don't require pulling stubs for cloud models (#14574)

* don't require pulling stubs for cloud models This is a first in a series of PRs that will better integrate Ollama's cloud into the API and CLI. Previously we used to have a layer of indirection where you'd first have to pull a "stub" model that contains a reference to a cloud model. With this change, you don't have to pull first, you can just use a cloud model in various routes like `/api/chat` and `/api/show`. This change respects <https://github.com/ollama/ollama/pull/14221>, so if cloud is disabled, these models won't be accessible. There's also a new, simpler pass-through proxy that doesn't convert the requests ahead of hitting the cloud models, which they themselves already support various formats (e.g., `v1/chat/completions` or Open Responses, etc.). This will help prevent issues caused by double converting (e.g., `v1/chat/completions` converted to `api/chat` on the client, then calling cloud and converting back to a `v1/chat/completions` response instead of the cloud model handling the original `v1/chat/completions` request first). There's now a notion of "source tags", which can be mixed with existing tags. So instead of having different formats like`gpt-oss:20b-cloud` vs. `kimi-k2.5:cloud` (`-cloud` suffix vs. `:cloud`), you can now specify cloud by simply appending `:cloud`. This PR doesn't change model resolution yet, but sets us up to allow for things like omitting the non-source tag, which would make something like `ollama run gpt-oss:cloud` work the same way that `ollama run gpt-oss` already works today. More detailed changes: - Added a shared model selector parser in `types/modelselector`: - supports `:cloud` and `:local` - accepts source tags in any position - supports legacy `:<tag>-cloud` - rejects conflicting source tags - Integrated selector handling across server inference/show routes: - `GenerateHandler`, `ChatHandler`, `EmbedHandler`, `EmbeddingsHandler`, `ShowHandler` - Added explicit-cloud passthrough proxy for ollama.com: - same-endpoint forwarding for `/api/*`, `/v1/*`, and `/v1/messages` - normalizes `model` (and `name` for `/api/show`) before forwarding - forwards request headers except hop-by-hop/proxy-managed headers - uses bounded response-header timeout - handles auth failures in a friendly way - Preserved cloud-disable behavior (`OLLAMA_NO_CLOUD`) - Updated create flow to support `FROM ...:cloud` model sources (though this flow uses the legacy proxy still, supporting Modelfile overrides is more complicated with the direct proxy approach) - Updated CLI/TUI/config cloud detection to use shared selector logic - Updated CLI preflight behavior so explicit cloud requests do not auto-pull local stubs What's next? - Cloud discovery/listing and cache-backed `ollama ls` / `/api/tags` - Modelfile overlay support for virtual cloud models on OpenAI/Anthropic request families - Recommender/default-selection behavior for ambiguous model families - Fully remove the legacy flow Fixes: https://github.com/ollama/ollama/issues/13801 * consolidate pull logic into confirmAndPull helper pullIfNeeded and ShowOrPull shared identical confirm-and-pull logic. Extract confirmAndPull to eliminate the duplication. * skip local existence checks for cloud models ModelExists and the TUI's modelExists both check the local model list, which causes cloud models to appear missing. Return true early for explicit cloud models so the TUI displays them beside the integration name and skips re-prompting the model picker on relaunch. * support optionally pulling stubs for newly-style names We now normalize names like `<family>:<size>:cloud` into legacy-style names like `<family>:<size>-cloud` for pulling and deleting (this also supports stripping `:local`). Support for pulling cloud models is temporary, once we integrate properly into `/api/tags` we won't need this anymore. * Fix server alias syncing * Update cmd/cmd.go Co-authored-by: Parth Sareen <parth.sareen@ollama.com> * address comments * improve some naming --------- Co-authored-by: ParthSareen <parth.sareen@ollama.com>
2026-04-23 09:15:44 +02:00 · 2026-03-03 10:46:33 -08:00
parent ad16bffc7d
commit 8207e55ec7
23 changed files with 2849 additions and 114 deletions
--- a/cmd/config/claude.go
+++ b/cmd/config/claude.go
@@ -107,15 +107,12 @@ func (c *Claude) ConfigureAliases(ctx context.Context, model string, existingAli
 	}

 	if !force && aliases["primary"] != "" {
-		client, _ := api.ClientFromEnvironment()
-		if isCloudModel(ctx, client, aliases["primary"]) {
-			if isCloudModel(ctx, client, aliases["fast"]) {
-				return aliases, false, nil
-			}
-		} else {
-			delete(aliases, "fast")
+		if isCloudModelName(aliases["primary"]) {
+			aliases["fast"] = aliases["primary"]
 			return aliases, false, nil
 		}
+		delete(aliases, "fast")
+		return aliases, false, nil
 	}

 	items, existingModels, cloudModels, client, err := listModels(ctx)
@@ -139,10 +136,8 @@ func (c *Claude) ConfigureAliases(ctx context.Context, model string, existingAli
 		aliases["primary"] = primary
 	}

-	if isCloudModel(ctx, client, aliases["primary"]) {
-		if aliases["fast"] == "" || !isCloudModel(ctx, client, aliases["fast"]) {
-			aliases["fast"] = aliases["primary"]
-		}
+	if isCloudModelName(aliases["primary"]) {
+		aliases["fast"] = aliases["primary"]
 	} else {
 		delete(aliases, "fast")
 	}
--- a/cmd/config/config.go
+++ b/cmd/config/config.go
@@ -233,6 +233,9 @@ func ModelExists(ctx context.Context, name string) bool {
 	if name == "" {
 		return false
 	}
+	if isCloudModelName(name) {
+		return true
+	}
 	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return false
--- a/cmd/config/droid.go
+++ b/cmd/config/droid.go
@@ -10,7 +10,6 @@ import (
 	"path/filepath"
 	"slices"

-	"github.com/ollama/ollama/api"
 	"github.com/ollama/ollama/envconfig"
 )

@@ -125,13 +124,12 @@ func (d *Droid) Edit(models []string) error {
 	}

 	// Build new Ollama model entries with sequential indices (0, 1, 2, ...)
-	client, _ := api.ClientFromEnvironment()

 	var newModels []any
 	var defaultModelID string
 	for i, model := range models {
 		maxOutput := 64000
-		if isCloudModel(context.Background(), client, model) {
+		if isCloudModelName(model) {
 			if l, ok := lookupCloudModelLimit(model); ok {
 				maxOutput = l.Output
 			}
--- a/cmd/config/integrations.go
+++ b/cmd/config/integrations.go
@@ -14,6 +14,7 @@ import (

 	"github.com/ollama/ollama/api"
 	internalcloud "github.com/ollama/ollama/internal/cloud"
+	"github.com/ollama/ollama/internal/modelref"
 	"github.com/ollama/ollama/progress"
 	"github.com/spf13/cobra"
 )
@@ -324,12 +325,7 @@ func SelectModelWithSelector(ctx context.Context, selector SingleSelector) (stri

 	// If the selected model isn't installed, pull it first
 	if !existingModels[selected] {
-		if cloudModels[selected] {
-			// Cloud models only pull a small manifest; no confirmation needed
-			if err := pullModel(ctx, client, selected); err != nil {
-				return "", fmt.Errorf("failed to pull %s: %w", selected, err)
-			}
-		} else {
+		if !isCloudModelName(selected) {
 			msg := fmt.Sprintf("Download %s?", selected)
 			if ok, err := confirmPrompt(msg); err != nil {
 				return "", err
@@ -524,7 +520,7 @@ func selectModelsWithSelectors(ctx context.Context, name, current string, single

 	var toPull []string
 	for _, m := range selected {
-		if !existingModels[m] {
+		if !existingModels[m] && !isCloudModelName(m) {
 			toPull = append(toPull, m)
 		}
 	}
@@ -550,12 +546,28 @@ func selectModelsWithSelectors(ctx context.Context, name, current string, single
 	return selected, nil
 }

+// TODO(parthsareen): consolidate pull logic from call sites
 func pullIfNeeded(ctx context.Context, client *api.Client, existingModels map[string]bool, model string) error {
-	if existingModels[model] {
+	if isCloudModelName(model) || existingModels[model] {
 		return nil
 	}
-	msg := fmt.Sprintf("Download %s?", model)
-	if ok, err := confirmPrompt(msg); err != nil {
+	return confirmAndPull(ctx, client, model)
+}
+
+// TODO(parthsareen): pull this out to tui package
+// ShowOrPull checks if a model exists via client.Show and offers to pull it if not found.
+func ShowOrPull(ctx context.Context, client *api.Client, model string) error {
+	if _, err := client.Show(ctx, &api.ShowRequest{Model: model}); err == nil {
+		return nil
+	}
+	if isCloudModelName(model) {
+		return nil
+	}
+	return confirmAndPull(ctx, client, model)
+}
+
+func confirmAndPull(ctx context.Context, client *api.Client, model string) error {
+	if ok, err := confirmPrompt(fmt.Sprintf("Download %s?", model)); err != nil {
 		return err
 	} else if !ok {
 		return errCancelled
@@ -567,26 +579,6 @@ func pullIfNeeded(ctx context.Context, client *api.Client, existingModels map[st
 	return nil
 }

-// TODO(parthsareen): pull this out to tui package
-// ShowOrPull checks if a model exists via client.Show and offers to pull it if not found.
-func ShowOrPull(ctx context.Context, client *api.Client, model string) error {
-	if _, err := client.Show(ctx, &api.ShowRequest{Model: model}); err == nil {
-		return nil
-	}
-	// Cloud models only pull a small manifest; skip the download confirmation
-	// TODO(parthsareen): consolidate with cloud config changes
-	if strings.HasSuffix(model, "cloud") {
-		return pullModel(ctx, client, model)
-	}
-	if ok, err := confirmPrompt(fmt.Sprintf("Download %s?", model)); err != nil {
-		return err
-	} else if !ok {
-		return errCancelled
-	}
-	fmt.Fprintf(os.Stderr, "\n")
-	return pullModel(ctx, client, model)
-}
-
 func listModels(ctx context.Context) ([]ModelItem, map[string]bool, map[string]bool, *api.Client, error) {
 	client, err := api.ClientFromEnvironment()
 	if err != nil {
@@ -731,10 +723,8 @@ func syncAliases(ctx context.Context, client *api.Client, ac AliasConfigurer, na
 	}
 	aliases["primary"] = model

-	if isCloudModel(ctx, client, model) {
-		if aliases["fast"] == "" || !isCloudModel(ctx, client, aliases["fast"]) {
-			aliases["fast"] = model
-		}
+	if isCloudModelName(model) {
+		aliases["fast"] = model
 	} else {
 		delete(aliases, "fast")
 	}
@@ -1020,7 +1010,7 @@ Examples:
 				existingAliases = aliases

 				// Ensure cloud models are authenticated
-				if isCloudModel(cmd.Context(), client, model) {
+				if isCloudModelName(model) {
 					if err := ensureAuth(cmd.Context(), client, map[string]bool{model: true}, []string{model}); err != nil {
 						return err
 					}
@@ -1209,7 +1199,7 @@ func buildModelList(existing []modelInfo, preChecked []string, current string) (
 	// When user has no models, preserve recommended order.
 	notInstalled := make(map[string]bool)
 	for i := range items {
-		if !existingModels[items[i].Name] {
+		if !existingModels[items[i].Name] && !cloudModels[items[i].Name] {
 			notInstalled[items[i].Name] = true
 			var parts []string
 			if items[i].Description != "" {
@@ -1303,7 +1293,8 @@ func IsCloudModelDisabled(ctx context.Context, name string) bool {
 }

 func isCloudModelName(name string) bool {
-	return strings.HasSuffix(name, ":cloud") || strings.HasSuffix(name, "-cloud")
+	// TODO(drifkin): Replace this wrapper with inlining once things stabilize a bit
+	return modelref.HasExplicitCloudSource(name)
 }

 func filterCloudModels(existing []modelInfo) []modelInfo {
--- a/cmd/config/integrations_test.go
+++ b/cmd/config/integrations_test.go
@@ -426,8 +426,14 @@ func TestBuildModelList_NoExistingModels(t *testing.T) {
 	}

 	for _, item := range items {
-		if !strings.HasSuffix(item.Description, "(not downloaded)") {
-			t.Errorf("item %q should have description ending with '(not downloaded)', got %q", item.Name, item.Description)
+		if strings.HasSuffix(item.Name, ":cloud") {
+			if strings.HasSuffix(item.Description, "(not downloaded)") {
+				t.Errorf("cloud model %q should not have '(not downloaded)' suffix, got %q", item.Name, item.Description)
+			}
+		} else {
+			if !strings.HasSuffix(item.Description, "(not downloaded)") {
+				t.Errorf("item %q should have description ending with '(not downloaded)', got %q", item.Name, item.Description)
+			}
 		}
 	}
 }
@@ -492,10 +498,14 @@ func TestBuildModelList_ExistingRecommendedMarked(t *testing.T) {
 			if strings.HasSuffix(item.Description, "(not downloaded)") {
 				t.Errorf("installed recommended %q should not have '(not downloaded)' suffix, got %q", item.Name, item.Description)
 			}
-		case "minimax-m2.5:cloud", "kimi-k2.5:cloud", "qwen3:8b":
+		case "qwen3:8b":
 			if !strings.HasSuffix(item.Description, "(not downloaded)") {
 				t.Errorf("non-installed recommended %q should have '(not downloaded)' suffix, got %q", item.Name, item.Description)
 			}
+		case "minimax-m2.5:cloud", "kimi-k2.5:cloud":
+			if strings.HasSuffix(item.Description, "(not downloaded)") {
+				t.Errorf("cloud model %q should not have '(not downloaded)' suffix, got %q", item.Name, item.Description)
+			}
 		}
 	}
 }
@@ -536,7 +546,13 @@ func TestBuildModelList_HasRecommendedCloudModel_OnlyNonInstalledAtBottom(t *tes
 	}

 	for _, item := range items {
-		if !slices.Contains([]string{"kimi-k2.5:cloud", "llama3.2"}, item.Name) {
+		isCloud := strings.HasSuffix(item.Name, ":cloud")
+		isInstalled := slices.Contains([]string{"kimi-k2.5:cloud", "llama3.2"}, item.Name)
+		if isInstalled || isCloud {
+			if strings.HasSuffix(item.Description, "(not downloaded)") {
+				t.Errorf("installed or cloud model %q should not have '(not downloaded)' suffix, got %q", item.Name, item.Description)
+			}
+		} else {
 			if !strings.HasSuffix(item.Description, "(not downloaded)") {
 				t.Errorf("non-installed %q should have '(not downloaded)' suffix, got %q", item.Name, item.Description)
 			}
@@ -1000,8 +1016,8 @@ func TestShowOrPull_ModelNotFound_ConfirmNo_Cancelled(t *testing.T) {
 	}
 }

-func TestShowOrPull_CloudModel_SkipsConfirmation(t *testing.T) {
-	// Confirm prompt should NOT be called for cloud models
+func TestShowOrPull_CloudModel_DoesNotPull(t *testing.T) {
+	// Confirm prompt should NOT be called for explicit cloud models
 	oldHook := DefaultConfirmPrompt
 	DefaultConfirmPrompt = func(prompt string) (bool, error) {
 		t.Error("confirm prompt should not be called for cloud models")
@@ -1032,8 +1048,115 @@ func TestShowOrPull_CloudModel_SkipsConfirmation(t *testing.T) {
 	if err != nil {
 		t.Errorf("ShowOrPull should succeed for cloud model, got: %v", err)
 	}
-	if !pullCalled {
-		t.Error("expected pull to be called for cloud model without confirmation")
+	if pullCalled {
+		t.Error("expected pull not to be called for cloud model")
+	}
+}
+
+func TestShowOrPull_CloudLegacySuffix_DoesNotPull(t *testing.T) {
+	// Confirm prompt should NOT be called for explicit cloud models
+	oldHook := DefaultConfirmPrompt
+	DefaultConfirmPrompt = func(prompt string) (bool, error) {
+		t.Error("confirm prompt should not be called for cloud models")
+		return false, nil
+	}
+	defer func() { DefaultConfirmPrompt = oldHook }()
+
+	var pullCalled bool
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		switch r.URL.Path {
+		case "/api/show":
+			w.WriteHeader(http.StatusNotFound)
+			fmt.Fprintf(w, `{"error":"model not found"}`)
+		case "/api/pull":
+			pullCalled = true
+			w.WriteHeader(http.StatusOK)
+			fmt.Fprintf(w, `{"status":"success"}`)
+		default:
+			w.WriteHeader(http.StatusNotFound)
+		}
+	}))
+	defer srv.Close()
+
+	u, _ := url.Parse(srv.URL)
+	client := api.NewClient(u, srv.Client())
+
+	err := ShowOrPull(context.Background(), client, "gpt-oss:20b-cloud")
+	if err != nil {
+		t.Errorf("ShowOrPull should succeed for cloud model, got: %v", err)
+	}
+	if pullCalled {
+		t.Error("expected pull not to be called for cloud model")
+	}
+}
+
+func TestPullIfNeeded_CloudModel_DoesNotPull(t *testing.T) {
+	oldHook := DefaultConfirmPrompt
+	DefaultConfirmPrompt = func(prompt string) (bool, error) {
+		t.Error("confirm prompt should not be called for cloud models")
+		return false, nil
+	}
+	defer func() { DefaultConfirmPrompt = oldHook }()
+
+	err := pullIfNeeded(context.Background(), nil, map[string]bool{}, "glm-5:cloud")
+	if err != nil {
+		t.Fatalf("expected no error for cloud model, got %v", err)
+	}
+
+	err = pullIfNeeded(context.Background(), nil, map[string]bool{}, "gpt-oss:20b-cloud")
+	if err != nil {
+		t.Fatalf("expected no error for cloud model with legacy suffix, got %v", err)
+	}
+}
+
+func TestSelectModelsWithSelectors_CloudSelection_DoesNotPull(t *testing.T) {
+	var pullCalled bool
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		switch r.URL.Path {
+		case "/api/status":
+			w.WriteHeader(http.StatusNotFound)
+			fmt.Fprintf(w, `{"error":"not found"}`)
+		case "/api/tags":
+			w.WriteHeader(http.StatusOK)
+			fmt.Fprintf(w, `{"models":[]}`)
+		case "/api/me":
+			w.WriteHeader(http.StatusOK)
+			fmt.Fprintf(w, `{"name":"test-user"}`)
+		case "/api/pull":
+			pullCalled = true
+			w.WriteHeader(http.StatusOK)
+			fmt.Fprintf(w, `{"status":"success"}`)
+		default:
+			w.WriteHeader(http.StatusNotFound)
+			fmt.Fprintf(w, `{"error":"not found"}`)
+		}
+	}))
+	defer srv.Close()
+	t.Setenv("OLLAMA_HOST", srv.URL)
+
+	single := func(title string, items []ModelItem, current string) (string, error) {
+		for _, item := range items {
+			if item.Name == "glm-5:cloud" {
+				return item.Name, nil
+			}
+		}
+		t.Fatalf("expected glm-5:cloud in selector items, got %v", items)
+		return "", nil
+	}
+
+	multi := func(title string, items []ModelItem, preChecked []string) ([]string, error) {
+		return nil, fmt.Errorf("multi selector should not be called")
+	}
+
+	selected, err := selectModelsWithSelectors(context.Background(), "codex", "", single, multi)
+	if err != nil {
+		t.Fatalf("selectModelsWithSelectors returned error: %v", err)
+	}
+	if !slices.Equal(selected, []string{"glm-5:cloud"}) {
+		t.Fatalf("unexpected selected models: %v", selected)
+	}
+	if pullCalled {
+		t.Fatal("expected cloud selection to skip pull")
 	}
 }

--- a/cmd/config/opencode.go
+++ b/cmd/config/opencode.go
@@ -12,8 +12,8 @@ import (
 	"slices"
 	"strings"

-	"github.com/ollama/ollama/api"
 	"github.com/ollama/ollama/envconfig"
+	"github.com/ollama/ollama/internal/modelref"
 )

 // OpenCode implements Runner and Editor for OpenCode integration
@@ -26,13 +26,13 @@ type cloudModelLimit struct {
 }

 // lookupCloudModelLimit returns the token limits for a cloud model.
-// It tries the exact name first, then strips the ":cloud" suffix.
+// It tries the exact name first, then strips explicit cloud suffixes.
 func lookupCloudModelLimit(name string) (cloudModelLimit, bool) {
 	if l, ok := cloudModelLimits[name]; ok {
 		return l, true
 	}
-	base := strings.TrimSuffix(name, ":cloud")
-	if base != name {
+	base, stripped := modelref.StripCloudSourceTag(name)
+	if stripped {
 		if l, ok := cloudModelLimits[base]; ok {
 			return l, true
 		}
@@ -152,7 +152,6 @@ func (o *OpenCode) Edit(modelList []string) error {
 		}
 	}

-	client, _ := api.ClientFromEnvironment()

 	for _, model := range modelList {
 		if existing, ok := models[model].(map[string]any); ok {
@@ -163,7 +162,7 @@ func (o *OpenCode) Edit(modelList []string) error {
 					existing["name"] = strings.TrimSuffix(name, " [Ollama]")
 				}
 			}
-			if isCloudModel(context.Background(), client, model) {
+			if isCloudModelName(model) {
 				if l, ok := lookupCloudModelLimit(model); ok {
 					existing["limit"] = map[string]any{
 						"context": l.Context,
@@ -177,7 +176,7 @@ func (o *OpenCode) Edit(modelList []string) error {
 			"name":    model,
 			"_launch": true,
 		}
-		if isCloudModel(context.Background(), client, model) {
+		if isCloudModelName(model) {
 			if l, ok := lookupCloudModelLimit(model); ok {
 				entry["limit"] = map[string]any{
 					"context": l.Context,