Turn Your Company Documents Into a Production AI Knowledge Engine in 60 Seconds
Replace fragmented vector pipelines with a single high-performance PostgreSQL + pgvector engine. Grounded answers, $0 local embeddings via Ollama, hybrid BM25 + HNSW search, and a 1-line script embed for any website.
One Sovereign Engine. Three Purpose-Built Workflows.
Whether you want an instant drop-in customer support bot, clean headless vector APIs for your microservices, or private VPC multi-tenant compliance, Retriever delivers without vendor lock-in.
1-Line Embed Copilot
Drop a single <script> tag onto Shopify, Webflow, WordPress, or Next.js. Turn stagnant FAQs, catalogs, and documentation into a 24/7 lead capture and support deflection assistant.
- ✓60-Second Drop-In: Zero code changes, paste anywhere in <head>
- ✓Intent Lead Capture: Gathers contact info & escalates hot leads
- ✓Zero Hallucinations: Strict vector grounding on your verified docs
- ✓Design Parity: Customizable fonts, colors, and prompt buttons
Headless Vector & REST API
Direct programmatic access with scoped API keys (ret_live_...). Full parity across TypeScript and Python SDKs, SSE streaming chat, and hybrid BM25 + HNSW + ColBERT MaxSim fusion.
- ✓REST & SSE Streaming: Clean /v1/tenants/{id}/search and /chat
- ✓$0 Embedding Cost: Built-in local Ollama nomic-embed-text engine
- ✓38 Production Batteries: Ebbinghaus memory, Graph-of-Thought, DSPy
- ✓Universal SDKs: Native TypeScript and Python client packages
Sovereign Multi-Tenancy & Privacy
PostgreSQL Row-Level Security (RLS) guarantees complete tenant isolation. Deploy in private VPCs, air-gapped enclaves, or dedicated Oracle Cloud VPS instances.
- ✓Strict DB Multi-Tenancy: PostgreSQL RLS prevents cross-tenant data leaks
- ✓Zero Data Retention: Your documents are never retained or trained on
- ✓Privacy Batteries: MPC enclave execution & ZKP proof attestation
- ✓Audit Posture: HIPAA, GDPR, and SOC2 aligned architecture
The Fragmented Stack vs. The Retriever Hexagonal Core
Compare the bloated, multi-vendor AI mesh of 2024 against Retriever's unified PostgreSQL pgvector engine.
The Fragmented AI Stack
6+ Disconnected VendorsRetriever Hexagonal Core
1 Unified EngineEngineered Without Bloat or Multi-Vendor Glue
Retriever integrates core production RAG capabilities into a single high-performance engine—eliminating brittle multi-vendor dependencies.
Dense pgvector HNSW
Hierarchical Navigable Small World graph indexing directly in PostgreSQL for ultra-fast approximate nearest neighbor vector search.
BM25 Okapi Sparse Search
Exact term frequency and inverse document frequency scoring for precise acronyms, code identifiers, and SKU matching.
ColBERT MaxSim Multi-Vector
Fine-grained token-level late interaction computing maximum cosine similarity across all query tokens.
Reciprocal Rank Fusion (RRF)
Rank-based score normalization combining sparse and dense candidate lists without arbitrary score-scale distortion.
Cross-Encoder Re-ranking
Deep transformer cross-attention scoring between query and top retrieved chunks for optimal context ranking.
Parent-Child Chunk Expansion
Indexes granular child chunks for precise vector hits while restoring full parent document context during generation.
Sub-15ms Semantic Vector Cache
Pre-indexes query vector embeddings with cosine similarity gating to instantly serve cached responses at zero LLM cost.
Hybrid Dense-Sparse Fusion
Dynamically balances semantic vector embeddings and exact keyword tokens based on query intent classification.
GraphRAG Leiden Clustering
Extracts entity-relation triples and clusters global knowledge communities for multi-hop analytical reasoning.
Graph-of-Thought (GoT) Planning
Generates non-linear reasoning graphs allowing thoughts to branch, merge, and backtrack when dead ends are hit.
Recursive Language Modeling (RLM)
Sandboxed Python execution environment allowing LLMs to run code, verify mathematical calculations, and iterate autonomously.
Multi-Agent Consensus Arbitration
Orchestrates multi-model debate between heterogeneous LLMs with automated majority voting and referee arbitration.
DSPy Automated Prompt Optimization
Algorithmic prompt compilation and bootstrap few-shot optimization maximizing downstream assertion metrics.
Self-Aware Corrective RAG (CRAG)
Evaluates document retrieval confidence before generation; rejects unbacked prompts instead of inventing answers.
Agentic Workflow DAG Orchestration
Directed acyclic graphs for complex multi-step enterprise workflows with deterministic state transitions and checkpoints.
Ebbinghaus Spacing Memory
Mathematically models human memory retention curves, prioritizing reinforced knowledge while gradually decaying stale state.
Dynamic Working Context Window
Dynamically allocates prompt token budgets between system constraints, active chat turns, and retrieved knowledge.
Episodic Memory Buffers
Stores key user preferences, historical outcomes, and resolved edge cases across asynchronous conversation threads.
Cross-Session Context Persistence
Preserves user agent memory across browser restarts and device transitions with strict tenant database isolation.
Semantic Vector State Eviction
Automatically cleans outdated and superseded document embeddings to maintain high index hygiene and low latency.
Docling Layout-Aware OCR
Parses complex multi-column PDFs, extracting bounding boxes, headings, and semantic hierarchy.
Baidu PP-OCRv4 Vision Pipeline
High-accuracy text and character recognition on low-resolution scanned documents and photographic inputs.
Structured Table & Markdown Extraction
Converts tabular data into structured Markdown and JSON representations, preserving column relationships.
Presigned Source PDF Citations
Generates secure, expiring download URLs allowing users to audit exact source documents directly from answers.
Cloud Data Ingestion Connectors
Background synchronization pipelines for Google Drive, Notion, REST endpoints, and custom event webhooks.
Live Web Grounding Fallback
Automatically queries real-time web search APIs when internal tenant knowledge is silent on breaking topics.
PostgreSQL RLS Multi-Tenancy
Enforces PostgreSQL Row-Level Security on every query; vectors and documents never leak across tenant boundaries.
Deterministic Grounding & Citations
Extracts exact character ranges and page coordinates so every answer can be audited against original source text.
SHA-256 Merkle Audit Verification
Builds cryptographic hash trees across document chunks to verify dataset integrity and detect unauthorized document edits.
Resilient Circuit Breaker & Failover
Monitors upstream LLM and embedding latency, automatically failing over to backup providers when thresholds exceed SLA.
Celery & RabbitMQ Task Swarm
Processes massive document ingestion batches concurrently without blocking synchronous chat API requests.
Local-First Ollama ($0 Embeddings)
Generates high-dimensional vector embeddings locally using nomic-embed-text with zero external API fees or token limits.
SHA-256 Tamper-Evident Audit Chains
Cryptographic hash-linked audit logging with automated 90-day retention pruning for enterprise compliance.
Real-Time SSE Streaming API
Streams tokens immediately via Server-Sent Events with structured citation events emitted dynamically during generation.
1-Line Embeddable Script Widget
Drop a single script tag onto any website, Shopify store, or web portal to deploy a production AI concierge in seconds.
Microsecond Structured Telemetry
Distributed trace IDs and microsecond latency tracking logged for every retrieval step and LLM token generation.
Automated Regression Benchmark
Statistical regression test suite asserting retrieval quality and relevance thresholds before releases.
Self-Querying JSONB Metadata Filters
Automatically translates conversational constraints into PostgreSQL JSONB queries for precise filtered retrieval.
Calculate Your Annual AI Infrastructure Savings
See how much your organization saves by replacing metered cloud vector databases and token embedding APIs with Retriever.
Empirical Load & Latency Benchmarks
Zero synthetic math, zero mocked responses. Real microsecond timings measured under concurrent load directly against our production Oracle Cloud VPS and PostgreSQL vector engine.
| Surface Tested | Requests | Avg Latency | P50 | P95 | Errors |
|---|---|---|---|---|---|
| PostgreSQL & Pool Readiness | 10 | 2377.92 ms | 2407.74 ms | 2440.66 ms | 0% |
| ASGI Gateway & Nginx Routing | 10 | 284.63 ms | 263.61 ms | 346.36 ms | 0% |
| Multi-Tenant Control Plane | 10 | 3770.91 ms | 3788.28 ms | 3902.16 ms | 0% |
Customize & Preview Your Live Widget
Configure colors, branding, and placement in seconds. Test the widget preview directly below or copy the 1-line script for your website.
<script
src="https://rag.prateeq.in/widget.js"
data-tenant="YOUR_TENANT_ID"
data-key="YOUR_API_KEY"
data-color="#2563eb"
data-title="Retriever Concierge"
data-position="bottom-right">
</script></body> tag on any website, Shopify store, or WordPress site.Engineered Beyond Basic ChatGPT Wrappers
Built for businesses that cannot afford hallucinations or slow customer service response times.
| Feature / Capability | 🚀 Retriever AI | Legacy Chatbot Wrappers |
|---|---|---|
| Retrieval Architecture | ✓ Hybrid pgvector HNSW + BM25 Keyword + Cohere Rerank | ✕ Standard Single Vector Search |
| Hallucination Control | ✓ Self-Aware CRAG Guardrails (Rejects Unbacked Queries) | ✕ None (Prone to inventing answers) |
| Source Verification | ✓ Presigned PDF Downloads with Exact Page Numbers | ✕ Plain Text / No Verifiable Citations |
| Latency & API Costs | ✓ Sub-50ms Semantic Cache (Saves up to 60% Token Costs) | ✕ Full LLM Inference & High Cost Every Query |
| Deployment Options | ✓ 1-Line Drop-in Script, REST API, & App Studio | ✕ Complex Setup or Locked Platform |
| LLM Provider Flexibility | ✓ Multi-Model BYOK (OpenAI, Gemini 3.6, Llama 3.3 70B, Claude 3.5, DeepSeek) | ✕ Locked to single vendor / hardcoded model |
| Knowledge Graph Traversal | ✓ GraphRAG Multi-Hop Triples (Subject -- Predicate --> Object) | ✕ None (Vector Similarity Only) |
| Multi-Modal Document Parsing | ✓ Baidu PP-OCRv4 Deep Learning + LLM Vision Engine | ✕ Plain Text Extraction / Fails on Scanned PDFs & Charts |
| Live Web Grounding Fallback | ✓ Automated Tavily & Brave Search Integration | ✕ None (Fails when internal docs lack context) |
| Metadata & Tag Filtering | ✓ Self-Querying Natural Language to SQL JSONB Filters | ✕ Manual Hardcoded Tags / No Structured Filters |
| Security & Compliance Audit | ✓ SHA-256 Cryptographic Audit Chain & 90-Day Retention Pruning | ✕ Unverified Plain Text Logs / Manual Deletion |
| Multi-Tenant Security | ✓ Isolated Tenant Schema + AES-256 Key Encryption | ✕ Shared Unencrypted Context |
Build Custom RAG Apps via REST API & SDK
Integrate vector search, custom system prompts, and streaming inference into your existing applications in minutes.
curl -X POST https://rag.prateeq.in/v1/tenants/YOUR_TENANT_ID/chat/sessions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "X-User-ID: YOUR_USER_ID" \
-H "Content-Type: application/json" \
-d '{"user_id": "YOUR_USER_ID"}'
# Stream chat response:
curl -N -X POST https://rag.prateeq.in/v1/tenants/YOUR_TENANT_ID/chat/sessions/SESSION_ID/messages \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "X-User-ID: YOUR_USER_ID" \
-H "Content-Type: application/json" \
-d '{"query": "What are your core platform security SLA requirements?", "stream": true, "web_search_grounding": true, "metadata_filter": {"year": 2026}}'Simple, Self-Serve Subscription Plans
Scale your knowledge base effortlessly. Zero hidden fees. Cancel anytime.
Starter
Ideal for small websites, blogs, and personal projects.
- ✓ 1 Workspace / Tenant
- ✓ 20 Documents (~50MB)
- ✓ 1,000 Chat Queries / mo
- ✓ Llama 3.3 70B & Gemini 2.5
- ✓ Standard Support
Pro
For growing businesses, legal teams, and e-commerce stores.
- ✓ 5 Workspaces / Tenants
- ✓ 100 Documents (~500MB)
- ✓ 5,000 Chat Queries / mo
- ✓ Remove 'Powered by' Branding
- ✓ Presigned Citation PDF Downloads
- ✓ Priority Hybrid Search & Re-ranking
Business
For agencies, medical networks, and high-traffic platforms.
- ✓ Unlimited Workspaces
- ✓ 500 Documents (~2.5GB)
- ✓ 20,000 Chat Queries / mo
- ✓ Dedicated Private Tenant RLS Isolation
- ✓ Custom Domain Mapping
- ✓ 99.9% Uptime SLA & 24/7 Support
Need a Dedicated or Private Cloud Cognitive Engine?
We engineer custom on-premise RAG pipelines, multi-modal vector search systems, and HIPAA/GDPR-compliant enterprise knowledge hubs tailored to your compliance boundaries.