feat: localchat, a Go + Templ + HTMX chat app for local Gemma models

Streams replies over SSE from any OpenAI-compatible server (oMLX,
llama.cpp, Ollama, ...). Single binary that can install itself as an
OS service; docker compose bundles llama.cpp + Gemma 4 E2B.

internal/web/views/views.templ

@@ -0,0 +1,122 @@
+package views
+
+import "git.b0b.be/bdeb/localchat/internal/chat"
+
+// Page is the full chat screen.
+templ Page(title string, msgs []chat.Message) {
+	<!DOCTYPE html>
+	<html lang="en">
+		<head>
+			<meta charset="utf-8"/>
+			<meta name="viewport" content="width=device-width, initial-scale=1"/>
+			<title>{ title }</title>
+			<link rel="stylesheet" href="/static/app.css"/>
+			<link rel="icon" href="/static/favicon.svg" type="image/svg+xml"/>
+			<script src="/static/htmx.min.js" defer></script>
+			<script src="/static/sse.min.js" defer></script>
+			<script src="/static/app.js" defer></script>
+		</head>
+		<body>
+			<header>
+				<h1>{ title }</h1>
+				@Health(HealthInfo{Checking: true})
+				<form hx-post="/chat/reset" hx-target="#messages" hx-swap="innerHTML">
+					<button type="submit" class="ghost" title="Start a new conversation">New chat</button>
+				</form>
+			</header>
+			<main id="messages" aria-live="polite">
+				@Messages(msgs)
+			</main>
+			<form id="composer" hx-post="/chat" hx-target="#messages" hx-swap="beforeend" hx-disabled-elt="find button">
+				<textarea name="message" rows="1" placeholder="Ask anything… (Enter to send, Shift+Enter for a new line)" required autofocus></textarea>
+				<button type="submit">Send</button>
+			</form>
+		</body>
+	</html>
+}
+
+// Messages renders a whole conversation, or the empty state.
+templ Messages(msgs []chat.Message) {
+	if len(msgs) == 0 {
+		<div class="empty">
+			<p class="big">Hi! 👋</p>
+			<p>Everything you type stays on this machine. The model runs locally.</p>
+		</div>
+	}
+	for _, m := range msgs {
+		if m.Role == "assistant" && m.State != chat.Done {
+			@StreamingReply(m)
+		} else {
+			@MessageView(m)
+		}
+	}
+}
+
+// Exchange is returned after sending: the user's message plus a reply that
+// streams in over server-sent events.
+templ Exchange(user, reply chat.Message) {
+	@MessageView(user)
+	@StreamingReply(reply)
+}
+
+// MessageView renders a finished message.
+templ MessageView(m chat.Message) {
+	<div class={ "msg", m.Role } id={ "msg-" + m.ID }>
+		<div class="bubble">
+			if m.Role == "assistant" {
+				<div class="content markdown">
+					@Markdown(m.Content)
+				</div>
+				if m.Err != "" {
+					<p class="error">⚠️ { m.Err }</p>
+				}
+			} else {
+				<div class="content plain">{ m.Content }</div>
+			}
+		</div>
+	</div>
+}
+
+// StreamingReply opens an SSE connection. "token" events append text as it
+// arrives; the final "done" event swaps in the rendered Markdown and closes
+// the connection.
+templ StreamingReply(m chat.Message) {
+	<div
+		class="msg assistant"
+		id={ "msg-" + m.ID }
+		hx-ext="sse"
+		sse-connect={ "/chat/stream/" + m.ID }
+		sse-swap="done"
+		hx-swap="outerHTML"
+		sse-close="done"
+	>
+		<div class="bubble">
+			<div class="content plain streaming" sse-swap="token" hx-swap="beforeend"></div>
+		</div>
+	</div>
+}
+
+// Token is one streamed chunk of text.
+templ Token(text string) {
+	<span>{ text }</span>
+}
+
+// HealthInfo describes the model server status shown in the header.
+type HealthInfo struct {
+	Checking bool
+	OK       bool
+	Model    string
+	Err      string
+}
+
+templ Health(h HealthInfo) {
+	<div id="health" class="health" hx-get="/health" hx-trigger={ healthTrigger(h) } hx-swap="outerHTML">
+		if h.Checking {
+			<span class="dot"></span> connecting…
+		} else if h.OK {
+			<span class="dot ok"></span> { h.Model }
+		} else {
+			<span class="dot bad"></span> <span title={ h.Err }>model offline</span>
+		}
+	</div>
+}