tweaks

README.md

@@ -67,6 +67,7 @@ go build -o whiskr
 - Adjust model, temperature, prompt, or message role from the controls in the bottom-left
 - Use the model search field to quickly find models (supports fuzzy matching)
 - Look for tags in the model list to see if a model supports tools, vision, or reasoning
+- Use `![alt](url)` in your message to display an image inline. If the model supports vision, the same image URL is passed to the model for multimodal input.
 
 ## License
 

chat.go

@@ -7,6 +7,7 @@ import (
 	"fmt"
 	"io"
 	"net/http"
+	"regexp"
 	"strings"
 
 	"github.com/revrost/go-openrouter"
@@ -117,13 +118,26 @@ func (r *Request) Parse() (*openrouter.ChatCompletionRequest, error) {
 
 	for index, message := range r.Messages {
 		switch message.Role {
-		case "system", "user":
+		case "system":
 			request.Messages = append(request.Messages, openrouter.ChatCompletionMessage{
 				Role: message.Role,
 				Content: openrouter.Content{
 					Text: message.Text,
 				},
 			})
+		case "user":
+			var content openrouter.Content
+
+			if model.Vision && strings.Contains(message.Text, "![") {
+				content.Multi = SplitImagePairs(message.Text)
+			} else {
+				content.Text = message.Text
+			}
+
+			request.Messages = append(request.Messages, openrouter.ChatCompletionMessage{
+				Role:    message.Role,
+				Content: content,
+			})
 		case "assistant":
 			msg := openrouter.ChatCompletionMessage{
 				Role: openrouter.ChatMessageRoleAssistant,
@@ -151,7 +165,7 @@ func (r *Request) Parse() (*openrouter.ChatCompletionRequest, error) {
 }
 
 func HandleChat(w http.ResponseWriter, r *http.Request) {
-	debug("new chat")
+	debug("parsing chat")
 
 	var raw Request
 
@@ -174,6 +188,9 @@ func HandleChat(w http.ResponseWriter, r *http.Request) {
 
 	request.Stream = true
 
+	dump("debug.json", request)
+	debug("preparing stream")
+
 	response, err := NewStream(w)
 	if err != nil {
 		RespondJson(w, http.StatusBadRequest, map[string]any{
@@ -325,3 +342,74 @@ func RunCompletion(ctx context.Context, response *Stream, request *openrouter.Ch
 
 	return tool, result.String(), nil
 }
+
+func SplitImagePairs(text string) []openrouter.ChatMessagePart {
+	rgx := regexp.MustCompile(`(?m)!\[[^\]]*]\((\S+?)\)`)
+
+	var (
+		index int
+		parts []openrouter.ChatMessagePart
+	)
+
+	push := func(str, end int) {
+		rest := text[str:end]
+
+		if rest == "" {
+			return
+		}
+
+		total := len(parts)
+
+		if total > 0 && parts[total-1].Type == openrouter.ChatMessagePartTypeText {
+			parts[total-1].Text += rest
+
+			return
+		}
+
+		parts = append(parts, openrouter.ChatMessagePart{
+			Type: openrouter.ChatMessagePartTypeText,
+			Text: rest,
+		})
+	}
+
+	for {
+		location := rgx.FindStringSubmatchIndex(text[index:])
+		if location == nil {
+			push(index, len(text)-1)
+
+			break
+		}
+
+		start := index + location[0]
+		end := index + location[1]
+
+		urlStart := index + location[2]
+		urlEnd := index + location[3]
+
+		url := text[urlStart:urlEnd]
+
+		if !strings.HasPrefix(url, "https://") && !strings.HasPrefix(url, "http://") {
+			push(index, end)
+
+			index = end
+
+			continue
+		}
+
+		if start > index {
+			push(index, start)
+		}
+
+		parts = append(parts, openrouter.ChatMessagePart{
+			Type: openrouter.ChatMessagePartTypeImageURL,
+			ImageURL: &openrouter.ChatMessageImageURL{
+				Detail: openrouter.ImageURLDetailAuto,
+				URL:    url,
+			},
+		})
+
+		index = end
+	}
+
+	return parts
+}

debug.go

@@ -1,5 +1,19 @@
 package main
 
+import (
+	"encoding/json"
+	"os"
+)
+
+func dump(name string, val any) {
+	if !Debug {
+		return
+	}
+
+	b, _ := json.MarshalIndent(val, "", "\t")
+	os.WriteFile(name, b, 0644)
+}
+
 func debug(format string, args ...any) {
 	if !Debug {
 		return

models.go

@@ -15,6 +15,7 @@ type Model struct {
 	Tags        []string `json:"tags,omitempty"`
 
 	Reasoning bool `json:"-"`
+	Vision    bool `json:"-"`
 	JSON      bool `json:"-"`
 	Tools     bool `json:"-"`
 }
@@ -78,6 +79,8 @@ func GetModelTags(model openrouter.Model, m *Model) {
 
 	for _, modality := range model.Architecture.InputModalities {
 		if modality == "image" {
+			m.Vision = true
+
 			m.Tags = append(m.Tags, "vision")
 		}
 	}

static/js/markdown.js

@@ -10,7 +10,13 @@
 		walkTokens: (token) => {
 			const { type, lang, text } = token;
 
-			if (type !== "code") {
+			if (type === "html") {
+				token.text = token.text.replace(/&/g, "&")
+				token.text = token.text.replace(/</g, "&lt;")
+				token.text = token.text.replace(/>/g, "&gt;")
+
+				return;
+			} else if (type !== "code") {
 				return;
 			}