Marketplace / Blueprint 03
LIVEAI Control Plane / intelligence

Universal AI Gateway

One governed API in front of cloud and local AI providers.

Review my build

Interactive platform map

Architecture in context

The focused blueprint, its required foundation, and declared recommendations.

Core Selected Automatic Required path
Blueprint 03 / AI Control Plane

Universal AI Gateway

One governed API in front of cloud and local AI providers.

1 vCPU1 GB RAM2 services
01 / Problem

What this replaces

Applications become coupled to one model vendor, credentials spread across apps, and teams lose centralized visibility into spend and routing.

02 / Outcome

What your team gains

A consistent OpenAI-style API with centralized authorization, model abstraction, routing, cost controls and provider resilience.

03 / Capability

What is inside the blueprint

100+ model/provider abstraction
OpenAI-compatible chat and Responses patterns
Embeddings, images, audio, batches and additional provider translations
Consistent response and error formats
Streaming
Retries and fallbacks
Application-level load balancing
Virtual keys
Authentication and authorization
Per-user/project spend tracking and budgets
Rate limits
Per-project logging, guardrails and caching
Admin dashboard
Observability callbacks
04 / Architecture

How it fits the platform

Applications/agents -> LiteLLM gateway -> policy/routing/budget layer -> OpenAI, Anthropic, Google, Azure, Ollama/local endpoints and other providers.

Included services

LiteLLM, ai-stack-postgres

Platform requirements
05 / Delivery

From prerequisites to operation

Prerequisites
  1. Provider credentials or local model endpoints
  2. Database for gateway state
  3. Defined model naming/routing policy
Deployment
  1. Deploy LiteLLM proxy
  2. Register providers and model aliases
  3. Create virtual keys/teams
  4. Enable spend logging and budgets
  5. Configure fallback chains and observability
Configuration
  1. Model aliases
  2. Fallback/retry policy
  3. Budget/rate limits
  4. Caching and guardrails
  5. Logging callbacks
  6. User/project keys
Operations
  1. Provider availability
  2. Spend drift
  3. Rate-limit events
  4. Schema/version compatibility
  5. Gateway DB backups
06 / Combinations

What this unlocks with other layers

Private AI Workspace + Universal AI Gateway

Private AI + Model Freedom

Users get one interface while builders can change providers underneath it without changing the experience.

07 / Technology

Technology behind this capability

LiteLLMRUNNINGPostgreSQLRUNNING - three separate DB instances in census