AIgilityX · Sovereign Suite
Full-Stack AI & Voice Engineering
FastAPI · LangChain · ChromaDB
aigilityx.com (Docker VPC)
Multi-Provider LLM Orchestration
Dynamic routing to optimal frontier model per task — synthesis, analysis, or vision.
Vector RAG Semantic Retrieval
ChromaDB cosine similarity search over embedded enterprise documents with sub-100ms P95.
Sovereign Containerized Deployment
One-command Docker Compose deployment in private VPC. Zero data leaves the sovereign perimeter.
Siloed LLM Wrappers vs. Resilient Autonomous Agent Teams
Modern enterprise teams need AI agents that can reason over internal documents, switch between frontier models, and converse naturally by voice — but existing solutions hardcode a single LLM vendor, lack vector-backed document retrieval, and ship disconnected pipelines for speech and text.
Single-Provider Fragility
Hardcoding a single LLM vendor prevents routing specialized tasks (CISO analysis vs. general synthesis) to optimal models, creating vendor lock-in.
Ungrounded Chat Interfaces
Standard wrappers lack vector-backed document ingestion, real-time WebSocket streaming, and low-latency voice transcription/synthesis pipelines.
Three Strictly Decoupled Production Tiers
Frontend Workstation
React 18 + TypeScript + Vite frontend with real-time ChatWindow token streaming, VoiceRecorder & AudioPlayer, AgentSelector persona switching, and Supabase auth.
FastAPI Async Gateway
REST & WebSocket endpoints with Pydantic v2 validation, structlog JSON telemetry, agent orchestration routing, and voice pipeline coordination.
Agent Core & Neural Voice
LangChain 0.3 multi-persona orchestrator with ChromaDB vector RAG, OpenAI Whisper STT, and ElevenLabs streaming TTS for end-to-end voice intelligence.
Production Engineering Tooling
FastAPI 0.115
Async Python · Pydantic v2 · Uvicorn
High-throughput async gateway handling REST endpoints, WebSocket streaming, and voice pipeline orchestration.
React 18
Vite · Zustand · Supabase Auth
Real-time ChatWindow with token streaming, VoiceRecorder, AudioPlayer, and AgentSelector persona switching.
TypeScript
Strict Mode · ESLint 9
End-to-end type safety across the React frontend, API contracts, and Zustand global state management.
LangChain 0.3
Multi-Chain · Agent Registry
Coordinates configurable agent personas across multi-provider LLMs with grounded RAG context injection.
ChromaDB
text-embedding-3-small · Cosine
Semantic vector search over ingested enterprise documents for grounded, hallucination-resistant responses.
MongoDB
Motor Async · TTL Indexes
Persistent async conversation memory enabling multi-turn context recall across agent interactions.
OpenAI Whisper
STT · Language Detection
Real-time speech-to-text transcription with automatic language detection and noise suppression.
ElevenLabs
Streaming TTS · Neural Voice
Low-latency neural voice synthesis streaming AI responses back to the browser audio player.
GPT-4o
128K Context · Vision · Tools
Primary LLM for general synthesis, document summarization, and multi-modal reasoning tasks.
Claude 3.5 Sonnet
200K Context · Extended Think
Specialized LLM for deep cybersecurity analysis, threat modeling, and compliance assessments.
Docker
Compose · Multi-Stage Build
One-command containerized deployment bundling FastAPI, React, ChromaDB, and MongoDB services.
Python 3.11
AsyncIO · Pydantic · structlog
Core backend runtime executing agent orchestration, voice pipelines, and telemetry logging.
Verified Production Topology Blueprint
The CyberX AI platform deployment topology — React frontend, FastAPI gateway, LangChain orchestrator, ChromaDB vector store, MongoDB session persistence, and dual voice pipelines — all containerized via Docker Compose.
Interactive Code Workbench
import requests
# CyberX AI — Multi-Agent Chat
resp = requests.post(
"https://aigilityx.com/api/v1/chat",
headers={"Authorization": "Bearer cyberx_sk_..."},
json={
"agent": "ciso",
"message": "Analyze SOC 2 compliance posture",
"stream": True,
"rag_enabled": True,
}
)
for chunk in resp.iter_lines():
print(chunk.decode())3 Active
CISO · Analyst · General — pluggable agent factory
GPT-4o + Claude
Dynamic multi-provider routing per task type
Whisper + ElevenLabs
Real-time STT → LLM → TTS voice loop
10-Step End-to-End Request Journey
Click any step to trace the journey of a multi-agent AI chat completion — from client dispatch through persona resolution, RAG retrieval, and voice synthesis:
Client Dispatch & Auth
Client sends POST to /api/v1/chat with Bearer token and selected agent persona (CISO, Analyst, or General).
Agent Persona Resolution
FastAPI gateway resolves the requested persona from the pluggable Agent Registry, loading system prompts and model routing rules.
Multi-Provider LLM Routing
Orchestrator routes to optimal model — GPT-4o for synthesis tasks, Claude 3.5 Sonnet for deep cybersecurity analysis.
ChromaDB RAG Retrieval
Query is embedded via text-embedding-3-small and matched against ChromaDB vector store for grounded document context.
LangChain 0.3 Orchestration
LangChain assembles the final prompt chain: system persona + RAG context + conversation history + user query.
WebSocket Token Streaming
Tokens stream in real-time over persistent WebSocket connection (/ws/chat/{session_id}) for instant UI rendering.
MongoDB Session Persistence
Full conversation history stored asynchronously via Motor driver to MongoDB, enabling multi-turn context recall.
Whisper Voice Transcription
Audio input transcribed via OpenAI Whisper STT with automatic language detection and noise suppression.
ElevenLabs Voice Synthesis
AI response streamed through ElevenLabs TTS pipeline for natural, low-latency voice output in browser.
Telemetry & Audit Logging
Request latency, token counts, model used, and agent persona logged via structlog JSON for production observability.
Want to deploy autonomous, multi-agent AI infrastructure?
We design and deploy production-grade agent platforms with multi-LLM routing, grounded RAG pipelines, neural voice I/O, and sovereign containerized deployments.