length} resources`);

Throws

• Error: If resourcePath does not exist.
• Error: If no mapper is configured.
• Error: If resourceType is not recognized.

---

async resolve(resourceId, nodeId, options?)

Resolves a node to extracted evidence. Extracts the physical segment from the resource (video clip, PDF pages, image crop, etc.) or returns a metadata-only reference.

async resolve(
    resourceId: string,
    nodeId: string,
    options?: { virtual?: boolean }
): Promise<ResolvedEvidence>

Parameters

Parameter	Type	Description
resourceId	string	The resource ID (from ingest() or listMaps()).
nodeId	string	The node ID to resolve (e.g., "chapter_1.scene_2").
options?.virtual	boolean | undefined	If true, returns metadata without extracting files (no FFmpeg/PDF library calls). Defaults to false.

Returns

• Promise<ResolvedEvidence>: An object containing:
  • output_path (string or undefined): Path to the extracted file (undefined if virtual=true)
  • address (string): CiteKit URI address (e.g., "video://lecture_01#t=10-20")
  • modality (string): The node's modality (e.g., "video", "document")
  • node (Node): The resolved node object
  • resource_id (string): The resource ID

Resolution Workflow

1. Map Lookup: Loads the resource map from storageDir.
2. Node Search: Recursively finds the node by ID in the hierarchical structure.
3. Address Building: Generates a CiteKit URI based on the node's location.
4. Virtual Check: If virtual=true, returns address without extraction.
5. Modality Dispatch: Selects the appropriate resolver (VideoResolver, DocumentResolver, etc.).
6. Physical Extraction: Resolver writes the extracted segment to outputDir (async I/O).

Examples

Virtual resolution (metadata only):

const client = new CiteKitClient({
    apiKey: process.env.GEMINI_API_KEY
});

const evidence = await client.resolve(
    'lecture_01',
    'chapter_1.intro',
    { virtual: true }
);

console.log(evidence.address);     // e.g., "video://lecture_01#t=145-285"
console.log(evidence.output_path); // undefined

Physical resolution (extracts file):

const evidence = await client.resolve('lecture_01', 'chapter_1.intro');

console.log(evidence.output_path);  // e.g., ".citekit_output/lecture_01_chapter_1_intro.mp4"
console.log(evidence.modality);     // "video"

Document page extraction:

const evidence = await client.resolve('textbook', 'chapter_2.definition');
// Output: ".citekit_output/textbook_chapter_2_definition.pdf"
// Contains only pages 12-15 (as specified in the node's location)

Resolve multiple nodes in parallel:

const nodeIds = ['chapter_1.intro', 'chapter_1.body', 'chapter_1.conclusion'];

const allEvidence = await Promise.all(
    nodeIds.map(id => client.resolve('lecture_01', id))
);

allEvidence.forEach(ev => {
    console.log(`${ev.node.title}: ${ev.output_path}`);
});

Throws

• Error: If the resource map doesn't exist.
• Error: If the node ID is not found.
• Error: If no resolver is available for the node's modality.

---

getMap(resourceId)

Loads a previously ingested resource map from local storage.

getMap(resourceId: string): ResourceMap

Parameters

Parameter	Type	Description
resourceId	string	The resource ID to retrieve.

Returns

• ResourceMap: The deserialized resource structure.

Example

const client = new CiteKitClient({
    apiKey: process.env.GEMINI_API_KEY
});

const resourceMap = client.getMap('lecture_01');
console.log(`Resource: ${resourceMap.title}`);
console.log(`Nodes: ${resourceMap.nodes.length}`);

Throws

• Error: If no map exists for the given resourceId.

---

listMaps()

Returns all resource IDs (ingested maps) currently stored locally.

listMaps(): string[]

Returns

• string[]: Array of resource IDs.

Example

const client = new CiteKitClient({
    apiKey: process.env.GEMINI_API_KEY
});

const maps = client.listMaps();
console.log(`Available resources: ${maps.join(', ')}`);
// Output: Available resources: lecture_01, textbook, codebase_v2

---

getStructure(resourceId)

Retrieves a resource map as a plain JavaScript object (JSON-serializable). Commonly used by MCP servers and integrations.

getStructure(resourceId: string): ResourceMap

Parameters

Parameter	Type	Description
resourceId	string	The resource ID to retrieve.

Returns

• ResourceMap: The resource map as a plain object (ready for JSON serialization).

Example

const client = new CiteKitClient({
    apiKey: process.env.GEMINI_API_KEY
});

const structure = client.getStructure('lecture_01');
// Can be serialized directly to JSON
const jsonStr = JSON.stringify(structure);
console.log(jsonStr);

Throws

• Error: If no map exists for the given resourceId.

---

Type Definitions

See Core Data Models for unified definitions across all implementations.

Quick Reference (TypeScript):

export interface ResourceMap {
    resource_id: string;    // Unique identifier
    type: "document" | "video" | "audio" | "image" | "text" | "virtual";
    title: string;          // Human-readable title
    source_path: string;    // Absolute path to the source file
    nodes: Node[];          // Hierarchical nodes
    metadata?: Record<string, string | number | null>;  // Custom metadata
    created_at: string;     // ISO 8601 timestamp
}

export interface Node {
    id: string;             // Unique within resource (e.g., "chapter_1.scene_2")
    title?: string;         // Display name
    type: string;           // "section", "scene", "chapter", "class", "function", etc.
    location: Location;     // Temporal/spatial bounds
    summary?: string;       // Brief description
    children?: Node[];      // Nested nodes (optional)
}

export interface Location {
    modality: "document" | "video" | "audio" | "image" | "text" | "virtual";
    start?: number;          // Video/Audio start (seconds)
    end?: number;            // Video/Audio end (seconds)
    pages?: number[];        // Document pages (1-indexed list)
    lines?: [number, number]; // Text lines (1-indexed)
    bbox?: [number, number, number, number];  // Image bbox [x1, y1, x2, y2] 0-1 normalized corners
    virtual_address?: string; // Virtual reference URI
}

export interface ResolvedEvidence {
    output_path?: string;   // Path to extracted file (undefined if virtual)
    modality: string;       // Node's modality
    address: string;        // CiteKit URI (e.g., "video://lecture_01#t=145.5-285.0")
    node: Node;             // The resolved node
    resource_id: string;    // The resource ID
}

All field names use snake_case (e.g., resource_id, not resourceId) for consistency with JSON serialization across all implementations (Python, JavaScript, MCP).

---

Error Handling

Common Errors

Missing mapper or API key:

try {
    const client = new CiteKitClient();  // No mapper, no apiKey
    await client.ingest('file.mp4', 'video');
} catch (error) {
    console.error(error.message);  // "GEMINI_API_KEY or custom 'mapper' required..."
}

Resource not found:

try {
    const map = client.getMap('nonexistent');
} catch (error) {
    console.error(error.message);  // "No map found for resource 'nonexistent'..."
}

Node not found:

try {
    const evidence = await client.resolve('lecture_01', 'invalid.node.id');
} catch (error) {
    console.error(error.message);  // "Node 'invalid.node.id' not found..."
}

---

Complete Example: Multi-Modal RAG Pipeline

import { CiteKitClient } from 'citekit';

async function ragPipeline() {
    // Initialize client
    const client = new CiteKitClient({
        apiKey: process.env.GEMINI_API_KEY
    });

    // 1. Ingest resources
    console.log('Ingesting lecture...');
    const videoMap = await client.ingest('lecture.mp4', 'video', { resourceId: 'lecture_01' });

    console.log('Ingesting textbook...');
    const docMap = await client.ingest('textbook.pdf', 'document', { resourceId: 'textbook' });

    // 2. List all resources
    const allResources = client.listMaps();
    console.log(`Mapped resources: ${allResources.join(', ')}`);

    // 3. Resolve specific nodes
    const nodeIds = ['chapter_1.intro', 'chapter_1.definition'];
    
    for (const nodeId of nodeIds) {
        console.log(`\nResolving ${nodeId}...`);

        // Virtual resolution (metadata only)
        const virtualEvidence = await client.resolve('lecture_01', nodeId, { virtual: true });
        console.log(`  Address: ${virtualEvidence.address}`);

        // Physical extraction
        const physicalEvidence = await client.resolve('lecture_01', nodeId);
        console.log(`  Extracted to: ${physicalEvidence.output_path}`);
    }

    // 4. Export a resource structure (e.g., for MCP)
    const structure = client.getStructure('lecture_01');
    console.log(`\nStructure JSON: ${JSON.stringify(structure, null, 2)}`);
}

ragPipeline().catch(console.error);