View evaluation backend - AgentKit 文档

eval backend resolves the evaluation backend for the current credentials and refreshes the local cache. Later evaluation commands use this result to choose the Coze or TEA evaluation platform. For TEA, the result includes the workspace, project, account, and session_key coordinates required by evaluation requests.

The backend result is cached by identity and evaluation gateway for 7 days. To bypass the cache, pass --cache-refresh on the eval command group or on the top-level dataset command group.

eval backend

Flag / Argument	Description	Default
`--json`	Output raw JSON.	`false`

agentkit eval backend

agentkit eval backend --json

Evaluation Gateway Configuration

The default evaluation gateway uses agentkit.cn-beijing.volcengineapi.com, signing service agentkit, and region cn-beijing. Set these environment variables when you need to point evaluation commands at a test environment.

Environment variable	Description	Default
`AGENTKIT_EVAL_HOST`	Evaluation gateway host.	`agentkit.cn-beijing.volcengineapi.com`
`AGENTKIT_EVAL_SERVICE`	Service name used for request signing.	`agentkit`
`AGENTKIT_EVAL_REGION`	Region used for request signing.	`cn-beijing`
`AGENTKIT_TEA_ACCOUNT_ID`	Account ID forwarded to TEA evaluation; when omitted, the CLI resolves it from the active SSO session or STS identity.	Resolved automatically
`EXTRA_HEADER`	Additional request headers, separated with semicolons and formatted as `Name: Value`.	—

AGENTKIT_EVAL_HOST=agentkit-ppe.cn-beijing.volcengineapi.com \
AGENTKIT_EVAL_SERVICE=agentkit_ppe \
agentkit eval backend --json

Debug Options

The eval command group and top-level dataset command group support these debug options. They are scoped to evaluation backend caching and TEA request troubleshooting, and are not global options.

Flag / Argument	Description	Default
`--cache-refresh`	Ignore the local cache and resolve the evaluation backend again.	`false`
`--verbose`	Print full TEA requests to stderr; the `session_key` cookie is masked.	`false`

agentkit eval --cache-refresh backend
agentkit eval --verbose dataset list

Last modified on September 19, 2026