Appendix D. Capability notes

Full-length chapter. Content is checked against the 2026.10 mainline code and the live console; menus and features depend on the current license, account permissions, and delivered environment.

Back to overview · Full table of contents · Download full PDF

The table collects capabilities listed in each chapter that the current release does not provide, and the alternatives.

Capability	What to do instead	See
Auto-discover model backend instances from a registry or Kubernetes	Register multiple instances of the same inference service as multiple accounts of the same provider; the gateway assigns by concurrency and account status	Chapter 4
Pass non-standard protocol models through on arbitrary paths and count them as model usage	Standard model endpoints accept only standard paths; pass-through as-is uses 「API services」	Chapter 4, Chapter 11
Provider-level model-name redirect and system-prompt injection	Provided as traffic-rule actions, and they take effect by condition	Chapter 15
Scheduled upstream authenticity checks	Trigger verification manually, and keep watching passive traffic signals	Chapter 4
Capacity-planning tools, pre-install kernel parameters, and distro detection	Insufficient disk space blocks install and lists candidate mount points; size capacity from actual usage in the pilot	Chapter 26
gRPC proxy and transcoding	Use a standard model endpoint or 「API services」	Chapter 10
Batch async task APIs (`/batches`, `/files`)	Synchronous calls plus caller-side queuing; batch judgements use `/decisions`	Chapter 10
Semantic routing, intelligent load balancing, and multi-region nearest-node dispatch	Write rules with deterministic match conditions (headers, model name, caller, CIDR, time window)	Chapter 15
Percentage-based gradual rollout and one-click version rollback	Use a rollout cookie / request header for targeted canaries; roll back by disabling the rule or restoring the old configuration	Chapter 15, Chapter 24
mTLS mutual authentication from the gateway to the upstream	Authenticate on the upstream with an API key or custom headers	Chapter 4
Vulnerability rule-library (CNVD/CNNVD) subscription	Threat intelligence supports only malicious IP and domain lists; the tenant connects its own malicious sources	Chapter 14
Private-model fence	This capability is no longer provided; tighten with guardrail rules and available scope	Chapter 14
Automatic degrade after consecutive guardrail timeouts	Only a per-call allow / reject stance; there is no automatic degrade across requests	Chapter 14
Custom roles and menu-level permission sets	Roles are a fixed set; tighten authorization with functional groups + available scope	Chapter 3
Online agent session debug panel	Use operations fields in agent operations and multi-turn debugging in the trial bench	Chapter 12
MCP tool-change approval, per-tool MCP limits, and MCP user allow/deny lists	Narrow by MCP method and tool name in traffic rules, and control who can use the service with available scope	Chapter 11, Chapter 15
Plugin sandbox isolation and timeout circuit breaking	Plugins run as gateway extensions and follow the overall stance of the rules	Chapter 12
Billing-plan configuration, subscription, and plan usage	Control cost with quotas and budgets; attribute usage by model, model group, and tag	Chapter 17
Custom log-export format, sampling, and keyword alerts	Use the existing metrics on Operations overview and Alert rules	Chapter 22
Search failed sign-in records by account	「Authentication」 audit records show time and reason, not the attempted account	Chapter 22, Chapter 27

The table reflects the current delivered release. If a later release adds a capability, follow that release's user documentation and console.

Previous: Appendix C. Error codes and handling suggestions