Core Concepts#

This page introduces the main concepts, capabilities, and terms used in OptScale AI.

For how these components work together during request processing, see Architecture Overview. For Admin Console navigation and setup order, see Admin Console Overview.

Key capabilities#

• AI Gateway — Provides a governed access layer for AI requests, connecting applications and users to configured providers and models. Supports routing, fallbacks, and traffic distribution.

• FinOps and usage analytics — Tracks token usage, cost, and provider activity across teams and workloads. Optimization savings are available in Analytics → Optimizations.

• Governance — Applies policies and guardrails to control how AI requests and responses are evaluated and handled according to organization requirements.

• Context compression — Reduces the number of tokens sent to AI providers while preserving relevant conversation context. Compression can be configured for users, teams, or agents. See Context compression.

• MCP integrations — Connects OptScale AI to approved external tools and data sources through Model Context Protocol servers.

• Observability — Provides traces, usage data, cost information, and request-level diagnostics for monitoring and troubleshooting AI workloads.

Interfaces and integrations#

• Admin Console — Administrative interface for configuring providers, access controls, policies, guardrails, MCP servers, teams, and other organization settings.

• Chat — End-user workspace for interacting with approved AI models, tools, files, web access, and conversation history.

• External tools — Compatible clients such as OpenCode and Cursor that connect to OptScale AI Gateway using connection values provided in Chat. See External Tools.

Glossary#

Admin Console — Administrative interface for configuring and monitoring OptScale AI. See Admin Console Overview.

Agent — Application identity configured with provider access and optimization settings under Config → Agents.

AI Gateway — Governed access layer that routes Chat, API, and external-client requests to configured AI providers. See AI request flow.

Allowed providers — Providers assigned to a user and available for use in Chat and related access paths. See Allowed providers.

API key — Connection credential shown in Chat and used by compatible external clients to access OptScale AI Gateway. See Get connection values from Chat.

Chat — End-user workspace for interacting with approved AI models and tools. See Interface Overview.

Context compression — Optimization that reduces request token volume while preserving relevant context. Configured for users, teams, or agents. See Context compression.

Guardrail — Reusable control linked to a policy to evaluate or handle request or response content. See Guardrails.

MCP server — Connection to external tools or data sources through the Model Context Protocol. See MCP Servers.

Member — Organization role that uses Chat but does not access the Admin Console or manage organization resources. See Organization roles.

Model name — Identifier used to select a model through OptScale AI and configure compatible external clients. See Model names.

Organization — Top-level workspace that contains providers, policies, teams, usage data, and other shared configuration. See First Steps.

Organization manager — Organization role with administrative access to configuration and resources. Can manage resources, invite users, and change organization settings. See Organization roles.

Policy — Rule that defines when and how AI requests are evaluated, including conditions and evaluation settings. See Policies.

Project — Chat workspace used to group related conversations. See Manage Chats and Projects.

Provider — Configured connection to an AI provider, including endpoint settings, credentials, availability, and supported models. See Providers.

Routing rule — Rule that directs matching requests to one or more providers and models using conditions, priorities, weights, and optional fallbacks. See Routing Rules.

Team — Group within an organization used to organize users and apply access or configuration settings. See Config → Teams.

Trace — Record of an AI request lifecycle used for debugging and observability. See Analytics → Traces.

See also#

• Architecture Overview — Platform structure and request flow.
• Introduction — Product overview and main use cases.
• Admin Console Overview — Administrative navigation and setup order.