Vesonix AI Hub — Enterprise Intelligence Platform

Platform Overview

Everything your platform needs.
Nothing you'll never use.

Six tightly integrated capability layers that sit inside your existing Vesonix environment — no new SaaS subscriptions to wire together.

Multi-Agent Orchestration

Knowledge RAG Platform

Live Data Intelligence

Vector Store Abstraction

AI Usage Analytics

Access-Controlled AI

Platform Capability

Multi-Agent Orchestration

Intent classification, LLM-driven routing, dependency-aware execution, and multi-turn conversation state management.

Intent classified across 9 message types

LLM-driven agent selection at runtime

Multi-turn conversation state preserved

Explore Section

Intent Types

GPT-4o Claude Gemini Router

Platform Capability

Knowledge RAG Platform

Document ingestion from 12+ source types, hybrid semantic + keyword retrieval, and LLM answer synthesis with citations.

12+ document source connectors

Hybrid semantic + keyword retrieval

Cited answers from source documents

Explore Section

12+

Sources

PDF Sheets Databases APIs

Platform Capability

Live Data Intelligence

Natural language to SQL queries. Natural language to REST API calls. Secure, SELECT-only, timeout-enforced.

Natural language → SQL, zero SQL knowledge needed

Natural language → REST API call

SELECT-only, timeout-enforced safety rails

Explore Section

Agent Tools

RAG SQL REST API Entities

Platform Capability

Vector Store Abstraction

Pluggable adapter for pgvector, Pinecone, Weaviate, Azure AI Search, and Qdrant — switchable at runtime with no code changes.

pgvector, Pinecone, Weaviate, Azure AI Search, Qdrant

Switch backends at runtime — zero code changes

Hybrid retrieval with relevance scoring

Explore Section

Backends

pgvector Pinecone Weaviate Qdrant

Platform Capability

AI Usage Analytics

30-day dashboards with query counts, token accounting, USD cost estimation, per-agent invocation tracking, and latency metrics.

Query volume + token usage dashboards (30-day)

USD cost estimate per provider and per agent

Response latency tracked per agent invocation

Explore Section

30d

History

Queries Tokens Cost Latency

Platform Capability

Access-Controlled AI

Agent-level data source restrictions, AppActor RBAC, per-conversation source scoping, and superuser-gated admin endpoints.

Agent-level data source restrictions enforced

AppActor RBAC tied to existing platform roles

Full audit trail for every AI interaction

Explore Section

RBAC

Enforced

API Keys SQL Guard Roles Audit

AI Agents

Four capabilities.
Unlimited possibilities.

Every AI agent you configure has access to four powerful capabilities. Combine them to match exactly how your team works.

What agents can do

RAG

Smart Document Search

Searches every uploaded file, PDF, and knowledge base using AI — returns exact passages with source citations.

SQL

Live Database Query

Ask questions in plain language, get real answers from your live database — no SQL knowledge required.

API

Real-Time Data Fetch

Pulls live data from any connected system — stock levels, exchange rates, and third-party platform data.

Entities

Smart Record Lookup

Finds and filters platform records — customers, vendors, inventory — and respects each user's access rights.

Ready-to-use agent configurations

🛒

Procurement Advisor

Supply & Sourcing

Answers sourcing questions, checks vendor contracts, reviews purchase history, and recommends reorder quantities.

Sample queries

Which of our approved vendors is currently offering the lowest unit price for SKU-4421, and do we have any existing contract terms that lock us into a higher rate?

Show me all open purchase orders above $50,000 that have been approved but not yet received, grouped by vendor and sorted by order date.

We need to restock raw material RM-078 urgently — compare lead times and last quoted prices from our top 3 suppliers and recommend the best option.

List every vendor whose payment invoices are overdue by more than 45 days and flag any that have penalty clauses in their contracts.

Which supplier contracts are due for renewal in the next 60 days, and what were the original pricing terms we agreed on?

Document Search Database Query Record Lookup

👥

People & Policy Guide

HR & Compliance

Answers HR policy questions, explains leave entitlements, and helps employees navigate company guidelines.

Sample queries

I joined in March last year and have taken 8 days of casual leave so far — how many carry-forward privilege leaves am I still entitled to use this quarter?

Our team has a contractor who works 4 days a week remotely — does the current remote work policy cover them the same way it does full-time employees, and are there any approval steps needed?

If an employee on a mid-level grade resigns today, what is the official notice period, and can the company waive part of it if a replacement is found early?

What is the full process for raising a formal grievance against a manager, and within how many business days is HR required to respond under our policy?

Our team has been working weekends for the last month on a project deadline — does company policy allow overtime to be claimed as cash payment, or is it only converted to comp-off days?

Document Search

📊

Operations Analyst

Data & Reporting

Runs on-demand reports, queries live operational data, and pulls real-time metrics from connected systems.

Sample queries

Give me a breakdown of last month's top 10 products by revenue, and highlight which ones had a higher return rate than 5% so I can flag them for the quality review.

What is the current USD to INR exchange rate from the payment gateway API, and how does it compare to the rate we used when we last processed international invoices?

Which warehouses are currently operating above 85% utilisation capacity, and which SKUs are taking up the most space relative to their monthly movement rate?

How many customer orders placed in the last 30 days were fulfilled more than 2 days after the committed delivery date, and which fulfilment centres had the highest delay rate?

Show me the daily order volume trend for the past 6 weeks, split by sales channel, and highlight any day where volume dropped more than 20% compared to the 7-day average.

Database Query Real-Time Fetch Document Search

Orchestration Engine

Every message.
Classified, routed,
executed.

The orchestrator classifies incoming messages across 9 intent types, selects the right agent, resolves tool dependencies, and synthesises a coherent answer — all in one streaming round-trip.

9-class Intent Classifier Greetings, follow-ups, clarifications, action requests and more — each routed before any tool is called.

LLM-driven Agent Router Selects the best agent combination per conversation. Agent descriptions shape what the router picks at runtime.

Multi-turn Conversation State Every message stores its full trace — agents selected, tools called, tokens used — enabling genuine follow-up reasoning.

9:41

Vesonix AI 9 Intent Types

👋

Greeting / Help

"Hi! What can you help me with?"

💬

New Question

"What is my total inventory value?"

🔄

Follow-up

"Can you break that down by region?"

✅

Confirmation

"Yes, go ahead with that."

✖

Rejection

"No, show me something different."

🎯

Option Selection

"Option 2 — the quarterly view."

💡

Clarification

"I meant last quarter, not last year."

⚡

Action Request

"Create the purchase order now."

🔕

No Action Needed

"Got it, thanks for letting me know."

🏠

🔍

📊

⚙️

Knowledge Platform

Connect any data source.
Search all of it in plain English.

Ingest documents from 12+ source types, index them into any vector backend, and let your team ask questions in natural language — with cited answers.

12+

Source connectors

Vector backends

Hybrid

Semantic + keyword search

Cited

Answers with sources

📁

Local Folder

File Source

☁️

SharePoint

Microsoft

🗂️

Google Drive

Google

💬

Slack

Communication

📧

Communication

🏭

SAP DMS

Enterprise DMS

🐘

PostgreSQL

Database

🐬

MySQL

Database

🪟

Microsoft SQL

Database

🍃

MongoDB

Database

❄️

Snowflake

Cloud Warehouse

🌐

REST API

Custom Integration

📁

Local Folder

File Source

☁️

SharePoint

Microsoft

🗂️

Google Drive

Google

💬

Slack

Communication

📧

Communication

🏭

SAP DMS

Enterprise DMS

🐘

PostgreSQL

Database

🐬

MySQL

Database

🪟

Microsoft SQL

Database

🍃

MongoDB

Database

❄️

Snowflake

Cloud Warehouse

🌐

REST API

Custom Integration

🌐

REST API

Custom Integration

❄️

Snowflake

Cloud Warehouse

🍃

MongoDB

Database

🪟

Microsoft SQL

Database

🐬

MySQL

Database

🐘

PostgreSQL

Database

🏭

SAP DMS

Enterprise DMS

📧

Communication

💬

Slack

Communication

🗂️

Google Drive

Google

☁️

SharePoint

Microsoft

📁

Local Folder

File Source

🌐

REST API

Custom Integration

❄️

Snowflake

Cloud Warehouse

🍃

MongoDB

Database

🪟

Microsoft SQL

Database

🐬

MySQL

Database

🐘

PostgreSQL

Database

🏭

SAP DMS

Enterprise DMS

📧

Communication

💬

Slack

Communication

🗂️

Google Drive

Google

☁️

SharePoint

Microsoft

📁

Local Folder

File Source

Pluggable vector backends Switch without code changes

Built-in Zero Config

Python-side cosine similarity on the DocumentEmbedding table. Always available, zero configuration.

pgvector PostgreSQL Extension

Approximate nearest-neighbor search within your existing PostgreSQL database infrastructure.

Pinecone Managed Cloud

Fully managed serverless vector search. Configured with index name, endpoint, and API key.

Weaviate Self-hosted / Cloud

Open-source vector database. Supports self-hosted or Weaviate Cloud deployments.

Azure AI Search Azure Service

Enterprise-grade vector search from Microsoft Azure. REST endpoint + API key from environment.

Qdrant Self-hosted / Cloud

High-performance vector database. Supports self-hosted or Qdrant Cloud.

Model Layer

Bring your own AI provider.
Switch without code changes.

The AI Hub sits between your agents and the model APIs. Swap any provider by changing one environment variable — no deployment needed.

4 chat providers 3 embedding providers Mix & match freely API-key based config

AI Hub — Core Infrastructure

Unified Model Router

Routes every agent request to your configured LLM for chat and to your embedding provider for semantic search — independently, at runtime.

Chat & Completion

Vector Embedding

Chat & Completion Providers LLM

Used for every agent conversation turn — generates answers, reasoning, and responses.

OpenAI

GPT-4o mini

Fast & precise Default

Anthropic

Claude Haiku 4.5

Deep reasoning

Google Gemini

Gemini 2.5 Flash

Long context

Azure OpenAI

Custom deployment

Enterprise VPC

Embedding Providers Vectors

Used when indexing documents and running semantic search across your knowledge base.

OpenAI

text-embedding-3-small

1,536 dims Default

Google Gemini

gemini-embedding-001

768 dims

Azure OpenAI

Custom deployment

Custom dims

Anthropic

No embedding API offered

Use OpenAI or Gemini

Security & Access Control

AI that knows its
boundaries.

Every AI interaction passes through four independent security layers — from key storage to query filtering to access control to an immutable audit trail.

Defense in depth

Protected

Layer 1

API Keys in Environment Only

LLM and vector store keys stored as environment variables — never in the database. Only the variable name is persisted.

Layer 2

Role-Based Access Control

Agents and document sources restricted to specific AppActors via M2M. EntityRole governs per-user record access — no separate permission setup.

Layer 3

SQL Injection Prevention

Only SELECT statements allowed. Every query is validated before execution — with a hard 50-row result limit and 10-second timeout enforced.

Layer 4

Immutable Audit Trail

Messages are append-only — never edited. Every entry stores provider telemetry, token counts, agent used, and the full request trace.

Common Questions

Questions people actually ask.

Click any question to read the answer.

Can I switch LLM providers without redeploying?

AI Hub

Yes. The provider is stored in the AIProviderConfig database singleton. Update the environment variable or patch the config via the admin API — the platform syncs at next startup with no redeployment required.

Are API keys ever stored in the database?

AI Hub

Never. LLM and embedding API keys are read exclusively from environment variables. Vector store configs store only the variable name — the adapter resolves the key at runtime. This applies to all providers.

How does agent routing work? Can I control which agents answer which questions?

AI Hub

The orchestrator sends the user query along with all accessible agents (name + description) to the routing LLM. The router selects relevant agents and assigns per-agent sub-queries. Write clear agent descriptions to control routing. Restrict agents further with actor-based access control.

What protects against SQL injection when querying external databases?

AI Hub

The query_database tool explicitly rejects any statement that is not a SELECT before execution. Queries run with a 50-row result cap and a 10-second timeout. Database credentials are never shared with the LLM.

How does hybrid retrieval work in Doc Hub?

AI Hub

Each chunk is scored by semantic similarity (cosine distance) and keyword overlap. Final score = semantic × 0.7 + keyword × 0.3. Both weights are configurable via DOCHUB_RERANK_SEMANTIC_WEIGHT and DOCHUB_RERANK_KEYWORD_WEIGHT environment variables.

Can I use Anthropic Claude for embeddings?

AI Hub

No — Anthropic does not provide a native embedding API. Configure OpenAI or Google Gemini as your embedding provider independently. The platform supports separate LLM and embedding provider selection for exactly this reason.

AI for your documents,
systems and workflows.

Everything your platform needs.
Nothing you'll never use.

Multi-Agent Orchestration

Knowledge RAG Platform

Live Data Intelligence

Vector Store Abstraction

AI Usage Analytics

Access-Controlled AI

Four capabilities.
Unlimited possibilities.

Every message.
Classified, routed,
executed.

Connect any data source.
Search all of it in plain English.

Bring your own AI provider.
Switch without code changes.

Full visibility into
every AI interaction.

AI that knows its
boundaries.

Questions people actually ask.

Your platform.
Now with AI.

AI for your documents,systems and workflows.

Everything your platform needs.Nothing you'll never use.

Multi-Agent Orchestration

Knowledge RAG Platform

Live Data Intelligence

Vector Store Abstraction

AI Usage Analytics

Access-Controlled AI

Four capabilities.Unlimited possibilities.

Every message.Classified, routed,executed.

Connect any data source.Search all of it in plain English.

Bring your own AI provider.Switch without code changes.

Full visibility intoevery AI interaction.

AI that knows itsboundaries.

Questions people actually ask.

Your platform.Now with AI.

AI for your documents,
systems and workflows.

Everything your platform needs.
Nothing you'll never use.

Four capabilities.
Unlimited possibilities.

Every message.
Classified, routed,
executed.

Connect any data source.
Search all of it in plain English.

Bring your own AI provider.
Switch without code changes.

Full visibility into
every AI interaction.

AI that knows its
boundaries.

Your platform.
Now with AI.