Step 01
Ingest your knowledge
Point the Knowledge Engine at your documents, databases, live APIs, or custom data feeds. Any format. Any source. Ingestion is asynchronous and idempotent.
Knowledge Engine
The Knowledge Engine ingests, indexes, and retrieves your exact knowledge on every query. No fine-tuning. No retraining. Just accurate, cited retrieval from your data.
LLMs don't know your business, your policies, or your customers. The Knowledge Engine bridges the gap between your data and your model, without any training or fine-tuning.
Sources supported
Any format
PDFs, DBs, APIs, custom feeds
Retrieval mode
Hybrid
Sparse + dense in a single call
Context output
Cited
Every chunk includes a source reference
How it works
Step 01
Point the Knowledge Engine at your documents, databases, live APIs, or custom data feeds. Any format. Any source. Ingestion is asynchronous and idempotent.
Step 02
KE parses, chunks, and generates dense vector embeddings alongside sparse keyword indexes. The result is a queryable knowledge graph that reflects your exact data, not generic training data.
Step 03
On each request, KE performs hybrid retrieval: dense semantic search plus sparse BM25, fused and re-ranked by context relevance. The response includes ranked chunks with source citations.
Capabilities
Dense vector similarity search surfaces the most contextually relevant chunks, even when query and document use different phrasing.
Combines dense semantic vectors with sparse BM25 keyword matching. Reciprocal Rank Fusion merges both result sets for best-of-both recall.
Ingest from PDFs, Word docs, Markdown, relational databases, REST APIs, and custom webhook feeds. Sources stay in sync automatically.
Document updates are re-indexed incrementally. No full re-index required. Latency from ingest to queryable is under 60 seconds for most sources.
Every retrieved chunk includes its source document, section, and a relevance score. Your LLM can cite exactly which document it drew from.
Partition your knowledge base into namespaces. Filter retrieval by namespace, tag, or metadata field. Teams and tenants stay isolated.
API
Call the retrieval API with a natural-language query and get back ranked, cited context chunks ready to pass directly to any LLM.
import { Lynk } from "docklee";const lynk = new Lynk({ engineId: "eng_policies" });// Retrieve grounded context for a user queryconst chunks = await lynk.retrieve("What is our refund policy for enterprise accounts?",{ topK: 5 },);// chunks[] → pass directly to your LLMconsole.log(chunks[0].content);// "Enterprise accounts receive a full refund within 60 days..."console.log(chunks[0].section);// "Section 3.2 · Enterprise Refunds"
Use cases
The Knowledge Engine is the retrieval backbone for every context-dependent AI product.
Ship support bots, onboarding assistants, and FAQ agents that answer from your actual docs. Grounded in your data, with citations.
Give employees a natural-language search layer over every internal document, wiki, and database. Ranked results, not keyword lists.
Ground each action your agent takes in verified knowledge. Agents that retrieve before they act are agents you can trust in production.
In-product assistants that pull from your help docs, runbooks, or product data on every suggestion. Context-aware, not context-blind.
Knowledge Engine
Ingest your first knowledge source in under 5 minutes. No infrastructure to manage. No model to fine-tune.