feat(llm): add Opper provider

docs/llm.md

@@ -172,6 +172,7 @@ A gateway preset is a built-in, ready-made custom provider for a well-known Open
 | `api-route` | `https://global.api-route.com/v1` | `gpt-6.1-sol` | `API_ROUTE_API_KEY` | `frontier` |
 | `cheaperinference` | `https://api.cheaperinference.com/v1` | `gpt-5.4-mini` | `CHEAPER_INFERENCE_API_KEY` | `frontier` |
 | `llm-buzz` | `https://api.llm.buzz/api/v1` | `openai/gpt-5.4-mini` | `LLM_BUZZ_API_KEY` | `frontier` |
+| `opper` | `https://api.opper.ai/v3/compat` | `gpt-6.1-terra` | `OPPER_API_KEY` | `frontier` |
 | `requesty` | `https://router.requesty.ai/v1` (EU: `https://router.eu.requesty.ai/v1`) | `openai/gpt-4o-mini` | `REQUESTY_API_KEY` | `frontier` |
 
 When `llm.provider` names `X`, gortex resolves it in this order: a true built-in provider `X` → your custom provider `X` (global `providers.json`, the opt-in repo-local one, or inline `llm.custom.X`) → the preset `X`. Your own entry always wins over a same-named preset, as a whole — no preset field leaks into it. `GORTEX_LLM_MODEL` / `GORTEX_LLM_EFFORT` apply on top of whichever entry is resolved.

internal/llm/presets.go

@@ -47,6 +47,16 @@ var gatewayPresets = map[string]CustomProvider{
 		SchemaMode:    "json_schema",
 		PromptProfile: "frontier",
 	},
+	// Opper (https://docs.opper.ai) is an EU-hosted OpenAI-compatible
+	// gateway at /v3/compat (no /v1); model ids are bare pool names, and a
+	// "provider/model" id such as "azure/gpt-5.5" pins one route.
+	"opper": {
+		BaseURL:       "https://api.opper.ai/v3/compat",
+		Model:         "gpt-6.1-terra",
+		APIKeyEnv:     "OPPER_API_KEY",
+		SchemaMode:    "json_schema",
+		PromptProfile: "frontier",
+	},
 	// Requesty (https://docs.requesty.ai) fronts many vendors behind one
 	// key; model ids carry a vendor prefix. Regional routers:
 	// https://router.eu.requesty.ai/v1 (and .us / .ap).

internal/llm/presets_test.go

@@ -75,6 +75,12 @@ func TestConfig_ActiveCustom_PresetResolution(t *testing.T) {
 		t.Errorf("llm-buzz preset: %+v ok=%v profile=%v", cp, ok, ProfileFor(buzz))
 	}
 
+	opper := Config{Provider: "opper"}
+	cp, ok = opper.ActiveCustom()
+	if !ok || cp.BaseURL != "https://api.opper.ai/v3/compat" || cp.Model != "gpt-6.1-terra" || cp.APIKeyEnv != "OPPER_API_KEY" || cp.SchemaMode != "json_schema" || ProfileFor(opper) != ProfileFrontier {
+		t.Errorf("opper preset: %+v ok=%v profile=%v", cp, ok, ProfileFor(opper))
+	}
+
 	// A user custom provider of the same name always wins, wholesale —
 	// no preset field leaks into it.
 	c.Custom = map[string]CustomProvider{"testgw": {BaseURL: "https://mine/v1", Model: "my-model"}}

internal/llm/provider/provider_custom_test.go

@@ -56,6 +56,40 @@ func TestNew_CustomProviderDispatches(t *testing.T) {
 	}
 }
 
+// TestNew_CustomBaseURLPathIsKept proves a base URL whose path is not
+// /v1 (Opper's /v3/compat) keeps that path when /chat/completions is
+// appended.
+func TestNew_CustomBaseURLPathIsKept(t *testing.T) {
+	var gotPath string
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		gotPath = r.URL.Path
+		_, _ = io.WriteString(w, `{"choices":[{"message":{"content":"ok"}}]}`)
+	}))
+	defer srv.Close()
+
+	t.Setenv("MY_GATEWAY_KEY", "secret")
+	cfg := llm.Config{
+		Provider: "mygw",
+		Custom: map[string]llm.CustomProvider{
+			"mygw": {BaseURL: srv.URL + "/v3/compat", Model: "some-model", APIKeyEnv: "MY_GATEWAY_KEY"},
+		},
+	}
+	p, err := New(cfg)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer p.Close()
+
+	if _, err := p.Complete(context.Background(), llm.CompletionRequest{
+		Messages: []llm.Message{{Role: llm.RoleUser, Content: "hi"}},
+	}); err != nil {
+		t.Fatal(err)
+	}
+	if gotPath != "/v3/compat/chat/completions" {
+		t.Errorf("path=%q want /v3/compat/chat/completions", gotPath)
+	}
+}
+
 func TestNew_CustomKeylessOmitsAuth(t *testing.T) {
 	var gotAuth string
 	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {