Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.
launch/opencode: add image modalities for vision models (#15922)
Eva H3af1a008e2dd
diff --git a/cmd/launch/opencode.go b/cmd/launch/opencode.goindex 18b04eb90..3fcaec902 100644--- a/cmd/launch/opencode.go+++ b/cmd/launch/opencode.go@@ -1,6 +1,7 @@package launchimport (+ "context""encoding/json""fmt""os"@@ -8,9 +9,12 @@ import ("path/filepath""runtime""slices"+ "time"+ "github.com/ollama/ollama/api""github.com/ollama/ollama/cmd/internal/fileutil""github.com/ollama/ollama/envconfig"+ "github.com/ollama/ollama/types/model")// OpenCode implements Runner and Editor for OpenCode integration.@@ -20,6 +24,8 @@ type OpenCode struct {configContent string // JSON config built by Edit, passed to Run via env var}+const openCodeModelShowTimeout = 2 * time.Second+func (o *OpenCode) String() string { return "OpenCode" }// findOpenCode returns the opencode binary path, checking PATH first then the@@ -176,6 +182,12 @@ func buildInlineConfig(primary string, models []string) (string, error) {if primary == "" || len(models) == 0 {return "", fmt.Errorf("buildInlineConfig: primary and models are required")}++ client, err := api.ClientFromEnvironment()+ if err != nil {+ client = nil+ }+config := map[string]any{"$schema": "https://opencode.ai/config.json","provider": map[string]any{@@ -185,7 +197,7 @@ func buildInlineConfig(primary string, models []string) (string, error) {"options": map[string]any{"baseURL": envconfig.Host().String() + "/v1",},- "models": buildModelEntries(models),+ "models": buildModelEntries(context.Background(), client, models),},},"model": "ollama/" + primary,@@ -228,21 +240,39 @@ func readModelJSONModels() []string {return models}-func buildModelEntries(modelList []string) map[string]any {+func buildModelEntries(ctx context.Context, client *api.Client, modelList []string) map[string]any {+ if client != nil {+ var cancel context.CancelFunc+ if _, hasDeadline := ctx.Deadline(); !hasDeadline {+ ctx, cancel = context.WithTimeout(ctx, openCodeModelShowTimeout)+ defer cancel()+ }+ }+models := make(map[string]any)- for _, model := range modelList {+ for _, modelID := range modelList {entry := map[string]any{- "name": model,+ "name": modelID,+ }+ if client != nil {+ if resp, err := client.Show(ctx, &api.ShowRequest{Model: modelID}); err == nil {+ if slices.Contains(resp.Capabilities, model.CapabilityVision) {+ entry["modalities"] = map[string]any{+ "input": []string{"text", "image"},+ "output": []string{"text"},+ }+ }+ }}- if isCloudModelName(model) {- if l, ok := lookupCloudModelLimit(model); ok {+ if isCloudModelName(modelID) {+ if l, ok := lookupCloudModelLimit(modelID); ok {entry["limit"] = map[string]any{"context": l.Context,"output": l.Output,}}}- models[model] = entry+ models[modelID] = entry}return models}diff --git a/cmd/launch/opencode_test.go b/cmd/launch/opencode_test.goindex 9d3b378a9..54d721e85 100644--- a/cmd/launch/opencode_test.go+++ b/cmd/launch/opencode_test.go@@ -1,12 +1,21 @@package launchimport (+ "context""encoding/json""fmt"+ "io"+ "net/http"+ "net/url""os""path/filepath""runtime"+ "strings"+ "sync""testing"+ "time"++ "github.com/ollama/ollama/api")func TestOpenCodeIntegration(t *testing.T) {@@ -157,6 +166,105 @@ func TestOpenCodeEdit(t *testing.T) {t.Errorf("local model should not have limit, got %v", entry["limit"])}})++ t.Run("vision model gets image input modalities", func(t *testing.T) {+ u, err := url.Parse("http://ollama.example")+ if err != nil {+ t.Fatalf("parse test URL: %v", err)+ }+ client := api.NewClient(u, &http.Client{Transport: roundTripFunc(func(req *http.Request) (*http.Response, error) {+ if req.URL.Path != "/api/show" {+ return &http.Response{+ StatusCode: http.StatusNotFound,+ Body: io.NopCloser(strings.NewReader("not found")),+ Header: make(http.Header),+ }, nil+ }++ var body struct {+ Model string `json:"model"`+ }+ if err := json.NewDecoder(req.Body).Decode(&body); err != nil {+ t.Fatalf("decode show request: %v", err)+ }+ if body.Model != "gemma4:26b" {+ t.Fatalf("show request model = %q, want gemma4:26b", body.Model)+ }++ return &http.Response{+ StatusCode: http.StatusOK,+ Body: io.NopCloser(strings.NewReader(`{"capabilities":["vision"],"model_info":{}}`)),+ Header: make(http.Header),+ }, nil+ })})++ models := buildModelEntries(context.Background(), client, []string{"gemma4:26b"})+ entry, _ := models["gemma4:26b"].(map[string]any)+ modalities, _ := entry["modalities"].(map[string]any)+ input, _ := modalities["input"].([]string)+ output, _ := modalities["output"].([]string)++ if len(input) != 2 || input[0] != "text" || input[1] != "image" {+ t.Fatalf("modalities.input = %v, want [text image]", input)+ }+ if len(output) != 1 || output[0] != "text" {+ t.Fatalf("modalities.output = %v, want [text]", output)+ }+ })+}++func TestBuildModelEntries(t *testing.T) {+ t.Run("defaults to model name when capabilities cannot be probed", func(t *testing.T) {+ models := buildModelEntries(context.Background(), nil, []string{"llama3.2"})+ entry, _ := models["llama3.2"].(map[string]any)+ if entry["name"] != "llama3.2" {+ t.Fatalf("name = %v, want llama3.2", entry["name"])+ }+ if _, ok := entry["modalities"]; ok {+ t.Fatalf("modalities should not be set without an API client, got %v", entry["modalities"])+ }+ })++ t.Run("uses one timeout budget across capability probes", func(t *testing.T) {+ u, err := url.Parse("http://ollama.example")+ if err != nil {+ t.Fatalf("parse test URL: %v", err)+ }++ var mu sync.Mutex+ waited := 0++ client := api.NewClient(u, &http.Client{Transport: roundTripFunc(func(req *http.Request) (*http.Response, error) {+ mu.Lock()+ if req.Context().Err() == nil {+ waited+++ }+ mu.Unlock()++ <-req.Context().Done()+ return nil, req.Context().Err()+ })})++ ctx, cancel := context.WithTimeout(context.Background(), 25*time.Millisecond)+ defer cancel()++ models := buildModelEntries(ctx, client, []string{"slow-1", "slow-2"})+ for _, modelID := range []string{"slow-1", "slow-2"} {+ entry, _ := models[modelID].(map[string]any)+ if entry["name"] != modelID {+ t.Fatalf("name for %q = %v, want %q", modelID, entry["name"], modelID)+ }+ if _, ok := entry["modalities"]; ok {+ t.Fatalf("modalities for %q should not be set after probe timeout, got %v", modelID, entry["modalities"])+ }+ }++ mu.Lock()+ defer mu.Unlock()+ if waited != 1 {+ t.Fatalf("expected shared timeout to block one probe, waited on %d probes", waited)+ }+ })}func TestOpenCodeModels_ReturnsNil(t *testing.T) {