Knowledge Engine

Searchable Knowledge for Every Agent

Upload documents, index them automatically, and give agents instant access through hybrid semantic + keyword search with built-in RAG.

Hybrid Search

Semantic + keyword search fused for best results

Built-in RAG

Automatic context injection into agent prompts

MCP Native

Agents search, retrieve, and browse knowledge natively

Core Features

From Documents to Agent Knowledge

Everything you need to build agents with deep domain knowledge and contextual understanding

Hybrid Search

Combines dense vector similarity with sparse BM25 keyword matching. Results merged via Reciprocal Rank Fusion for the best of both approaches.

Automatic Ingestion

Upload documents through API or dashboard. Content is extracted, chunked intelligently with overlap, embedded, and indexed in Qdrant automatically.

Intelligent Chunking

Semantic chunking with configurable target size (200-400 tokens), 10% overlap, and minimum token thresholds. Preserves context across chunk boundaries.

LLM Reranking

Optional LLM-based reranking fetches 3x candidates then selects the most relevant results. Configurable globally or per knowledge base.

Vector Embeddings

Top-tier embedding models for high-accuracy semantic understanding. Configurable dimensions and batch processing for efficient indexing at scale.

MCP Integration

Three tools exposed natively: search (hybrid retrieval), get_by_id (fetch specific chunk), and list (browse sources). Per-gateway KB scoping.

From Documents to Intelligence

Upload & Extract

Upload files through API or dashboard. Text and markdown are read directly. PDF, DOCX, PPTX, XLSX, and images are automatically converted to searchable content.

Chunk & Embed

Documents are split into semantic chunks with overlap. Each chunk is embedded into dense vectors and sparse BM25 representations, then stored in Qdrant.

Search & Retrieve

Agents query using natural language. Hybrid search combines semantic similarity and keyword matching with reciprocal rank fusion. Optional LLM reranking.

Inject & Generate

Retrieved chunks are automatically injected into agent prompts as context. Agents generate accurate, grounded responses via built-in RAG.

Technical Details

Built for Scale and Accuracy

Supported Formats

  • Text & Markdown — .txt, .md (native UTF-8)
  • PDF — automatic text extraction
  • Microsoft Office — DOCX, PPTX, XLSX
  • Images — PNG, JPG, SVG
  • Other formats — additional formats supported automatically

Search & Retrieval

  • Vector store — Qdrant Cloud (gRPC)
  • Fusion — Reciprocal Rank Fusion (40 prefetch per method)
  • Results — Up to 100 per query, 0.15 score threshold
  • Reranking — Optional LLM reranker (3x overfetch)
  • Scoping — Per-gateway and per-agent KB access control

What Teams Build With Knowledge Engine

Give agents domain expertise across any industry

Customer Support

Support agents that search product docs, FAQs, and troubleshooting guides to resolve issues with grounded, accurate answers.

Research & Analysis

Research agents that query across reports, papers, and internal documentation to synthesize insights from multiple sources.

Internal Knowledge

Agents that understand company policies, procedures, and institutional knowledge. Onboarding, HR, and compliance use cases.

Legal & Compliance

Search contracts, regulations, and case documents with semantic understanding of domain-specific language.

Frequently Asked Questions

What file formats are supported?

Plain text (.txt) and Markdown (.md) are read natively. PDF, DOCX, PPTX, XLSX, and image files (PNG, JPG, SVG) are automatically converted to searchable content.

How does hybrid search work?

Each query runs two parallel searches: dense vector similarity (semantic) and sparse BM25 (keyword). Results are merged using Reciprocal Rank Fusion in Qdrant to give you the best of both approaches.

What embedding models are available?

We use best-in-class embedding models optimized for accuracy and speed. Multiple model options available with configurable dimensions to balance precision and cost.

What is RAG and how does it work?

Retrieval Augmented Generation automatically searches your knowledge bases when agents run. Relevant chunks are injected into the agent prompt as context, so responses are grounded in your actual documents.

Can I control which agents access which knowledge bases?

Yes. Knowledge bases are scoped per agent and per gateway. An empty configuration means no access — KBs must be explicitly assigned. You can also set per-KB result limits and enable/disable reranking.

Give Your Agents Domain Knowledge

Upload documents, search semantically, and build grounded AI agents. Included in all tiers.