deployment

Explicit bindings, validation and candidates.

Integration from other modules

Integration here means other Go packages in this repository. Go internal import rules restrict imports from outside the repository; HTTP clients use mounted gateway endpoints.

ValidateDeployment during startup and management writes, then revalidate current configuration with Execution/AllowsEndpoint/ResolveHit during execution. Specify model_info.transport and endpoint_types explicitly.

Entry point or resource	Usage and result
ValidateDeployment(model)	Reject invalid configuration
Execution(model)	Return an explicitly bound transport and bool
Candidates / DeploymentEndpoints	Filter candidates and describe endpoint bindings

Integration demo

Save the code as demo/main.go inside the repository and run go run ./demo from the repository root. Remove the demo directory after use.

package main

import (
	"fmt"
	"github.com/sunqirui1987/xhub/internal/config"
	_ "github.com/sunqirui1987/xhub/internal/provider/all"
	"github.com/sunqirui1987/xhub/internal/provider/deployment"
)

// main 演示本页 API 的最小调用；无参数和返回值，打印上述预期结果，依赖与副作用见本页说明。
func main() {
	model := config.ModelEntry{ModelName: "demo-model", LiteLLMParams: map[string]any{"model": "demo-upstream"}, ModelInfo: map[string]any{"transport": "bypass_openai_responses", "endpoint_types": []string{"bypass:openai-responses"}}}
	if err := deployment.ValidateDeployment(model); err != nil {
		panic(err)
	}
	transport, ok := deployment.Execution(model)
	fmt.Println(ok, transport.ID)
}

Prints true bypass_openai_responses. It checks protocol binding, not URLs, credentials, or connectivity.

Errors, state, and side effects

Do not infer execution from api_base/custom_llm_provider when declarations are missing. Capability compatibility still requires disabled-state, authorization, budget, and weight checks.

Source and verification

Code entry points: bindings.go, candidates.go, catalog_binding.go, execution.go, registry.go, validate.go

Unit tests live beside source: bindings_test.go, candidates_test.go, catalog_binding_test.go, endpoint_redesign_test.go, registry_test.go, validate_test.go

go test ./internal/provider/deployment/...

Backend regressions live in cmd/regression; browser flows in frontend/e2e. Offline tests use local HTTP upstreams and do not establish production credential or vendor availability.

Optional Responses web search

Deployment metadata web_search_support accepts inherit (default; preserve native Responses behavior), supported (requires native Responses transport), or unsupported. web_search_fallback accepts reject (default) or explicit drop_optional. EncodeDialogue applies the same policy for candidate selection, preview and execution. Only optional Responses web_search/web_search_preview declarations are removed, including Codex additional_tools. Required/tool-selected search and other unsupported capabilities remain rejected. This does not execute searches or infer support from model names. Responses expose X-XHub-Dropped-Capabilities: web_search; request diagnostics persist dropped_capabilities, and the log drawer displays a localized downgrade notice.