From 019a5e6c80cfc53409e42f4125813f13504c9586 Mon Sep 17 00:00:00 2001
From: lcolok <425311101@qq.com>
Date: Wed, 4 Mar 2026 14:52:22 +0800
Subject: [PATCH] feat(anthropic): add vision/image support for native
 Anthropic provider

The merged #1020 added vision support to the openai_compat provider but
left the native Anthropic provider without image handling. This patch
closes the gap: when a Message carries Media (base64 data URLs), the
Anthropic adapter now builds NewImageBlockBase64 content blocks using
the official SDK, giving users of `anthropic/` prefixed models the
same vision capability.

Changes:
- buildParams: handle msg.Media in user messages, convert data URLs
  to anthropic.NewImageBlockBase64 content blocks
- parseDataURL: extract media type and base64 data from data URLs
- Tests for parseDataURL (8 cases) and media message building
---
 pkg/providers/anthropic/provider.go      | 31 +++++++++++
 pkg/providers/anthropic/provider_test.go | 67 ++++++++++++++++++++++++
 2 files changed, 98 insertions(+)

diff --git a/pkg/providers/anthropic/provider.go b/pkg/providers/anthropic/provider.go
index 1bb15f771..046fb4be4 100644
--- a/pkg/providers/anthropic/provider.go
+++ b/pkg/providers/anthropic/provider.go
@@ -132,6 +132,20 @@ func buildParams(
 				anthropicMessages = append(anthropicMessages,
 					anthropic.NewUserMessage(anthropic.NewToolResultBlock(msg.ToolCallID, msg.Content, false)),
 				)
+			} else if len(msg.Media) > 0 {
+				// Multimodal message with text + images
+				var blocks []anthropic.ContentBlockParamUnion
+				if msg.Content != "" {
+					blocks = append(blocks, anthropic.NewTextBlock(msg.Content))
+				}
+				for _, mediaURL := range msg.Media {
+					if mediaType, b64Data, ok := parseDataURL(mediaURL); ok {
+						blocks = append(blocks, anthropic.NewImageBlockBase64(mediaType, b64Data))
+					}
+				}
+				if len(blocks) > 0 {
+					anthropicMessages = append(anthropicMessages, anthropic.NewUserMessage(blocks...))
+				}
 			} else {
 				anthropicMessages = append(anthropicMessages,
 					anthropic.NewUserMessage(anthropic.NewTextBlock(msg.Content)),
@@ -257,6 +271,23 @@ func parseResponse(resp *anthropic.Message) *LLMResponse {
 	}
 }
 
+// parseDataURL extracts the media type and base64 data from a data URL.
+// Expected format: "data:<mediaType>;base64,<data>"
+func parseDataURL(url string) (mediaType string, b64Data string, ok bool) {
+	// data:image/jpeg;base64,/9j/4AAQ...
+	if !strings.HasPrefix(url, "data:") {
+		return "", "", false
+	}
+	rest := url[len("data:"):]
+	idx := strings.Index(rest, ";base64,")
+	if idx < 0 {
+		return "", "", false
+	}
+	mediaType = rest[:idx]
+	b64Data = rest[idx+len(";base64,"):]
+	return mediaType, b64Data, mediaType != "" && b64Data != ""
+}
+
 func normalizeBaseURL(apiBase string) string {
 	base := strings.TrimSpace(apiBase)
 	if base == "" {
diff --git a/pkg/providers/anthropic/provider_test.go b/pkg/providers/anthropic/provider_test.go
index 3d21c1d0b..06c5a6fd3 100644
--- a/pkg/providers/anthropic/provider_test.go
+++ b/pkg/providers/anthropic/provider_test.go
@@ -262,6 +262,73 @@ func TestProvider_ChatUsesTokenSource(t *testing.T) {
 	}
 }
 
+func TestParseDataURL(t *testing.T) {
+	tests := []struct {
+		name     string
+		url      string
+		wantType string
+		wantData string
+		wantOK   bool
+	}{
+		{"valid jpeg", "data:image/jpeg;base64,/9j/4AAQ", "image/jpeg", "/9j/4AAQ", true},
+		{"valid png", "data:image/png;base64,iVBOR", "image/png", "iVBOR", true},
+		{"valid gif", "data:image/gif;base64,R0lG", "image/gif", "R0lG", true},
+		{"valid webp", "data:image/webp;base64,UklG", "image/webp", "UklG", true},
+		{"no data prefix", "https://example.com/image.jpg", "", "", false},
+		{"no base64 marker", "data:image/jpeg,rawdata", "", "", false},
+		{"empty media type", "data:;base64,abc", "", "", false},
+		{"empty data", "data:image/jpeg;base64,", "", "", false},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			gotType, gotData, gotOK := parseDataURL(tt.url)
+			if gotOK != tt.wantOK {
+				t.Errorf("ok = %v, want %v", gotOK, tt.wantOK)
+				return
+			}
+			if gotOK {
+				if gotType != tt.wantType {
+					t.Errorf("type = %q, want %q", gotType, tt.wantType)
+				}
+				if gotData != tt.wantData {
+					t.Errorf("data = %q, want %q", gotData, tt.wantData)
+				}
+			}
+		})
+	}
+}
+
+func TestBuildParams_MediaMessage(t *testing.T) {
+	messages := []Message{
+		{
+			Role:    "user",
+			Content: "What's in this image?",
+			Media:   []string{"data:image/jpeg;base64,/9j/4AAQ"},
+		},
+	}
+	params, err := buildParams(messages, nil, "claude-sonnet-4.6", map[string]any{})
+	if err != nil {
+		t.Fatalf("buildParams() error: %v", err)
+	}
+	if len(params.Messages) != 1 {
+		t.Fatalf("len(Messages) = %d, want 1", len(params.Messages))
+	}
+}
+
+func TestBuildParams_MediaMessageTextOnly(t *testing.T) {
+	// Message without media should still work as text-only
+	messages := []Message{
+		{Role: "user", Content: "Hello"},
+	}
+	params, err := buildParams(messages, nil, "claude-sonnet-4.6", map[string]any{})
+	if err != nil {
+		t.Fatalf("buildParams() error: %v", err)
+	}
+	if len(params.Messages) != 1 {
+		t.Fatalf("len(Messages) = %d, want 1", len(params.Messages))
+	}
+}
+
 func createAnthropicTestClient(baseURL, token string) *anthropic.Client {
 	c := anthropic.NewClient(
 		anthropicoption.WithAuthToken(token),