Whitepaper v1.0

OMRA AI Gateway

Multi-tenant LLM orchestration platform with intelligent routing, O₮ token economy, and autonomous agents

April 2026 — Version 1.0

10+
LLM Providers
26+
Integrations
O₮
Token Economy
70/30
Revenue Share

Table of Contents

1. The Problem

For Business

  • Vendor lock-in — you integrate GPT-4, and three months later DeepSeek is 10x cheaper for the same tasks. Rewriting the integration is expensive.
  • Uncontrolled costs — a misconfigured agent can burn your budget overnight.
  • Siloed tools — CRM, ERP, marketplaces, accounting are all through different APIs. The AI assistant can only answer, not act.
  • Security — API keys in code, no audit trail, no approval gates for critical operations.

For Developers

  • Every LLM provider has its own SDK, response format, and rate limits.
  • RAG needs to be built from scratch for each project.
  • No standard way to create a tool marketplace.

2. The Solution

OAG is an AI Operating System for B2B that solves these problems architecturally.

Unified API Gateway

Clients use the standard OpenAI SDK — OAG is compatible at the Assistants v2 protocol level. Switching models is a single config change.

Client (OpenAI SDK) → OAG Gateway → Routing Engine → LLM Provider                                     ├── Billing Middleware                                     ├── RAG Engine                                     └── Tool Sandbox

Intelligent Routing

content Images → Gemini Flash
volume > 50k tokens → long-context
time 22:00–06:00 → DeepSeek
client Enterprise → Claude Opus
fallback Provider down → auto-failover

Orchestrator

Real-time chat via WebSocket. Full prompt stack: system → company → persona → RAG → tools. Token streaming straight to UI.

Sub-Agents

Autonomous workers on cron schedules or events. Lightweight prompt, fire-and-forget, full logging.

26+ Integrations

E-Commerce
Ozon, Wildberries, Class365
Payments
YooKassa, Tinkoff, MercadoPago
Logistics
CDEK
Productivity
Google Workspace
Tax
AFIP/ARCA
Media & AI
OCR, Vision, Scraper

3. Architecture

Technology Stack

Runtime: Python 3.12 async
API: FastAPI + Uvicorn
ORM: SQLAlchemy 2.0
Database: PostgreSQL 16 + pgvector
Cache: Redis 7
Storage: MinIO (S3)
Frontend: Jinja2 + HTMX + Tailwind
Encryption: AES-256-GCM
Scheduling: APScheduler + croniter

Multi-Tenancy

Every resource is scoped to company_id. No shared namespaces. Full data isolation between companies.

Security

  • API keys encrypted at rest (AES-256-GCM)
  • Group + Tag ACL for tools and files
  • Immutable audit log (90-day retention)
  • HITL approval gates for critical ops

4. O₮ Token Economy

O₮ (Omra Token) is the platform's internal currency, normalizing costs across providers to a single scale.

O₮ USD
1$0.001
1,000$1.00
100,000$100.00

Revenue Sharing

Creator
Platform
External call
70%
30%
Own call
0%
100%

5. Multi-Agent System

Delegation

Synchronous map-reduce. Parent spawns workers, collects results.

Event Bus

Async Pub/Sub. Agents subscribe to topics via glob patterns.

Webhooks

External systems trigger agents via HTTP. Zero coupling.

HITL Approval Gates

For high-stakes operations (payments, VEP generation, bulk mailings) agents request human approval. Notifications via Telegram, decisions in Cockpit UI or inline buttons.


6. RAG Pipeline

Upload
PDF, DOCX, XLSX, TXT, MD
Parse
Chunk
Embed
pgvector 1536-dim
Retrieve
cosine ≥ 0.28
Inject
system prompt

7. Target Markets

E-Commerce (RU/LATAM)

Marketplace sync (Ozon, Wildberries, MercadoLibre), order management, pricing automation.

Finance & Tax

Tax reporting automation (AFIP/ARCA), debt monitoring, VEP generation.

B2B SaaS

White-label AI assistant — connect a provider, upload knowledge, configure tools. Full data isolation.

AI Tool Creators

Tool marketplace with revenue sharing — create a tool, earn 70% per call.


8. Competitive Advantages

Aspect OAG Alternatives
Multi-provider 10 providers, single API 1-2 providers
Routing Rules by content, time, budget Manual model selection
Billing Native O₮ token economy External billing or none
Tools Sandbox + marketplace + revenue sharing Hardcoded function calls
Agents Cron + Event Bus + HITL + delegation Interactive chat only
Self-hosted Docker Compose, full control SaaS only

9. Conclusion

OAG is not a ChatGPT wrapper. It's a full-fledged operating system for AI agents in business:

  • One API instead of ten SDKs
  • Smart routing instead of manual model selection
  • Token economy instead of uncontrolled costs
  • Autonomous agents instead of chat-only
  • Tool marketplace instead of hardcoded integrations
  • Full audit trail instead of a black box

The platform is already running in production, serving e-commerce and financial processes for companies in Russia and Latin America.

Ready to get started?

Leave your details — we'll get back to you within 24 hours.