External Tools

Connect external AI tools to OptScale AI through a governed access layer. All supported integrations use the same OpenAI-compatible LLM proxy (llm_proxy/<organization_id>/v1) and an employee virtual key from AI Access, referred to as the API key throughout this guide.

Before you start, complete baseline setup in First Steps (organization, provider, etc). For Admin UI context, see Core Services — AI Access and Providers.

Integration	Guide
OpenCode	Connect OpenCode
Cursor	Connect Cursor

Get connection values from Chat

Use OptScale AI Chat to obtain the Base URL, API key, and Model name required for OpenCode or Cursor. The dialog displays the values expected by the LLM proxy for your current selection; Cursor requires one Base URL path adjustment described below.

Steps

1. Open Chat. For layout context, see Interface Overview — Chat UI.
2. In the top header, select the Provider and Model you plan to use in the external tool. Only providers and models available to your account appear in these lists.
3. Click Connect to external tools .
4. The Connect this model to external tools dialog appears. Use the copy control next to each field to retrieve the required values:

Field in Chat	Use in OpenCode / Cursor
Base URL	Base URL (OpenCode) or Override OpenAI Base URL (Cursor). For Cursor, replace the trailing /v1 with /cursor before saving.
API key	API key (OpenCode) or OpenAI API Key (Cursor)
Model name	Models

The dialog indicates that requests are routed through OptScale AI Gateway. Treat the API key as a password. Requests made with this key are attributed to your account.

If multiple providers expose the same model name, use the provider_proxy_name/model_name format described in Model identifiers.

Model identifiers

When OpenCode or Cursor asks for model names, use identifiers exposed by OptScale AI:

Format	Example	When to use
<model_name>	gpt-oss:120b	Single provider or unambiguous model name
<provider_proxy_name>/<model_name>	ollama/llama3.1:8b	Same model name on multiple providers

The provider proxy name helps distinguish between models with identical names exposed by different providers. Confirm available names on the provider detail page under Available models—see Provider detail page.

Shared limitations

Table 1: OptScale AI external tools — OpenCode and Cursor integration requirements and limits

Topic	Detail
Virtual key scope	Requests use providers assigned to the key on AI Access
Tool execution	Agent tools require models that support tool calls; not all models do
TLS	The gateway URL must be served over HTTPS with a trusted certificate
Monitoring	Traffic appears in Traces, Usage and Cost, and policy evaluation when configured

Related documentation

Topic	Where to read more
Chat UI and connection control	Interface Overview
Initial setup	First Steps
Virtual keys and AI Access	Core Services — AI Access
Provider catalog and models	Core Services — Providers
Policies and guardrails example	End-to-end example — Policies + Guardrails
Platform architecture	Architecture Overview