Knowledge Engine

Your AI knows
what you tell it.

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.

  • Ingest PDFs, databases, APIs, and custom data
  • Semantic and hybrid retrieval in one call
  • Every response includes a citation and source score
PDFs & Docs
Databases
Live APIs & URLs
Custom Data
Ingest
Knowledge
Engine
Serve
Retrieval API
Query API
Context Bundle

Live context pipeline

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

Three steps from data to grounded AI.

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.

Step 02

Index and embed

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

Retrieve on every query

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.

Knowledge Engine pipeline
  1. Source ingestionPDF, DB, API, custom
  2. Parsing + chunkingStructure-aware splitting
  3. Embedding generationDense vector + sparse index
  4. Knowledge graphEntity relationships indexed
  5. Hybrid retrievalDense + sparse, fused + re-ranked
  6. Context bundleChunks + citations + scores

Capabilities

Built for production retrieval.

Semantic search

Dense vector similarity search surfaces the most contextually relevant chunks, even when query and document use different phrasing.

Hybrid retrieval

Combines dense semantic vectors with sparse BM25 keyword matching. Reciprocal Rank Fusion merges both result sets for best-of-both recall.

Multi-source ingestion

Ingest from PDFs, Word docs, Markdown, relational databases, REST APIs, and custom webhook feeds. Sources stay in sync automatically.

Real-time updates

Document updates are re-indexed incrementally. No full re-index required. Latency from ingest to queryable is under 60 seconds for most sources.

Citation support

Every retrieved chunk includes its source document, section, and a relevance score. Your LLM can cite exactly which document it drew from.

Namespace isolation

Partition your knowledge base into namespaces. Filter retrieval by namespace, tag, or metadata field. Teams and tenants stay isolated.

API

One call.
Grounded context returned.

Call the retrieval API with a natural-language query and get back ranked, cited context chunks ready to pass directly to any LLM.

  • JavaScript + Python SDKs
  • REST API with full OpenAPI spec
  • Streaming responses supported
  • P99 retrieval latency under 120ms
Live
import { Lynk } from "docklee";
const lynk = new Lynk({ engineId: "eng_policies" });
// Retrieve grounded context for a user query
const chunks = await lynk.retrieve(
"What is our refund policy for enterprise accounts?",
{ topK: 5 },
);
// chunks[] → pass directly to your LLM
console.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

What teams build with it.

The Knowledge Engine is the retrieval backbone for every context-dependent AI product.

Grounded chatbots

Ship support bots, onboarding assistants, and FAQ agents that answer from your actual docs. Grounded in your data, with citations.

Enterprise search

Give employees a natural-language search layer over every internal document, wiki, and database. Ranked results, not keyword lists.

Autonomous agents

Ground each action your agent takes in verified knowledge. Agents that retrieve before they act are agents you can trust in production.

AI copilots

In-product assistants that pull from your help docs, runbooks, or product data on every suggestion. Context-aware, not context-blind.

Knowledge Engine

The fastest way to make
your AI actually know things.

Ingest your first knowledge source in under 5 minutes. No infrastructure to manage. No model to fine-tune.