AI service management
The SASE LLM proxy gateway routes agent traffic through a central gateway, where you manage provider access and token quotas in one place.
Prerequisites
Before you begin, make sure the following requirements are met:
-
You have activated Agent Office Security module. see Billing overview.
-
You have obtained an API key from the target LLM provider (OpenAI, Anthropic, or DashScope).
Step 1: Add an AI service
Add an AI service to register an LLM provider with SASE. After you add the service, SASE generates a proxy address that your agents can use to access the LLM through the SASE gateway.
-
Log on to the SASE console.
-
In the left-side navigation pane, choose Agent Office Security > AI Service Management.
-
On the AI Service List tab, click Add AI Service.
-
In the Add AI Service panel, configure the following parameters.
Parameter
Description
Service Name
The name of the AI service, used for display in the console.
LLM Provider
The LLM provider. Valid values: OpenAI, Claude, and DashScope. The provider you select determines the available protocols and the default base URL.
Terms of service
The API protocol.
BaseUrl
The base URL of the LLM provider API.
API-KEY
The API key of the LLM provider. You can add multiple keys. SASE rotates among the keys for load balancing.
Custom Proxy Prefix
The prefix for the SASE proxy domain. Update the base URL in your agent configuration to use this proxy domain instead of the original provider URL.
Model Configuration
A glob pattern that specifies which model IDs this service handles.
*matches zero or more characters and?matches a single character. Default value:*(matches all models).AI Security Guardrail
When enabled, this feature integrates with AI Security Guardrail to perform security inspection on model requests and responses.
-
Inspect Request: Sends model requests to AI Security Guardrail for inspection. From the Service drop-down list, select an inspection policy, which is the protection service configured in AI Security Guardrail.
-
Inspect Responses: Sends model responses to AI Security Guardrail for inspection. From the Service drop-down list, select an inspection policy, which is the protection service configured in AI Security Guardrail. In Streaming Response Chunk Size, you can specify the number of characters sent to AI Security Guardrail for each inspection.
-
-
Click OK.
-
In the AI Service List, click the sync icon in the Synchronize Applications column to synchronize the AI service to the Office Application. This registers the AI service with the SASE gateway so that agent traffic can be routed to the service through the private network.
-
Copy the Proxy Address and paste it into the base URL configuration of the agent on your device (for example, Claude Code or OpenClaw).
Step 2: Configure a tokens control policy
Create a tokens control policy to set per-user token quotas for AI services. You can specify which users and services are subject to token limits.
-
In the left-side navigation pane, choose Agent Office Security > AI Service Management.
-
On the Tokens Management tab, click Create Policy.
-
In the Create Tokens Control Policy panel, configure the following parameters.
Parameter
Description
Policy Status
Whether to enable the policy.
Policy Type
The type of quota policy. Currently, only the built-in Tokens Management policy type is supported.
Priority
The priority of the policy. A smaller value indicates a higher priority. Valid values: 1 to 100.
Applicable User
The users to which the policy applies. Valid values: Specific User Group: Select one or more user groups. All Users: The policy applies to all users.
Applied Services
The AI services to which the policy applies. These are the services added in Step 1: Add an AI service. Valid values: Specified Services: Select one or more AI services. All Services: The policy applies to all AI services.
Quota Limit per User
The maximum number of LLM tokens that each user can consume within one reset period.
Reset Period
The interval at which the token quota resets. Valid values: Daily: The quota resets at 00:00 every day. Monthly: The quota resets at 00:00 on the first day of each month.
-
Click OK.
Step 3: Configure a zero-trust access control policy
Configure a zero-trust access control policy to allow agent traffic to flow through the SASE proxy to the AI service. Without this policy, the proxy blocks all requests.
For more information, see Configure a zero trust policy.
Next steps
Test that your agent can successfully connect and interact with the AI service through the SASE proxy.