Skip to content
Cloud Score Docs

Welcome to Cloud Score

Cloud Score sits between your application and the model providers, picks the right model for every request under rules you control, and records what each request cost.

On this page

What Cloud Score does

Your application sends its requests to Cloud Score instead of to a provider. For every request, Cloud Score scans the prompt for sensitive content, works out what kind of task it is, finds the enabled models that can do it, applies your policies and budgets, checks whether the answer is already in its cache, picks the winning model on cost, speed, and quality, runs it with retries and failover, checks the answer, and records exactly what happened and what it cost.

You manage all of that from the Cloud Score admin app: which providers are connected, which models may compete, the rules that shape each decision, and the keys your applications use.

Your application

Sends requests to Cloud Score with a dex_sk_ key, through the native API or the OpenAI-compatible endpoint.

Cloud Score gateway

Guard, classify, match, policy, cache, select, execute, record

Model providers

OpenAI, Anthropic, Google Gemini, Grok (xAI), OpenRouter, Together AI, Azure AI Foundry, Google Vertex AI, Amazon Bedrock

How a request travels

Every request runs the same sequence of stations. You will meet these names on the Playground's Route board and in every request's detail in Observability.

Guard inscan the prompt
Classifywhat kind of work
Matchwhich enabled models can do it
Estimatewhat each would cost
Policyyour rules and budgets
Resolvealready answered? serve from cache
Selectpick the winner
Executecall it, retry, fail over
Guard outscan the answer
Validatenon-empty, valid JSON
Recordwhat happened, what it cost

A request ends in one of three recorded states: completed when a model answered or a cached answer was served, blocked when a policy, a budget, or AI Security refused it, and failed when Cloud Score tried and could not deliver.

What is in your workspace

Each account is its own workspace. Overview and Playground sit above the Analytics, Gateway, Agents and Admin groups. This is what each is for. See The sidebar for every entry.

Sidebar entry or groupWhat it is for
OverviewRequests, total spend, provider latency, estimated savings and success rate, with Gateway insights, Agent usage and your saved dashboards.
PlaygroundSend a request with your gateway key and inspect its routing evidence.
AnalyticsOptimizations, Usage & cost, Budgets, Cost centers, ROI & Impact, and Observability (Requests and Gateway health).
GatewayProviders, Models, Routing, Guardrails and Optimization settings.
AgentsThe registry of agents that call the gateway.
AdminTeams and Members for Admins; API keys, Organization and Developer for every member.

Where to start

Related