Merge pull request 'feat: display inference statistics in chat UI' (#45 ) from feature/issue-43-inference-stats into main

Merge pull request 'feat: display inference statistics in chat UI' (#45)
chore: mark issue #43 plan as COMPLETED
2026-03-13 14:50:35 +01:00 · 2026-03-13 14:46:47 +01:00 · 2026-03-13 14:46:35 +01:00 · 2026-03-13 14:44:12 +01:00 · 2026-03-12 23:14:28 +01:00 · 2026-03-12 23:13:33 +01:00
36 changed files with 1227 additions and 692 deletions
--- a/README.md
+++ b/README.md
@@ -1,3 +1,97 @@
 # llm-multiverse-ui

-Web frontend for the llm-multiverse orchestration system
+Web frontend for the **llm-multiverse** orchestration system. Built with SvelteKit 5, TypeScript, Tailwind CSS v4, and gRPC-Web (Connect).
+
+## Prerequisites
+
+- Node.js 20+
+- A running llm-multiverse gRPC backend (the UI proxies requests through Vite/your production reverse proxy)
+
+## Getting Started
+
+```bash
+# Install dependencies
+npm install
+
+# Generate protobuf TypeScript definitions (only needed after .proto changes)
+npm run generate
+
+# Start the dev server
+npm run dev
+```
+
+The app starts at `http://localhost:5173` by default.
+
+## Scripts
+
+| Command             | Description                                      |
+| ------------------- | ------------------------------------------------ |
+| `npm run dev`       | Start Vite dev server with HMR                   |
+| `npm run build`     | Production build                                 |
+| `npm run preview`   | Preview the production build locally             |
+| `npm run check`     | Run svelte-check for TypeScript/Svelte errors    |
+| `npm run lint`      | Run ESLint                                       |
+| `npm run format`    | Format code with Prettier                        |
+| `npm run generate`  | Regenerate protobuf TypeScript from `.proto` files |
+
+## Using the UI
+
+### Landing Page
+
+The root page (`/`) links to every section of the app.
+
+### Chat (`/chat`)
+
+The main interface for interacting with the orchestrator.
+
+- **Sessions** are listed in the left sidebar. Click **+ New Chat** to start a session, or select an existing one. On mobile, tap the hamburger menu to open the sidebar.
+- Type a message and press **Enter** (or the **Send** button) to send it. While the orchestrator is processing, a progress indicator and thinking section show the current state.
+- **Session Config** (gear icon / "Config" button) opens a right sidebar where you can adjust the override level, disable specific tools, grant permissions, and save/load presets.
+- If a request fails, an error banner appears with a **Retry** button.
+- Navigation links to Lineage, Memory, and Audit are in the header (hidden on small screens).
+
+### Agent Lineage (`/lineage`)
+
+Visualizes the agent spawn tree as an interactive SVG diagram.
+
+- Click a node to view agent details (ID, type, depth, children) in a side panel.
+- A legend at the bottom shows agent type colors.
+- Supports dark mode with theme-aware SVG colors.
+
+### Memory Candidates (`/memory`)
+
+Lists memory candidates captured during orchestration, grouped by session.
+
+- **Source filter** — filter by Tool Output, Model Knowledge, or Web.
+- **Confidence slider** — set a minimum confidence threshold.
+- Each card shows the candidate content, source badge, and a confidence bar.
+
+### Audit Log (`/audit`)
+
+Timeline view of orchestration events for each session.
+
+- Select a session from the dropdown, then optionally filter by event type (State Change, Tool, Error, Message).
+- Events are displayed chronologically with date separators and color-coded badges.
+
+### Theme
+
+Use the theme toggle (sun/moon icon in any header) to cycle between **System**, **Light**, and **Dark** modes. The preference is persisted in localStorage.
+
+## Project Structure
+
+```
+src/
+  lib/
+    components/      Svelte components (Backdrop, PageHeader, MessageInput, ...)
+    composables/     Reactive composables (useOrchestration)
+    proto/           Generated protobuf TypeScript
+    services/        gRPC client (orchestrator)
+    stores/          Svelte 5 reactive stores (sessions, audit, memory, theme, ...)
+    types/           TypeScript types and helpers (lineage, resultSource)
+    utils/           Shared utilities (date formatting, session config)
+  routes/
+    chat/            Chat page
+    lineage/         Agent lineage visualization
+    memory/          Memory candidates browser
+    audit/           Audit log timeline
+```
--- a/implementation-plans/_index.md
+++ b/implementation-plans/_index.md
@@ -22,3 +22,4 @@
 | #18   | Dark/light theme toggle                                | COMPLETED    | [issue-018.md](issue-018.md) |
 | #19   | Responsive layout and mobile support                   | COMPLETED    | [issue-019.md](issue-019.md) |
 | #20   | Error handling and connection status                    | COMPLETED    | [issue-020.md](issue-020.md) |
+| #43   | Display inference statistics in chat UI                 | COMPLETED    | [issue-043.md](issue-043.md) |
--- a/implementation-plans/issue-043.md
+++ b/implementation-plans/issue-043.md
@@ -0,0 +1,60 @@
+# Issue #43: Display Inference Statistics in Chat UI
+
+**Status:** COMPLETED
+**Issue:** [#43](https://git.shahondin1624.de/llm-multiverse/llm-multiverse-ui/issues/43)
+**Branch:** `feature/issue-43-inference-stats`
+
+## Overview
+
+Add a collapsible UI panel that displays LLM inference statistics (token counts, context window utilization, throughput) below assistant messages after orchestration completes.
+
+## Phases
+
+### Phase 1: Proto Types
+
+**Files:**
+- `proto/upstream/proto/llm_multiverse/v1/common.proto` — Add `InferenceStats` message
+- `proto/upstream/proto/llm_multiverse/v1/orchestrator.proto` — Add optional `inference_stats` field to `ProcessRequestResponse`
+
+**InferenceStats message fields:**
+- `prompt_tokens` (uint32) — tokens in the prompt
+- `completion_tokens` (uint32) — tokens generated
+- `total_tokens` (uint32) — sum of prompt + completion
+- `context_window_size` (uint32) — model's maximum context length
+- `tokens_per_second` (float) — generation throughput
+
+**Then regenerate types:** `npm run generate`
+
+### Phase 2: Orchestration State
+
+**Files:**
+- `src/lib/composables/useOrchestration.svelte.ts` — Extract `inferenceStats` from response, expose via store getter
+
+### Phase 3: InferenceStatsPanel Component
+
+**Files:**
+- `src/lib/components/InferenceStatsPanel.svelte` — New component
+
+**Design:**
+- Follow `<details>` pattern from FinalResult.svelte
+- Collapsed by default
+- Summary line shows key stat (e.g., total tokens + tokens/sec)
+- Expanded content shows all stats in a grid layout
+- Context utilization shown as a progress bar
+- Blue/indigo color scheme (neutral, info-like)
+- Full dark mode support
+
+### Phase 4: Chat Page Integration
+
+**Files:**
+- `src/routes/chat/+page.svelte` — Render `InferenceStatsPanel` after `FinalResult` when stats available
+
+## Acceptance Criteria
+
+- [x] InferenceStats proto message defined and TypeScript types generated
+- [x] InferenceStatsPanel displays all required metrics
+- [x] Panel is collapsible, collapsed by default
+- [x] Context utilization shows visual progress bar
+- [x] Integrates cleanly into chat page below assistant message
+- [x] Dark mode support
+- [x] Build, lint, typecheck pass
--- a/src/app.html
+++ b/src/app.html
@@ -5,7 +5,7 @@
 		<meta name="viewport" content="width=device-width, initial-scale=1" />
 		%sveltekit.head%
 	</head>
-	<body data-sveltekit-preload-data="hover">
+	<body data-sveltekit-preload-data="tap">
 		<div style="display: contents">%sveltekit.body%</div>
 	</body>
 </html>
--- a/src/lib/components/AuditTimeline.svelte
+++ b/src/lib/components/AuditTimeline.svelte
@@ -1,5 +1,6 @@
 <script lang="ts">
-	import type { AuditEvent } from '$lib/stores/audit.svelte';
+	import type { AuditEvent, AuditEventType } from '$lib/stores/audit.svelte';
+	import { formatShortDate } from '$lib/utils/date';

 	let { events }: { events: AuditEvent[] } = $props();

@@ -7,53 +8,33 @@
 		return date.toLocaleTimeString([], { hour: '2-digit', minute: '2-digit', second: '2-digit' });
 	}

-	function formatDate(date: Date): string {
-		return date.toLocaleDateString([], { month: 'short', day: 'numeric', year: 'numeric' });
+	const eventTypeConfig: Record<AuditEventType, { badge: string; dot: string; label: string }> = {
+		state_change: {
+			badge: 'bg-blue-100 dark:bg-blue-900/40 text-blue-700 dark:text-blue-300',
+			dot: 'bg-blue-500',
+			label: 'State Change'
+		},
+		tool_invocation: {
+			badge: 'bg-green-100 dark:bg-green-900/40 text-green-700 dark:text-green-300',
+			dot: 'bg-green-500',
+			label: 'Tool'
+		},
+		error: {
+			badge: 'bg-red-100 dark:bg-red-900/40 text-red-700 dark:text-red-300',
+			dot: 'bg-red-500',
+			label: 'Error'
+		},
+		message: {
+			badge: 'bg-gray-100 dark:bg-gray-800 text-gray-600 dark:text-gray-400',
+			dot: 'bg-gray-400',
+			label: 'Message'
 		}
+	};

-	function typeBadgeClasses(eventType: string): string {
-		switch (eventType) {
-			case 'state_change':
-				return 'bg-blue-100 dark:bg-blue-900/40 text-blue-700 dark:text-blue-300';
-			case 'tool_invocation':
-				return 'bg-green-100 dark:bg-green-900/40 text-green-700 dark:text-green-300';
-			case 'error':
-				return 'bg-red-100 dark:bg-red-900/40 text-red-700 dark:text-red-300';
-			case 'message':
-				return 'bg-gray-100 dark:bg-gray-800 text-gray-600 dark:text-gray-400';
-			default:
-				return 'bg-gray-100 dark:bg-gray-800 text-gray-600 dark:text-gray-400';
-		}
-	}
+	const defaultEventConfig = { badge: 'bg-gray-100 dark:bg-gray-800 text-gray-600 dark:text-gray-400', dot: 'bg-gray-400', label: 'Unknown' };

-	function typeDotClasses(eventType: string): string {
-		switch (eventType) {
-			case 'state_change':
-				return 'bg-blue-500';
-			case 'tool_invocation':
-				return 'bg-green-500';
-			case 'error':
-				return 'bg-red-500';
-			case 'message':
-				return 'bg-gray-400';
-			default:
-				return 'bg-gray-400';
-		}
-	}
-
-	function typeLabel(eventType: string): string {
-		switch (eventType) {
-			case 'state_change':
-				return 'State Change';
-			case 'tool_invocation':
-				return 'Tool';
-			case 'error':
-				return 'Error';
-			case 'message':
-				return 'Message';
-			default:
-				return eventType;
-		}
+	function getEventConfig(eventType: string) {
+		return eventTypeConfig[eventType as AuditEventType] ?? defaultEventConfig;
 	}
 </script>

@@ -70,18 +51,19 @@

 		<ol class="space-y-4">
 			{#each events as event, i (event.id)}
-				{@const showDate = i === 0 || formatDate(event.timestamp) !== formatDate(events[i - 1].timestamp)}
+				{@const showDate = i === 0 || formatShortDate(event.timestamp) !== formatShortDate(events[i - 1].timestamp)}
+				{@const config = getEventConfig(event.eventType)}

 				{#if showDate}
 					<li class="relative pl-10 pt-2">
-						<span class="text-xs font-medium text-gray-400 dark:text-gray-500">{formatDate(event.timestamp)}</span>
+						<span class="text-xs font-medium text-gray-400 dark:text-gray-500">{formatShortDate(event.timestamp)}</span>
 					</li>
 				{/if}

 				<li class="group relative flex items-start pl-10">
 					<!-- Dot on the timeline -->
 					<div
-						class="absolute left-1.5 top-1.5 h-3 w-3 rounded-full ring-2 ring-white dark:ring-gray-900 {typeDotClasses(event.eventType)}"
+						class="absolute left-1.5 top-1.5 h-3 w-3 rounded-full ring-2 ring-white dark:ring-gray-900 {config.dot}"
 					></div>

 					<!-- Event card -->
@@ -91,9 +73,9 @@
 						<div class="flex flex-wrap items-center gap-2">
 							<span class="text-xs text-gray-400 dark:text-gray-500">{formatTimestamp(event.timestamp)}</span>
 							<span
-								class="inline-flex rounded-full px-2 py-0.5 text-xs font-medium {typeBadgeClasses(event.eventType)}"
+								class="inline-flex rounded-full px-2 py-0.5 text-xs font-medium {config.badge}"
 							>
-								{typeLabel(event.eventType)}
+								{config.label}
 							</span>
 							{#if event.state}
 								<span class="rounded-md bg-blue-50 dark:bg-blue-900/30 px-1.5 py-0.5 text-xs font-mono text-blue-600 dark:text-blue-400">
--- a/src/lib/components/Backdrop.svelte
+++ b/src/lib/components/Backdrop.svelte
@@ -0,0 +1,12 @@
+<script lang="ts">
+	let { onClose }: { onClose: () => void } = $props();
+</script>
+
+<div
+	class="fixed inset-0 z-40 bg-black/50 md:hidden"
+	role="button"
+	tabindex="-1"
+	onclick={onClose}
+	onkeydown={(e) => { if (e.key === 'Escape') onClose(); }}
+	aria-label="Close overlay"
+></div>
--- a/src/lib/components/ConfigSidebar.svelte
+++ b/src/lib/components/ConfigSidebar.svelte
@@ -4,6 +4,8 @@
 	import { SessionConfigSchema } from '$lib/proto/llm_multiverse/v1/orchestrator_pb';
 	import { create } from '@bufbuild/protobuf';
 	import { presetStore } from '$lib/stores/presets.svelte';
+	import { isNonDefaultConfig } from '$lib/utils/sessionConfig';
+	import Backdrop from '$lib/components/Backdrop.svelte';

 	let {
 		config,
@@ -37,11 +39,7 @@
 		{ value: OverrideLevel.ALL, label: 'All', description: 'No enforcement' }
 	];

-	const isNonDefault = $derived(
-		config.overrideLevel !== OverrideLevel.NONE ||
-			config.disabledTools.length > 0 ||
-			config.grantedPermissions.length > 0
-	);
+	const isNonDefault = $derived(isNonDefaultConfig(config));

 	function setOverrideLevel(level: OverrideLevel) {
 		const updated = create(SessionConfigSchema, {
@@ -120,16 +118,8 @@
 	}
 </script>

-<!-- Mobile overlay backdrop -->
 {#if onClose}
-	<div
-		class="fixed inset-0 z-40 bg-black/50 md:hidden"
-		role="button"
-		tabindex="-1"
-		onclick={onClose}
-		onkeydown={(e) => { if (e.key === 'Escape') onClose?.(); }}
-		aria-label="Close config sidebar"
-	></div>
+	<Backdrop onClose={onClose} />
 {/if}

 <aside
--- a/src/lib/components/FinalResult.svelte
+++ b/src/lib/components/FinalResult.svelte
@@ -1,6 +1,7 @@
 <script lang="ts">
 	import type { SubagentResult } from '$lib/proto/llm_multiverse/v1/common_pb';
-	import { ResultStatus, ResultQuality, ResultSource } from '$lib/proto/llm_multiverse/v1/common_pb';
+	import { ResultStatus, ResultQuality, ArtifactType } from '$lib/proto/llm_multiverse/v1/common_pb';
+	import { resultSourceConfig } from '$lib/types/resultSource';

 	let { result }: { result: SubagentResult } = $props();

@@ -26,18 +27,34 @@
 		}
 	});

-	const sourceLabel = $derived.by(() => {
-		switch (result.source) {
-			case ResultSource.TOOL_OUTPUT: return 'Tool Output';
-			case ResultSource.MODEL_KNOWLEDGE: return 'Model Knowledge';
-			case ResultSource.WEB: return 'Web';
-			default: return '';
+	const sourceBadge = $derived(resultSourceConfig(result.source));
+
+	const LINE_COLLAPSE_THRESHOLD = 20;
+
+	function contentLineCount(content: string): number {
+		return content.split('\n').length;
+	}
+
+	const URL_RE = /(https?:\/\/[^\s<>"')\]]+)/g;
+
+	function linkify(text: string): string {
+		return text.replace(URL_RE, '<a href="$1" target="_blank" rel="noopener noreferrer" class="text-blue-600 dark:text-blue-400 underline hover:text-blue-800 dark:hover:text-blue-300 break-all">$1</a>');
+	}
+
+	let copiedIndex: number | null = $state(null);
+
+	async function copyToClipboard(content: string, index: number) {
+		await navigator.clipboard.writeText(content);
+		copiedIndex = index;
+		setTimeout(() => {
+			if (copiedIndex === index) copiedIndex = null;
+		}, 2000);
 	}
-	});
 </script>

-<div class="mx-4 mb-3 rounded-xl border {statusConfig.border} {statusConfig.bg} p-4">
-	<div class="mb-2 flex items-center gap-2">
+<details class="mx-4 mb-3 rounded-xl border {statusConfig.border} {statusConfig.bg}">
+	<summary class="flex cursor-pointer items-center gap-2 px-4 py-2.5 select-none">
+		<svg class="chevron h-4 w-4 shrink-0 text-gray-500 dark:text-gray-400 transition-transform" fill="none" viewBox="0 0 24 24" stroke-width="2" stroke="currentColor"><path stroke-linecap="round" stroke-linejoin="round" d="M8.25 4.5l7.5 7.5-7.5 7.5" /></svg>
 		<span class="rounded-full px-2.5 py-0.5 text-xs font-medium {statusConfig.bg} {statusConfig.text}">
 			{statusConfig.label}
 		</span>
@@ -46,15 +63,19 @@
 				{qualityLabel}
 			</span>
 		{/if}
-		{#if sourceLabel}
-			<span class="rounded-full bg-purple-100 dark:bg-purple-900/40 px-2.5 py-0.5 text-xs font-medium text-purple-800 dark:text-purple-300">
-				{sourceLabel}
+		{#if sourceBadge.label !== 'Unspecified'}
+			<span class="rounded-full {sourceBadge.bg} px-2.5 py-0.5 text-xs font-medium {sourceBadge.text}">
+				{sourceBadge.label}
 			</span>
 		{/if}
-	</div>
-
 		{#if result.summary}
-		<p class="text-sm {statusConfig.text}">{result.summary}</p>
+			<span class="ml-1 truncate text-xs {statusConfig.text}">{result.summary}</span>
+		{/if}
+	</summary>
+
+	<div class="max-h-96 overflow-y-auto border-t {statusConfig.border} px-4 pb-4 pt-3">
+		{#if result.summary}
+			<p class="text-sm {statusConfig.text}">{@html linkify(result.summary)}</p>
 		{/if}

 		{#if result.failureReason}
@@ -64,14 +85,98 @@
 		{#if result.artifacts.length > 0}
 			<div class="mt-3 border-t {statusConfig.border} pt-3">
 				<p class="mb-1.5 text-xs font-medium text-gray-600 dark:text-gray-400">Artifacts</p>
-			<ul class="space-y-1">
-				{#each result.artifacts as artifact (artifact)}
-					<li class="flex items-center gap-2 text-sm">
-						<span class="text-gray-400 dark:text-gray-500">&#128196;</span>
-						<span class="font-mono text-xs text-gray-700 dark:text-gray-300">{artifact}</span>
-					</li>
+				<div class="space-y-3">
+					{#each result.artifacts as artifact, i (artifact.label + i)}
+						{#if artifact.artifactType === ArtifactType.CODE}
+							<!-- Code artifact: syntax-highlighted block with filename header -->
+							<div class="rounded-lg border border-gray-200 dark:border-gray-700 overflow-hidden">
+								<div class="flex items-center gap-2 bg-gray-100 dark:bg-gray-800 px-3 py-1.5 text-xs">
+									<span class="text-blue-600 dark:text-blue-400">&#128196;</span>
+									<span class="font-mono font-medium text-gray-700 dark:text-gray-300">{artifact.label}</span>
+									<div class="ml-auto flex items-center gap-1.5">
+										{#if artifact.metadata?.language}
+											<span class="rounded bg-gray-200 dark:bg-gray-700 px-1.5 py-0.5 text-[10px] text-gray-500 dark:text-gray-400">{artifact.metadata.language}</span>
+										{/if}
+										<button
+											type="button"
+											onclick={() => copyToClipboard(artifact.content, i)}
+											class="rounded p-0.5 text-gray-400 hover:bg-gray-200 hover:text-gray-600 dark:text-gray-500 dark:hover:bg-gray-700 dark:hover:text-gray-300 transition-colors"
+											title="Copy code"
+										>
+											{#if copiedIndex === i}
+												<svg class="h-3.5 w-3.5 text-green-500" fill="none" viewBox="0 0 24 24" stroke-width="2" stroke="currentColor"><path stroke-linecap="round" stroke-linejoin="round" d="M4.5 12.75l6 6 9-13.5" /></svg>
+											{:else}
+												<svg class="h-3.5 w-3.5" fill="none" viewBox="0 0 24 24" stroke-width="2" stroke="currentColor"><path stroke-linecap="round" stroke-linejoin="round" d="M15.666 3.888A2.25 2.25 0 0 0 13.5 2.25h-3c-1.03 0-1.9.693-2.166 1.638m7.332 0c.055.194.084.4.084.612v0a.75.75 0 0 1-.75.75H9.75a.75.75 0 0 1-.75-.75v0c0-.212.03-.418.084-.612m7.332 0c.646.049 1.288.11 1.927.184 1.1.128 1.907 1.077 1.907 2.185V19.5a2.25 2.25 0 0 1-2.25 2.25H6.75A2.25 2.25 0 0 1 4.5 19.5V6.257c0-1.108.806-2.057 1.907-2.185a48.208 48.208 0 0 1 1.927-.184" /></svg>
+											{/if}
+										</button>
+									</div>
+								</div>
+								{#if contentLineCount(artifact.content) > LINE_COLLAPSE_THRESHOLD}
+									<details>
+										<summary class="cursor-pointer bg-gray-50 dark:bg-gray-900 px-3 py-1 text-xs text-gray-500 dark:text-gray-400 hover:text-gray-700 dark:hover:text-gray-300">
+											Show {contentLineCount(artifact.content)} lines
+										</summary>
+										<pre class="overflow-x-auto bg-gray-50 dark:bg-gray-900 p-3 text-xs leading-relaxed text-gray-800 dark:text-gray-200"><code class="language-{artifact.metadata?.language ?? ''}">{artifact.content}</code></pre>
+									</details>
+								{:else}
+									<pre class="overflow-x-auto bg-gray-50 dark:bg-gray-900 p-3 text-xs leading-relaxed text-gray-800 dark:text-gray-200"><code class="language-{artifact.metadata?.language ?? ''}">{artifact.content}</code></pre>
+								{/if}
+							</div>
+
+						{:else if artifact.artifactType === ArtifactType.COMMAND_OUTPUT}
+							<!-- Command output: terminal-style block -->
+							<div class="rounded-lg border border-gray-700 dark:border-gray-600 overflow-hidden">
+								<div class="flex items-center gap-2 bg-gray-800 dark:bg-gray-900 px-3 py-1.5 text-xs">
+									<span class="text-green-400">&#9654;</span>
+									<span class="font-mono text-gray-300">{artifact.label}</span>
+								</div>
+								{#if contentLineCount(artifact.content) > LINE_COLLAPSE_THRESHOLD}
+									<details>
+										<summary class="cursor-pointer bg-gray-900 dark:bg-black px-3 py-1 text-xs text-gray-400 hover:text-gray-200">
+											Show {contentLineCount(artifact.content)} lines
+										</summary>
+										<pre class="overflow-x-auto bg-gray-900 dark:bg-black p-3 font-mono text-xs leading-relaxed text-green-300 dark:text-green-400">{artifact.content}</pre>
+									</details>
+								{:else}
+									<pre class="overflow-x-auto bg-gray-900 dark:bg-black p-3 font-mono text-xs leading-relaxed text-green-300 dark:text-green-400">{artifact.content}</pre>
+								{/if}
+							</div>
+
+						{:else if artifact.artifactType === ArtifactType.SEARCH_RESULT}
+							<!-- Search result: card with query title -->
+							<div class="rounded-lg border border-purple-200 dark:border-purple-800 bg-purple-50 dark:bg-purple-900/20 p-3">
+								<div class="mb-1 flex items-center gap-2 text-xs">
+									<span class="text-purple-500">&#128269;</span>
+									<span class="font-medium text-purple-700 dark:text-purple-300">{artifact.label}</span>
+								</div>
+								{#if contentLineCount(artifact.content) > LINE_COLLAPSE_THRESHOLD}
+									<details>
+										<summary class="cursor-pointer text-xs text-purple-500 dark:text-purple-400 hover:text-purple-700 dark:hover:text-purple-300">
+											Show full results
+										</summary>
+										<p class="mt-1 whitespace-pre-wrap text-xs text-gray-700 dark:text-gray-300">{@html linkify(artifact.content)}</p>
+									</details>
+								{:else}
+									<p class="whitespace-pre-wrap text-xs text-gray-700 dark:text-gray-300">{@html linkify(artifact.content)}</p>
+								{/if}
+							</div>
+
+						{:else}
+							<!-- Text / finding artifact: plain text -->
+							<div class="flex items-start gap-2 text-sm">
+								<span class="mt-0.5 text-gray-400 dark:text-gray-500">&#128196;</span>
+								<span class="text-xs text-gray-700 dark:text-gray-300">{@html linkify(artifact.content)}</span>
+							</div>
+						{/if}
 					{/each}
-			</ul>
+				</div>
 			</div>
 		{/if}
 	</div>
+</details>
+
+<style>
+	details[open] > summary .chevron {
+		transform: rotate(90deg);
+	}
+</style>
--- a/src/lib/components/InferenceStatsPanel.svelte
+++ b/src/lib/components/InferenceStatsPanel.svelte
@@ -0,0 +1,94 @@
+<script lang="ts">
+	import type { InferenceStats } from '$lib/proto/llm_multiverse/v1/common_pb';
+
+	let { stats }: { stats: InferenceStats } = $props();
+
+	const utilizationPct = $derived(
+		stats.contextWindowSize > 0
+			? Math.min(100, (stats.totalTokens / stats.contextWindowSize) * 100)
+			: 0
+	);
+
+	const utilizationColor = $derived.by(() => {
+		if (utilizationPct >= 90) return { bar: 'bg-red-500', text: 'text-red-700 dark:text-red-400' };
+		if (utilizationPct >= 70) return { bar: 'bg-amber-500', text: 'text-amber-700 dark:text-amber-400' };
+		return { bar: 'bg-blue-500', text: 'text-blue-700 dark:text-blue-400' };
+	});
+
+	function formatNumber(n: number): string {
+		return n.toLocaleString();
+	}
+</script>
+
+<details class="mx-4 mb-3 rounded-xl border border-indigo-200 dark:border-indigo-800 bg-indigo-50 dark:bg-indigo-900/20">
+	<summary class="flex cursor-pointer items-center gap-2 px-4 py-2.5 select-none">
+		<svg class="chevron h-4 w-4 shrink-0 text-gray-500 dark:text-gray-400 transition-transform" fill="none" viewBox="0 0 24 24" stroke-width="2" stroke="currentColor"><path stroke-linecap="round" stroke-linejoin="round" d="M8.25 4.5l7.5 7.5-7.5 7.5" /></svg>
+		<span class="rounded-full bg-indigo-100 dark:bg-indigo-900/40 px-2.5 py-0.5 text-xs font-medium text-indigo-800 dark:text-indigo-300">
+			Stats
+		</span>
+		<span class="truncate text-xs text-indigo-700 dark:text-indigo-400">
+			{formatNumber(stats.totalTokens)} tokens
+			{#if stats.tokensPerSecond > 0}
+				&middot; {stats.tokensPerSecond.toFixed(1)} tok/s
+			{/if}
+			{#if stats.contextWindowSize > 0}
+				&middot; {utilizationPct.toFixed(0)}% context
+			{/if}
+		</span>
+	</summary>
+
+	<div class="border-t border-indigo-200 dark:border-indigo-800 px-4 pb-4 pt-3">
+		<div class="grid grid-cols-2 gap-x-6 gap-y-3 sm:grid-cols-3">
+			<div>
+				<p class="text-[10px] font-medium uppercase tracking-wider text-gray-500 dark:text-gray-400">Prompt</p>
+				<p class="text-sm font-semibold text-gray-900 dark:text-gray-100">{formatNumber(stats.promptTokens)}</p>
+			</div>
+			<div>
+				<p class="text-[10px] font-medium uppercase tracking-wider text-gray-500 dark:text-gray-400">Completion</p>
+				<p class="text-sm font-semibold text-gray-900 dark:text-gray-100">{formatNumber(stats.completionTokens)}</p>
+			</div>
+			<div>
+				<p class="text-[10px] font-medium uppercase tracking-wider text-gray-500 dark:text-gray-400">Total</p>
+				<p class="text-sm font-semibold text-gray-900 dark:text-gray-100">{formatNumber(stats.totalTokens)}</p>
+			</div>
+			{#if stats.tokensPerSecond > 0}
+				<div>
+					<p class="text-[10px] font-medium uppercase tracking-wider text-gray-500 dark:text-gray-400">Throughput</p>
+					<p class="text-sm font-semibold text-gray-900 dark:text-gray-100">{stats.tokensPerSecond.toFixed(1)} tok/s</p>
+				</div>
+			{/if}
+			{#if stats.contextWindowSize > 0}
+				<div class="col-span-2">
+					<p class="text-[10px] font-medium uppercase tracking-wider text-gray-500 dark:text-gray-400">Context Window</p>
+					<p class="text-sm font-semibold text-gray-900 dark:text-gray-100">{formatNumber(stats.contextWindowSize)}</p>
+				</div>
+			{/if}
+		</div>
+
+		{#if stats.contextWindowSize > 0}
+			<div class="mt-3 border-t border-indigo-200 dark:border-indigo-800 pt-3">
+				<div class="flex items-center justify-between text-xs">
+					<span class="font-medium text-gray-600 dark:text-gray-400">Context Utilization</span>
+					<span class="font-semibold {utilizationColor.text}">{utilizationPct.toFixed(1)}%</span>
+				</div>
+				<div class="mt-1.5 h-2 w-full overflow-hidden rounded-full bg-gray-200 dark:bg-gray-700">
+					<div
+						class="h-full rounded-full transition-all duration-300 {utilizationColor.bar}"
+						style="width: {utilizationPct}%"
+						role="progressbar"
+						aria-valuenow={utilizationPct}
+						aria-valuemin={0}
+						aria-valuemax={100}
+						aria-label="Context window utilization"
+					></div>
+				</div>
+			</div>
+		{/if}
+	</div>
+</details>
+
+<style>
+	details[open] > summary .chevron {
+		transform: rotate(90deg);
+	}
+</style>
--- a/src/lib/components/LineageTree.svelte
+++ b/src/lib/components/LineageTree.svelte
@@ -40,12 +40,23 @@
 	}

 	/**
-	 * Recursively counts the total number of leaf nodes in a subtree.
-	 * A leaf is a node with no children; it counts as 1 vertical slot.
+	 * Computes leaf counts for every node in a single bottom-up pass.
+	 * Stores results in a map keyed by node identity.
 	 */
-	function countLeaves(node: LineageNode): number {
-		if (node.children.length === 0) return 1;
-		return node.children.reduce((sum, child) => sum + countLeaves(child), 0);
+	function computeLeafCounts(roots: LineageNode[]): Map<LineageNode, number> {
+		// eslint-disable-next-line svelte/prefer-svelte-reactivity -- internal computation map, not reactive state
+		const counts = new Map<LineageNode, number>();
+		function walk(node: LineageNode): number {
+			if (node.children.length === 0) {
+				counts.set(node, 1);
+				return 1;
+			}
+			const total = node.children.reduce((sum, child) => sum + walk(child), 0);
+			counts.set(node, total);
+			return total;
+		}
+		for (const root of roots) walk(root);
+		return counts;
 	}

 	/**
@@ -57,7 +68,8 @@
 		totalWidth: number;
 		totalHeight: number;
 	} {
-		const totalLeaves = roots.reduce((sum, r) => sum + countLeaves(r), 0);
+		const leafCounts = computeLeafCounts(roots);
+		const totalLeaves = roots.reduce((sum, r) => sum + (leafCounts.get(r) ?? 1), 0);
 		const totalHeight = totalLeaves * (NODE_HEIGHT + VERTICAL_GAP) - VERTICAL_GAP;

 		function positionNode(
@@ -73,7 +85,7 @@
 				return { node, x, y, children: [] };
 			}

-			const childLeafCounts = node.children.map(countLeaves);
+			const childLeafCounts = node.children.map((c) => leafCounts.get(c) ?? 1);
 			const totalChildLeaves = childLeafCounts.reduce((a, b) => a + b, 0);

 			let currentY = startY;
@@ -98,7 +110,7 @@
 		let currentY = 0;
 		const positioned: PositionedNode[] = [];
 		for (const root of roots) {
-			const rootLeaves = countLeaves(root);
+			const rootLeaves = leafCounts.get(root) ?? 1;
 			const rootHeight = (rootLeaves / totalLeaves) * totalHeight;
 			positioned.push(positionNode(root, 0, currentY, rootHeight));
 			currentY += rootHeight;
@@ -222,6 +234,7 @@
 				<!-- Nodes -->
 				{#each allNodes as pn (pn.node.id)}
 					{@const colors = agentTypeColor(pn.node.agentType)}
+					{@const colorSet = themeStore.isDark ? colors.dark : colors.light}
 					{@const isSelected = selectedNodeId === pn.node.id}

 					<g
@@ -241,8 +254,8 @@
 							height={NODE_HEIGHT}
 							rx="8"
 							ry="8"
-							fill={colors.fill}
-							stroke={isSelected ? colors.stroke : defaultStroke}
+							fill={colorSet.fill}
+							stroke={isSelected ? colorSet.stroke : defaultStroke}
 							stroke-width={isSelected ? 2.5 : 1.5}
 						/>

@@ -251,7 +264,7 @@
 							x={NODE_WIDTH / 2}
 							y="22"
 							text-anchor="middle"
-							fill={colors.text}
+							fill={colorSet.text}
 							font-size="12"
 							font-weight="600"
 						>
--- a/src/lib/components/MemoryCandidateCard.svelte
+++ b/src/lib/components/MemoryCandidateCard.svelte
@@ -1,21 +1,10 @@
 <script lang="ts">
-	import { ResultSource } from '$lib/proto/llm_multiverse/v1/common_pb';
 	import type { StoredMemoryCandidate } from '$lib/stores/memory.svelte';
+	import { resultSourceConfig } from '$lib/types/resultSource';

 	let { candidate }: { candidate: StoredMemoryCandidate } = $props();

-	const sourceBadge = $derived.by(() => {
-		switch (candidate.source) {
-			case ResultSource.TOOL_OUTPUT:
-				return { label: 'Tool Output', bg: 'bg-blue-100 dark:bg-blue-900/40', text: 'text-blue-800 dark:text-blue-300' };
-			case ResultSource.MODEL_KNOWLEDGE:
-				return { label: 'Model Knowledge', bg: 'bg-purple-100 dark:bg-purple-900/40', text: 'text-purple-800 dark:text-purple-300' };
-			case ResultSource.WEB:
-				return { label: 'Web', bg: 'bg-green-100 dark:bg-green-900/40', text: 'text-green-800 dark:text-green-300' };
-			default:
-				return { label: 'Unspecified', bg: 'bg-gray-100 dark:bg-gray-800', text: 'text-gray-800 dark:text-gray-300' };
-		}
-	});
+	const sourceBadge = $derived(resultSourceConfig(candidate.source));

 	const confidencePct = $derived(Math.round(candidate.confidence * 100));

--- a/src/lib/components/MessageInput.svelte
+++ b/src/lib/components/MessageInput.svelte
@@ -36,7 +36,7 @@
 </script>

 <div class="border-t border-gray-200 dark:border-gray-700 bg-white dark:bg-gray-900 p-3 md:p-4">
-	<form onsubmit={handleSubmit} class="flex items-end gap-2">
+	<form onsubmit={(e) => { e.preventDefault(); handleSubmit(); }} class="flex items-end gap-2">
 		<textarea
 			bind:this={textarea}
 			bind:value={input}
@@ -51,8 +51,7 @@
 				placeholder:text-gray-400 dark:placeholder:text-gray-500"
 		></textarea>
 		<button
-			type="button"
-			onclick={handleSubmit}
+			type="submit"
 			disabled={disabled || !input.trim()}
 			class="min-h-[44px] min-w-[44px] rounded-xl bg-blue-600 px-4 py-2.5 text-sm font-medium text-white
 				hover:bg-blue-700 focus:ring-2 focus:ring-blue-500 focus:ring-offset-2 focus:outline-none
--- a/src/lib/components/OrchestrationProgress.svelte
+++ b/src/lib/components/OrchestrationProgress.svelte
@@ -29,7 +29,7 @@
 						{status === 'completed'
 						? 'bg-green-500 text-white'
 						: status === 'active'
-							? 'bg-blue-500 text-white ring-2 ring-blue-300 dark:ring-blue-600 ring-offset-1 dark:ring-offset-gray-800'
+							? 'bg-blue-500 text-white ring-2 ring-blue-300 dark:ring-blue-600 ring-offset-1 dark:ring-offset-gray-800 animate-pulse-ring'
 							: 'bg-gray-200 dark:bg-gray-600 text-gray-500 dark:text-gray-400'}"
 				>
 					{#if status === 'completed'}
@@ -46,11 +46,45 @@
 				</span>
 			</div>
 			{#if i < phases.length - 1}
-				<div
-					class="mb-5 h-0.5 flex-1 transition-colors duration-300
-						{getStatus(phases[i + 1].state) !== 'pending' ? 'bg-green-500' : 'bg-gray-200 dark:bg-gray-600'}"
-				></div>
+				{@const nextStatus = getStatus(phases[i + 1].state)}
+				<div class="relative mb-5 h-0.5 flex-1 bg-gray-200 dark:bg-gray-600 overflow-hidden">
+					{#if nextStatus !== 'pending'}
+						<!-- Completed: solid green fill -->
+						<div class="absolute inset-0 bg-green-500"></div>
+					{:else if status === 'active'}
+						<!-- Active: animated fill shimmer -->
+						<div class="absolute inset-0 animate-progress-fill bg-gradient-to-r from-blue-500 via-blue-400 to-blue-500 bg-[length:200%_100%]"></div>
+					{/if}
+				</div>
 			{/if}
 		{/each}
 	</div>
 </div>
+
+<style>
+	@keyframes pulse-ring {
+		0%, 100% {
+			box-shadow: 0 0 0 0 rgba(59, 130, 246, 0.5);
+		}
+		50% {
+			box-shadow: 0 0 0 6px rgba(59, 130, 246, 0);
+		}
+	}
+
+	@keyframes progress-fill {
+		0% {
+			background-position: 200% 0;
+		}
+		100% {
+			background-position: -200% 0;
+		}
+	}
+
+	:global(.animate-pulse-ring) {
+		animation: pulse-ring 2s ease-in-out infinite;
+	}
+
+	:global(.animate-progress-fill) {
+		animation: progress-fill 2s linear infinite;
+	}
+</style>
--- a/src/lib/components/PageHeader.svelte
+++ b/src/lib/components/PageHeader.svelte
@@ -0,0 +1,38 @@
+<script lang="ts">
+	import type { Snippet } from 'svelte';
+	import ThemeToggle from '$lib/components/ThemeToggle.svelte';
+
+	let {
+		title,
+		backHref,
+		backLabel = 'Chat',
+		children
+	}: {
+		title: string;
+		backHref?: string;
+		backLabel?: string;
+		children?: Snippet;
+	} = $props();
+</script>
+
+<header class="flex items-center justify-between border-b border-gray-200 dark:border-gray-700 bg-white dark:bg-gray-900 px-4 py-3 md:px-6">
+	<div class="flex items-center gap-2 md:gap-4">
+		{#if backHref}
+			<!-- eslint-disable svelte/no-navigation-without-resolve -- resolveRoute is resolve; plugin does not recognize the alias -->
+			<a
+				href={backHref}
+				class="rounded-lg px-2.5 py-1.5 text-sm text-gray-600 dark:text-gray-400 hover:bg-gray-100 dark:hover:bg-gray-700 hover:text-gray-900 dark:hover:text-gray-100"
+			>
+				&larr; {backLabel}
+			</a>
+			<!-- eslint-enable svelte/no-navigation-without-resolve -->
+		{/if}
+		<h1 class="text-base md:text-lg font-semibold text-gray-900 dark:text-gray-100">{title}</h1>
+	</div>
+	<div class="flex items-center gap-2">
+		{#if children}
+			{@render children()}
+		{/if}
+		<ThemeToggle />
+	</div>
+</header>
--- a/src/lib/components/SessionSidebar.svelte
+++ b/src/lib/components/SessionSidebar.svelte
@@ -1,5 +1,7 @@
 <script lang="ts">
 	import { sessionStore } from '$lib/stores/sessions.svelte';
+	import { formatRelativeDate } from '$lib/utils/date';
+	import Backdrop from '$lib/components/Backdrop.svelte';

 	let {
 		onSelectSession,
@@ -41,27 +43,11 @@
 		onClose?.();
 	}

-	function formatDate(date: Date): string {
-		const now = new Date();
-		const diff = now.getTime() - date.getTime();
-		const days = Math.floor(diff / (1000 * 60 * 60 * 24));
-		if (days === 0) return 'Today';
-		if (days === 1) return 'Yesterday';
-		if (days < 7) return `${days} days ago`;
-		return date.toLocaleDateString();
-	}
+
 </script>

-<!-- Mobile overlay backdrop -->
 {#if open}
-	<div
-		class="fixed inset-0 z-40 bg-black/50 md:hidden"
-		role="button"
-		tabindex="-1"
-		onclick={onClose}
-		onkeydown={(e) => { if (e.key === 'Escape') onClose?.(); }}
-		aria-label="Close sidebar"
-	></div>
+	<Backdrop onClose={() => onClose?.()} />
 {/if}

 <!-- Sidebar: always visible on md+, slide-in drawer on mobile when open -->
@@ -105,7 +91,7 @@
 				>
 					<div class="min-w-0 flex-1">
 						<p class="truncate text-sm font-medium text-gray-900 dark:text-gray-100">{session.title}</p>
-						<p class="mt-0.5 text-xs text-gray-500 dark:text-gray-400">{formatDate(session.createdAt)}</p>
+						<p class="mt-0.5 text-xs text-gray-500 dark:text-gray-400">{formatRelativeDate(session.createdAt)}</p>
 					</div>
 					<button
 						type="button"
--- a/src/lib/composables/useOrchestration.svelte.ts
+++ b/src/lib/composables/useOrchestration.svelte.ts
@@ -0,0 +1,174 @@
+import type { ChatMessage } from '$lib/types';
+import type { SubagentResult, InferenceStats } from '$lib/proto/llm_multiverse/v1/common_pb';
+import type { SessionConfig } from '$lib/proto/llm_multiverse/v1/orchestrator_pb';
+import { processRequest, OrchestratorError, friendlyMessage } from '$lib/services/orchestrator';
+import { OrchestrationState } from '$lib/proto/llm_multiverse/v1/orchestrator_pb';
+import { sessionStore } from '$lib/stores/sessions.svelte';
+import { memoryStore } from '$lib/stores/memory.svelte';
+import { auditStore } from '$lib/stores/audit.svelte';
+import { logger } from '$lib/utils/logger';
+
+export function createOrchestration() {
+	let isStreaming = $state(false);
+	let error: string | null = $state(null);
+	let lastFailedContent: string | null = $state(null);
+	let orchestrationState: OrchestrationState = $state(OrchestrationState.UNSPECIFIED);
+	let intermediateResult: string = $state('');
+	let finalResult: SubagentResult | null = $state(null);
+	let inferenceStats: InferenceStats | null = $state(null);
+
+	async function send(
+		sessionId: string,
+		content: string,
+		config: SessionConfig,
+		messages: ChatMessage[]
+	): Promise<ChatMessage[]> {
+		error = null;
+		lastFailedContent = null;
+		orchestrationState = OrchestrationState.UNSPECIFIED;
+		intermediateResult = '';
+		finalResult = null;
+		inferenceStats = null;
+
+		let lastAuditState = OrchestrationState.UNSPECIFIED;
+
+		const userMessage: ChatMessage = {
+			id: crypto.randomUUID(),
+			role: 'user',
+			content,
+			// eslint-disable-next-line svelte/prefer-svelte-reactivity -- plain data, not reactive state
+			timestamp: new Date()
+		};
+		messages.push(userMessage);
+
+		const assistantMessage: ChatMessage = {
+			id: crypto.randomUUID(),
+			role: 'assistant',
+			content: '',
+			// eslint-disable-next-line svelte/prefer-svelte-reactivity -- plain data, not reactive state
+			timestamp: new Date()
+		};
+		messages.push(assistantMessage);
+
+		sessionStore.updateMessages(sessionId, messages);
+		isStreaming = true;
+
+		try {
+			for await (const response of processRequest(sessionId, content, config)) {
+				orchestrationState = response.state;
+
+				if (response.state !== lastAuditState) {
+					const stateLabel = OrchestrationState[response.state] ?? String(response.state);
+					logger.debug('useOrchestration', `State: ${stateLabel}`, {
+						sessionId,
+						state: response.state
+					});
+					auditStore.addEvent(sessionId, {
+						eventType: 'state_change',
+						details: response.message || `State changed to ${stateLabel}`,
+						state: stateLabel
+					});
+					lastAuditState = response.state;
+				}
+
+				if (response.intermediateResult) {
+					intermediateResult = response.intermediateResult;
+				}
+				if (response.inferenceStats) {
+					inferenceStats = response.inferenceStats;
+				}
+				if (response.finalResult) {
+					finalResult = response.finalResult;
+					if (response.finalResult.newMemoryCandidates.length > 0) {
+						memoryStore.addCandidates(
+							sessionId,
+							response.finalResult.newMemoryCandidates.map((mc) => ({
+								content: mc.content,
+								source: mc.source,
+								confidence: mc.confidence
+							}))
+						);
+					}
+				}
+				const idx = messages.length - 1;
+				messages[idx] = {
+					...messages[idx],
+					content: response.message
+				};
+			}
+		} catch (err) {
+			const isOrcErr = err instanceof OrchestratorError;
+			const code = isOrcErr ? err.code : 'unknown';
+			const details = isOrcErr ? err.details : undefined;
+			const friendlyMsg = isOrcErr
+				? friendlyMessage(err.code)
+				: 'An unexpected error occurred';
+			const displayMsg =
+				code && code !== 'unknown'
+					? `${friendlyMsg} (${code})`
+					: friendlyMsg;
+			error = displayMsg;
+			lastFailedContent = content;
+			logger.error('useOrchestration', 'Request failed', { code, details });
+			auditStore.addEvent(sessionId, {
+				eventType: 'error',
+				details: `${friendlyMsg} | code=${code}${details ? ` | details=${details}` : ''}`
+			});
+			const idx = messages.length - 1;
+			messages[idx] = {
+				...messages[idx],
+				content: `\u26A0 ${displayMsg}`
+			};
+		} finally {
+			isStreaming = false;
+			sessionStore.updateMessages(sessionId, messages);
+		}
+
+		return messages;
+	}
+
+	function retry(
+		sessionId: string,
+		config: SessionConfig,
+		messages: ChatMessage[]
+	): { messages: ChatMessage[]; sending: boolean } {
+		if (!lastFailedContent) return { messages, sending: false };
+		const content = lastFailedContent;
+		// Remove the failed user+assistant pair before retrying
+		if (
+			messages.length >= 2 &&
+			messages[messages.length - 2].role === 'user' &&
+			messages[messages.length - 1].role === 'assistant' &&
+			messages[messages.length - 1].content.startsWith('\u26A0')
+		) {
+			messages = messages.slice(0, -2);
+		}
+		send(sessionId, content, config, messages);
+		return { messages, sending: true };
+	}
+
+	function reset() {
+		error = null;
+		finalResult = null;
+		inferenceStats = null;
+	}
+
+	return {
+		get isStreaming() { return isStreaming; },
+		get error() { return error; },
+		get lastFailedContent() { return lastFailedContent; },
+		get orchestrationState() { return orchestrationState; },
+		get intermediateResult() { return intermediateResult; },
+		get finalResult() { return finalResult; },
+		get inferenceStats() { return inferenceStats; },
+		send,
+		retry,
+		reset
+	};
+}
+
+/**
+ * Module-level singleton so orchestration state survives tab/route changes.
+ * The chat page uses this instead of calling createOrchestration() per mount.
+ */
+export const orchestrationStore = createOrchestration();
--- a/src/lib/proto/llm_multiverse/v1/common_pb.ts
+++ b/src/lib/proto/llm_multiverse/v1/common_pb.ts
@@ -12,7 +12,47 @@ import type { Message } from "@bufbuild/protobuf";
 * Describes the file llm_multiverse/v1/common.proto.
 */
 export const file_llm_multiverse_v1_common: GenFile = /*@__PURE__*/
-  fileDesc("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", [file_google_protobuf_timestamp]);
+  fileDesc("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", [file_google_protobuf_timestamp]);
+
+/**
+ * A concrete output produced by an agent (code, command output, etc.).
+ *
+ * @generated from message llm_multiverse.v1.Artifact
+ */
+export type Artifact = Message<"llm_multiverse.v1.Artifact"> & {
+  /**
+   * Display name (filename, query, etc.)
+   *
+   * @generated from field: string label = 1;
+   */
+  label: string;
+
+  /**
+   * Full content
+   *
+   * @generated from field: string content = 2;
+   */
+  content: string;
+
+  /**
+   * @generated from field: llm_multiverse.v1.ArtifactType artifact_type = 3;
+   */
+  artifactType: ArtifactType;
+
+  /**
+   * language, path, tool_name, exit_code, etc.
+   *
+   * @generated from field: map<string, string> metadata = 4;
+   */
+  metadata: { [key: string]: string };
+};
+
+/**
+ * Describes the message llm_multiverse.v1.Artifact.
+ * Use `create(ArtifactSchema)` to create a new message.
+ */
+export const ArtifactSchema: GenMessage<Artifact> = /*@__PURE__*/
+  messageDesc(file_llm_multiverse_v1_common, 0);

 /**
 * Identifies a single agent in the lineage chain.
@@ -41,7 +81,7 @@ export type AgentIdentifier = Message<"llm_multiverse.v1.AgentIdentifier"> & {
 * Use `create(AgentIdentifierSchema)` to create a new message.
 */
 export const AgentIdentifierSchema: GenMessage<AgentIdentifier> = /*@__PURE__*/
-  messageDesc(file_llm_multiverse_v1_common, 0);
+  messageDesc(file_llm_multiverse_v1_common, 1);

 /**
 * Ordered chain of agents from orchestrator (index 0) to current agent.
@@ -61,7 +101,7 @@ export type AgentLineage = Message<"llm_multiverse.v1.AgentLineage"> & {
 * Use `create(AgentLineageSchema)` to create a new message.
 */
 export const AgentLineageSchema: GenMessage<AgentLineage> = /*@__PURE__*/
-  messageDesc(file_llm_multiverse_v1_common, 1);
+  messageDesc(file_llm_multiverse_v1_common, 2);

 /**
 * Carried in every gRPC request for audit trail and broker enforcement.
@@ -100,7 +140,7 @@ export type SessionContext = Message<"llm_multiverse.v1.SessionContext"> & {
 * Use `create(SessionContextSchema)` to create a new message.
 */
 export const SessionContextSchema: GenMessage<SessionContext> = /*@__PURE__*/
-  messageDesc(file_llm_multiverse_v1_common, 2);
+  messageDesc(file_llm_multiverse_v1_common, 3);

 /**
 * Structured error detail for gRPC error responses.
@@ -129,7 +169,7 @@ export type ErrorDetail = Message<"llm_multiverse.v1.ErrorDetail"> & {
 * Use `create(ErrorDetailSchema)` to create a new message.
 */
 export const ErrorDetailSchema: GenMessage<ErrorDetail> = /*@__PURE__*/
-  messageDesc(file_llm_multiverse_v1_common, 3);
+  messageDesc(file_llm_multiverse_v1_common, 4);

 /**
 * A candidate memory entry proposed by a subagent for persistence.
@@ -158,7 +198,56 @@ export type MemoryCandidate = Message<"llm_multiverse.v1.MemoryCandidate"> & {
 * Use `create(MemoryCandidateSchema)` to create a new message.
 */
 export const MemoryCandidateSchema: GenMessage<MemoryCandidate> = /*@__PURE__*/
-  messageDesc(file_llm_multiverse_v1_common, 4);
+  messageDesc(file_llm_multiverse_v1_common, 5);
+
+/**
+ * Inference statistics surfaced from model-gateway through the orchestrator.
+ *
+ * @generated from message llm_multiverse.v1.InferenceStats
+ */
+export type InferenceStats = Message<"llm_multiverse.v1.InferenceStats"> & {
+  /**
+   * Number of tokens in the prompt.
+   *
+   * @generated from field: uint32 prompt_tokens = 1;
+   */
+  promptTokens: number;
+
+  /**
+   * Number of tokens generated.
+   *
+   * @generated from field: uint32 completion_tokens = 2;
+   */
+  completionTokens: number;
+
+  /**
+   * Sum of prompt + completion tokens.
+   *
+   * @generated from field: uint32 total_tokens = 3;
+   */
+  totalTokens: number;
+
+  /**
+   * Model's maximum context length.
+   *
+   * @generated from field: uint32 context_window_size = 4;
+   */
+  contextWindowSize: number;
+
+  /**
+   * Generation throughput (tokens per second).
+   *
+   * @generated from field: float tokens_per_second = 5;
+   */
+  tokensPerSecond: number;
+};
+
+/**
+ * Describes the message llm_multiverse.v1.InferenceStats.
+ * Use `create(InferenceStatsSchema)` to create a new message.
+ */
+export const InferenceStatsSchema: GenMessage<InferenceStats> = /*@__PURE__*/
+  messageDesc(file_llm_multiverse_v1_common, 6);

 /**
 * Standardized return value from any subagent to its parent.
@@ -179,9 +268,11 @@ export type SubagentResult = Message<"llm_multiverse.v1.SubagentResult"> & {
  summary: string;

  /**
-   * @generated from field: repeated string artifacts = 3;
+   * Structured artifacts produced during the agent loop.
+   *
+   * @generated from field: repeated llm_multiverse.v1.Artifact artifacts = 3;
   */
-  artifacts: string[];
+  artifacts: Artifact[];

  /**
   * @generated from field: llm_multiverse.v1.ResultQuality result_quality = 4;
@@ -209,7 +300,7 @@ export type SubagentResult = Message<"llm_multiverse.v1.SubagentResult"> & {
 * Use `create(SubagentResultSchema)` to create a new message.
 */
 export const SubagentResultSchema: GenMessage<SubagentResult> = /*@__PURE__*/
-  messageDesc(file_llm_multiverse_v1_common, 5);
+  messageDesc(file_llm_multiverse_v1_common, 7);

 /**
 * Agent types with distinct tool permission manifests.
@@ -450,3 +541,49 @@ export enum ResultSource {
 export const ResultSourceSchema: GenEnum<ResultSource> = /*@__PURE__*/
  enumDesc(file_llm_multiverse_v1_common, 5);

+/**
+ * Type of artifact produced by an agent during its tool call loop.
+ *
+ * @generated from enum llm_multiverse.v1.ArtifactType
+ */
+export enum ArtifactType {
+  /**
+   * @generated from enum value: ARTIFACT_TYPE_UNSPECIFIED = 0;
+   */
+  UNSPECIFIED = 0,
+
+  /**
+   * Code written via fs_write
+   *
+   * @generated from enum value: ARTIFACT_TYPE_CODE = 1;
+   */
+  CODE = 1,
+
+  /**
+   * Plain text / file content from fs_read
+   *
+   * @generated from enum value: ARTIFACT_TYPE_TEXT = 2;
+   */
+  TEXT = 2,
+
+  /**
+   * Output from run_code / run_shell
+   *
+   * @generated from enum value: ARTIFACT_TYPE_COMMAND_OUTPUT = 3;
+   */
+  COMMAND_OUTPUT = 3,
+
+  /**
+   * Web search results
+   *
+   * @generated from enum value: ARTIFACT_TYPE_SEARCH_RESULT = 4;
+   */
+  SEARCH_RESULT = 4,
+}
+
+/**
+ * Describes the enum llm_multiverse.v1.ArtifactType.
+ */
+export const ArtifactTypeSchema: GenEnum<ArtifactType> = /*@__PURE__*/
+  enumDesc(file_llm_multiverse_v1_common, 6);
+
--- a/src/lib/proto/llm_multiverse/v1/orchestrator_pb.ts
+++ b/src/lib/proto/llm_multiverse/v1/orchestrator_pb.ts
@@ -4,7 +4,7 @@

 import type { GenEnum, GenFile, GenMessage, GenService } from "@bufbuild/protobuf/codegenv2";
 import { enumDesc, fileDesc, messageDesc, serviceDesc } from "@bufbuild/protobuf/codegenv2";
-import type { AgentType, OverrideLevel, SessionContext, SubagentResult, ToolType } from "./common_pb";
+import type { AgentType, InferenceStats, OverrideLevel, SessionContext, SubagentResult, ToolType } from "./common_pb";
 import { file_llm_multiverse_v1_common } from "./common_pb";
 import type { Message } from "@bufbuild/protobuf";

@@ -12,7 +12,7 @@ import type { Message } from "@bufbuild/protobuf";
 * Describes the file llm_multiverse/v1/orchestrator.proto.
 */
 export const file_llm_multiverse_v1_orchestrator: GenFile = /*@__PURE__*/
-  fileDesc("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", [file_llm_multiverse_v1_common]);
+  fileDesc("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", [file_llm_multiverse_v1_common]);

 /**
 * Per-session configuration for override control.
@@ -199,6 +199,13 @@ export type ProcessRequestResponse = Message<"llm_multiverse.v1.ProcessRequestRe
   * @generated from field: optional llm_multiverse.v1.SubagentResult final_result = 4;
   */
  finalResult?: SubagentResult;
+
+  /**
+   * Inference statistics from the model-gateway (on the final streamed message).
+   *
+   * @generated from field: optional llm_multiverse.v1.InferenceStats inference_stats = 5;
+   */
+  inferenceStats?: InferenceStats;
 };

 /**
--- a/src/lib/services/orchestrator.ts
+++ b/src/lib/services/orchestrator.ts
@@ -1,4 +1,4 @@
-import { createClient } from '@connectrpc/connect';
+import { createClient, ConnectError, Code } from '@connectrpc/connect';
 import { createGrpcWebTransport } from '@connectrpc/connect-web';
 import { OrchestratorService } from '$lib/proto/llm_multiverse/v1/orchestrator_pb';
 import type {
@@ -9,6 +9,7 @@ import { create } from '@bufbuild/protobuf';
 import { ProcessRequestRequestSchema } from '$lib/proto/llm_multiverse/v1/orchestrator_pb';
 import { connectionStore } from '$lib/stores/connection.svelte';
 import { toastStore } from '$lib/stores/toast.svelte';
+import { logger } from '$lib/utils/logger';

 /**
 * Application-level error wrapping gRPC status codes.
@@ -79,15 +80,18 @@ function getTransport(endpoint?: string) {
 /**
 * Reset the transport (useful for reconfiguring the endpoint).
 */
-export function resetTransport(): void {
+export function resetTransport(newEndpoint?: string): void {
 	transport = null;
+	if (newEndpoint !== undefined) {
+		transport = createGrpcWebTransport({ baseUrl: newEndpoint });
+	}
 }

 /**
 * Create a configured orchestrator client.
 */
-function getClient(endpoint?: string) {
-	return createClient(OrchestratorService, getTransport(endpoint));
+function getClient() {
+	return createClient(OrchestratorService, getTransport());
 }

 /**
@@ -109,10 +113,35 @@ function backoffDelay(attempt: number): number {

 const MAX_RETRIES = 3;

+/**
+ * Map numeric Code enum values to lowercase string names matching GRPC_USER_MESSAGES keys.
+ */
+const CODE_TO_STRING: Record<number, string> = {
+	[Code.Canceled]: 'cancelled',
+	[Code.Unknown]: 'unknown',
+	[Code.InvalidArgument]: 'failed_precondition',
+	[Code.DeadlineExceeded]: 'deadline_exceeded',
+	[Code.NotFound]: 'not_found',
+	[Code.AlreadyExists]: 'already_exists',
+	[Code.PermissionDenied]: 'permission_denied',
+	[Code.ResourceExhausted]: 'resource_exhausted',
+	[Code.FailedPrecondition]: 'failed_precondition',
+	[Code.Aborted]: 'aborted',
+	[Code.OutOfRange]: 'failed_precondition',
+	[Code.Unimplemented]: 'unimplemented',
+	[Code.Internal]: 'internal',
+	[Code.Unavailable]: 'unavailable',
+	[Code.DataLoss]: 'data_loss',
+	[Code.Unauthenticated]: 'unauthenticated'
+};
+
 /**
 * Extract gRPC error code from an error, normalising to lowercase string.
 */
 function extractCode(err: unknown): string {
+	if (err instanceof ConnectError) {
+		return CODE_TO_STRING[err.code] ?? 'unknown';
+	}
 	if (err instanceof Error && 'code' in err) {
 		const raw = (err as { code: unknown }).code;
 		if (typeof raw === 'string') return raw.toLowerCase();
@@ -126,6 +155,14 @@ function extractCode(err: unknown): string {
 */
 function toOrchestratorError(err: unknown): OrchestratorError {
 	if (err instanceof OrchestratorError) return err;
+	if (err instanceof ConnectError) {
+		const code = extractCode(err);
+		const details = [
+			err.rawMessage,
+			err.cause ? String(err.cause) : ''
+		].filter(Boolean).join('; ');
+		return new OrchestratorError(friendlyMessage(code), code, details || undefined);
+	}
 	if (err instanceof Error) {
 		const code = extractCode(err);
 		return new OrchestratorError(friendlyMessage(code), code, err.message);
@@ -133,6 +170,13 @@ function toOrchestratorError(err: unknown): OrchestratorError {
 	return new OrchestratorError(friendlyMessage('unknown'), 'unknown');
 }

+/**
+ * Append a diagnostic code suffix to a message, e.g. "(code: unavailable)".
+ */
+function diagnosticSuffix(err: OrchestratorError): string {
+	return err.code && err.code !== 'unknown' ? ` (code: ${err.code})` : '';
+}
+
 /**
 * Send a request to the orchestrator and yield streaming responses.
 *
@@ -147,8 +191,7 @@ function toOrchestratorError(err: unknown): OrchestratorError {
 export async function* processRequest(
 	sessionId: string,
 	userMessage: string,
-	sessionConfig?: SessionConfig,
-	endpoint?: string
+	sessionConfig?: SessionConfig
 ): AsyncGenerator<ProcessRequestResponse> {
 	const request = create(ProcessRequestRequestSchema, {
 		sessionId,
@@ -156,24 +199,41 @@ export async function* processRequest(
 		sessionConfig
 	});

+	logger.debug('orchestrator', 'processRequest', {
+		sessionId,
+		messageLength: userMessage.length
+	});
+
 	let lastError: OrchestratorError | null = null;

 	for (let attempt = 0; attempt <= MAX_RETRIES; attempt++) {
+		// Skip retries if already known disconnected
+		if (connectionStore.status === 'disconnected' && attempt > 0) {
+			break;
+		}
+
 		if (attempt > 0) {
 			connectionStore.setReconnecting();
 			const delay = backoffDelay(attempt - 1);
+			logger.warn('orchestrator', `Retry attempt ${attempt}/${MAX_RETRIES}`, {
+				sessionId,
+				previousCode: lastError?.code,
+				delay
+			});
 			await sleep(delay);
 		}

 		try {
-			const client = getClient(endpoint);
+			const client = getClient();
 			for await (const response of client.processRequest(request)) {
 				connectionStore.reportSuccess();
 				yield response;
 			}
+			logger.debug('orchestrator', 'Stream completed', { sessionId });
 			// Completed successfully — no retry needed
 			return;
 		} catch (err: unknown) {
+			logger.grpcError('orchestrator', `Request failed (attempt ${attempt + 1}/${MAX_RETRIES + 1})`, err);
 			lastError = toOrchestratorError(err);
 			const code = lastError.code;

@@ -185,7 +245,12 @@ export async function* processRequest(

 			// Non-transient or exhausted retries
 			connectionStore.reportFailure();
-			toastStore.addToast({ message: lastError.message, type: 'error' });
+			const suffix = diagnosticSuffix(lastError);
+			logger.error('orchestrator', 'Request failed permanently', {
+				code: lastError.code,
+				details: lastError.details
+			});
+			toastStore.addToast({ message: lastError.message + suffix, type: 'error' });
 			throw lastError;
 		}
 	}
--- a/src/lib/stores/audit.svelte.ts
+++ b/src/lib/stores/audit.svelte.ts
@@ -40,168 +40,9 @@ function saveEvents(events: SvelteMap<string, AuditEvent[]>) {
 	localStorage.setItem(STORAGE_KEY, JSON.stringify([...events.entries()]));
 }

-function getSampleData(): SvelteMap<string, AuditEvent[]> {
-	const samples = new SvelteMap<string, AuditEvent[]>();
-	const now = Date.now();
-
-	samples.set('session-demo-alpha', [
-		{
-			id: 'evt-001',
-			sessionId: 'session-demo-alpha',
-			timestamp: new Date(now - 300_000),
-			eventType: 'state_change',
-			details: 'Orchestration started — decomposing user request',
-			state: 'DECOMPOSING'
-		},
-		{
-			id: 'evt-002',
-			sessionId: 'session-demo-alpha',
-			timestamp: new Date(now - 280_000),
-			eventType: 'tool_invocation',
-			details: 'Invoked web_search tool for "SvelteKit best practices"',
-			state: 'DECOMPOSING'
-		},
-		{
-			id: 'evt-003',
-			sessionId: 'session-demo-alpha',
-			timestamp: new Date(now - 260_000),
-			eventType: 'state_change',
-			details: 'Dispatching subtasks to subagents',
-			state: 'DISPATCHING'
-		},
-		{
-			id: 'evt-004',
-			sessionId: 'session-demo-alpha',
-			timestamp: new Date(now - 240_000),
-			eventType: 'message',
-			details: 'Researcher agent assigned to subtask "Gather TypeScript patterns"'
-		},
-		{
-			id: 'evt-005',
-			sessionId: 'session-demo-alpha',
-			timestamp: new Date(now - 220_000),
-			eventType: 'state_change',
-			details: 'Subagents executing tasks',
-			state: 'EXECUTING'
-		},
-		{
-			id: 'evt-006',
-			sessionId: 'session-demo-alpha',
-			timestamp: new Date(now - 200_000),
-			eventType: 'tool_invocation',
-			details: 'Invoked code_analysis tool on project source files'
-		},
-		{
-			id: 'evt-007',
-			sessionId: 'session-demo-alpha',
-			timestamp: new Date(now - 180_000),
-			eventType: 'error',
-			details: 'Timeout waiting for code_analysis response (retrying)'
-		},
-		{
-			id: 'evt-008',
-			sessionId: 'session-demo-alpha',
-			timestamp: new Date(now - 160_000),
-			eventType: 'tool_invocation',
-			details: 'Retry: code_analysis tool succeeded'
-		},
-		{
-			id: 'evt-009',
-			sessionId: 'session-demo-alpha',
-			timestamp: new Date(now - 140_000),
-			eventType: 'state_change',
-			details: 'Compacting results from subagents',
-			state: 'COMPACTING'
-		},
-		{
-			id: 'evt-010',
-			sessionId: 'session-demo-alpha',
-			timestamp: new Date(now - 120_000),
-			eventType: 'state_change',
-			details: 'Orchestration complete',
-			state: 'COMPLETE'
-		}
-	]);
-
-	samples.set('session-demo-beta', [
-		{
-			id: 'evt-011',
-			sessionId: 'session-demo-beta',
-			timestamp: new Date(now - 600_000),
-			eventType: 'state_change',
-			details: 'Orchestration started — decomposing deployment request',
-			state: 'DECOMPOSING'
-		},
-		{
-			id: 'evt-012',
-			sessionId: 'session-demo-beta',
-			timestamp: new Date(now - 580_000),
-			eventType: 'state_change',
-			details: 'Dispatching to sysadmin agent',
-			state: 'DISPATCHING'
-		},
-		{
-			id: 'evt-013',
-			sessionId: 'session-demo-beta',
-			timestamp: new Date(now - 560_000),
-			eventType: 'state_change',
-			details: 'Executing deployment analysis',
-			state: 'EXECUTING'
-		},
-		{
-			id: 'evt-014',
-			sessionId: 'session-demo-beta',
-			timestamp: new Date(now - 540_000),
-			eventType: 'tool_invocation',
-			details: 'Invoked kubectl_check tool for cluster status'
-		},
-		{
-			id: 'evt-015',
-			sessionId: 'session-demo-beta',
-			timestamp: new Date(now - 520_000),
-			eventType: 'error',
-			details: 'Permission denied: kubectl access not granted for this session'
-		},
-		{
-			id: 'evt-016',
-			sessionId: 'session-demo-beta',
-			timestamp: new Date(now - 500_000),
-			eventType: 'message',
-			details: 'Agent requested elevated permissions for cluster access'
-		},
-		{
-			id: 'evt-017',
-			sessionId: 'session-demo-beta',
-			timestamp: new Date(now - 480_000),
-			eventType: 'state_change',
-			details: 'Compacting partial results',
-			state: 'COMPACTING'
-		},
-		{
-			id: 'evt-018',
-			sessionId: 'session-demo-beta',
-			timestamp: new Date(now - 460_000),
-			eventType: 'state_change',
-			details: 'Orchestration complete with warnings',
-			state: 'COMPLETE'
-		}
-	]);
-
-	return samples;
-}
-
 function createAuditStore() {
 	const events = $state<SvelteMap<string, AuditEvent[]>>(loadEvents());

-	// Seed sample data if store is empty
-	if (events.size === 0) {
-		const samples = getSampleData();
-		for (const [id, evts] of samples) {
-			events.set(id, evts);
-		}
-		saveEvents(events);
-	}
-
 	function addEvent(
 		sessionId: string,
 		event: Omit<AuditEvent, 'id' | 'sessionId' | 'timestamp'>
--- a/src/lib/stores/memory.svelte.ts
+++ b/src/lib/stores/memory.svelte.ts
@@ -13,9 +13,11 @@ export interface SessionMemory {
 }

 const STORAGE_KEY = 'llm-multiverse-memory-candidates';
+const SAMPLE_CLEANUP_KEY = 'llm-multiverse-sample-data-cleaned';

 function loadMemory(): SvelteMap<string, StoredMemoryCandidate[]> {
 	if (typeof localStorage === 'undefined') return new SvelteMap();
+	_cleanupSampleData();
 	try {
 		const raw = localStorage.getItem(STORAGE_KEY);
 		if (!raw) return new SvelteMap();
@@ -26,79 +28,37 @@ function loadMemory(): SvelteMap<string, StoredMemoryCandidate[]> {
 	}
 }

+/** One-time removal of previously seeded sample sessions. */
+function _cleanupSampleData() {
+	if (typeof localStorage === 'undefined') return;
+	if (localStorage.getItem(SAMPLE_CLEANUP_KEY)) return;
+	try {
+		const raw = localStorage.getItem(STORAGE_KEY);
+		if (raw) {
+			const arr: [string, unknown[]][] = JSON.parse(raw);
+			const filtered = arr.filter(([id]) => !id.startsWith('session-demo-'));
+			localStorage.setItem(STORAGE_KEY, JSON.stringify(filtered));
+		}
+		// Also clean audit store
+		const auditKey = 'llm-multiverse-audit-events';
+		const auditRaw = localStorage.getItem(auditKey);
+		if (auditRaw) {
+			const arr: [string, unknown[]][] = JSON.parse(auditRaw);
+			const filtered = arr.filter(([id]) => !id.startsWith('session-demo-'));
+			localStorage.setItem(auditKey, JSON.stringify(filtered));
+		}
+	} catch { /* ignore */ }
+	localStorage.setItem(SAMPLE_CLEANUP_KEY, '1');
+}
+
 function saveMemory(memory: SvelteMap<string, StoredMemoryCandidate[]>) {
 	if (typeof localStorage === 'undefined') return;
 	localStorage.setItem(STORAGE_KEY, JSON.stringify([...memory.entries()]));
 }

-function getSampleData(): SvelteMap<string, StoredMemoryCandidate[]> {
-	const samples = new SvelteMap<string, StoredMemoryCandidate[]>();
-	samples.set('session-demo-alpha', [
-		{
-			content: 'The user prefers TypeScript strict mode with no implicit any.',
-			source: ResultSource.MODEL_KNOWLEDGE,
-			confidence: 0.92
-		},
-		{
-			content: 'Project uses SvelteKit with Tailwind CSS v4 and Svelte 5 runes.',
-			source: ResultSource.TOOL_OUTPUT,
-			confidence: 0.98
-		},
-		{
-			content: 'Preferred testing framework is Vitest with Playwright for e2e.',
-			source: ResultSource.MODEL_KNOWLEDGE,
-			confidence: 0.75
-		},
-		{
-			content: 'The gRPC backend runs on port 50051 behind a Caddy reverse proxy.',
-			source: ResultSource.TOOL_OUTPUT,
-			confidence: 0.85
-		}
-	]);
-	samples.set('session-demo-beta', [
-		{
-			content: 'User asked about deploying to Kubernetes with Helm charts.',
-			source: ResultSource.WEB,
-			confidence: 0.67
-		},
-		{
-			content: 'The container registry is at registry.example.com.',
-			source: ResultSource.TOOL_OUTPUT,
-			confidence: 0.91
-		},
-		{
-			content: 'Deployment target is a 3-node k3s cluster running on ARM64.',
-			source: ResultSource.WEB,
-			confidence: 0.58
-		}
-	]);
-	samples.set('session-demo-gamma', [
-		{
-			content: 'Database schema uses PostgreSQL with pgvector extension for embeddings.',
-			source: ResultSource.TOOL_OUTPUT,
-			confidence: 0.95
-		},
-		{
-			content: 'The embedding model is all-MiniLM-L6-v2 with 384 dimensions.',
-			source: ResultSource.MODEL_KNOWLEDGE,
-			confidence: 0.82
-		}
-	]);
-	return samples;
-}
-
 function createMemoryStore() {
 	const memory = $state<SvelteMap<string, StoredMemoryCandidate[]>>(loadMemory());

-	// Seed sample data if store is empty
-	if (memory.size === 0) {
-		const samples = getSampleData();
-		for (const [id, candidates] of samples) {
-			memory.set(id, candidates);
-		}
-		saveMemory(memory);
-	}
-
 	function addCandidates(sessionId: string, candidates: StoredMemoryCandidate[]) {
 		if (candidates.length === 0) return;
 		const existing = memory.get(sessionId) ?? [];
--- a/src/lib/stores/sessions.svelte.ts
+++ b/src/lib/stores/sessions.svelte.ts
@@ -1,3 +1,4 @@
+import { SvelteMap } from 'svelte/reactivity';
 import type { ChatMessage } from '$lib/types';

 export interface Session {
@@ -10,24 +11,24 @@ export interface Session {
 const STORAGE_KEY = 'llm-multiverse-sessions';
 const ACTIVE_SESSION_KEY = 'llm-multiverse-active-session';

-function loadSessions(): Map<string, Session> {
-	if (typeof localStorage === 'undefined') return new Map();
+function loadSessions(): SvelteMap<string, Session> {
+	if (typeof localStorage === 'undefined') return new SvelteMap();
 	try {
 		const raw = localStorage.getItem(STORAGE_KEY);
-		if (!raw) return new Map();
+		if (!raw) return new SvelteMap();
 		const arr: [string, Session][] = JSON.parse(raw);
-		return new Map(
+		return new SvelteMap(
 			arr.map(([id, s]) => [
 				id,
 				{ ...s, createdAt: new Date(s.createdAt), messages: s.messages.map(m => ({ ...m, timestamp: new Date(m.timestamp) })) }
 			])
 		);
 	} catch {
-		return new Map();
+		return new SvelteMap();
 	}
 }

-function saveSessions(sessions: Map<string, Session>) {
+function saveSessions(sessions: SvelteMap<string, Session>) {
 	if (typeof localStorage === 'undefined') return;
 	localStorage.setItem(STORAGE_KEY, JSON.stringify([...sessions.entries()]));
 }
@@ -43,7 +44,7 @@ function saveActiveSessionId(id: string) {
 }

 function createSessionStore() {
-	const sessions = $state<Map<string, Session>>(loadSessions());
+	const sessions = $state<SvelteMap<string, Session>>(loadSessions());
 	let activeSessionId = $state<string | null>(loadActiveSessionId());

 	function createSession(id?: string): Session {
@@ -66,6 +67,10 @@ function createSessionStore() {
 			saveActiveSessionId(id);
 			return sessions.get(id)!;
 		}
+		// No specific ID — prefer existing active session
+		if (!id && activeSessionId && sessions.has(activeSessionId)) {
+			return sessions.get(activeSessionId)!;
+		}
 		return createSession(id);
 	}

--- a/src/lib/stores/theme.svelte.ts
+++ b/src/lib/stores/theme.svelte.ts
@@ -5,6 +5,9 @@ const STORAGE_KEY = 'llm-multiverse-theme';
 function createThemeStore() {
 	let mode: ThemeMode = $state('system');
 	let resolvedDark = $state(false);
+	let initialized = false;
+	let mediaQuery: MediaQueryList | null = null;
+	let mediaListener: ((e: MediaQueryListEvent) => void) | null = null;

 	function applyTheme(isDark: boolean) {
 		if (typeof document === 'undefined') return;
@@ -27,8 +30,10 @@ function createThemeStore() {
 		return window.matchMedia('(prefers-color-scheme: dark)').matches;
 	}

-	function init() {
+	function init(): (() => void) | undefined {
 		if (typeof window === 'undefined') return;
+		if (initialized) return;
+		initialized = true;

 		// Load saved preference
 		const saved = localStorage.getItem(STORAGE_KEY) as ThemeMode | null;
@@ -39,18 +44,26 @@ function createThemeStore() {
 		}

 		// Listen for system preference changes
-		const mediaQuery = window.matchMedia('(prefers-color-scheme: dark)');
-		mediaQuery.addEventListener('change', (e) => {
+		mediaQuery = window.matchMedia('(prefers-color-scheme: dark)');
+		mediaListener = (e: MediaQueryListEvent) => {
 			if (mode === 'system') {
 				resolvedDark = e.matches;
 				applyTheme(resolvedDark);
 			}
-		});
+		};
+		mediaQuery.addEventListener('change', mediaListener);

 		// Apply initial theme
 		resolvedDark =
 			mode === 'dark' ? true : mode === 'light' ? false : getSystemPreference();
 		applyTheme(resolvedDark);
+
+		return () => {
+			if (mediaQuery && mediaListener) {
+				mediaQuery.removeEventListener('change', mediaListener);
+			}
+			initialized = false;
+		};
 	}

 	function setMode(newMode: ThemeMode) {
--- a/src/lib/types/lineage.ts
+++ b/src/lib/types/lineage.ts
@@ -42,57 +42,59 @@ export function agentTypeLabel(type: AgentType): string {
 	}
 }

-/**
- * Maps AgentType enum values to Tailwind-friendly color tokens.
- */
-export function agentTypeColor(type: AgentType): {
+export interface AgentColorSet {
 	fill: string;
 	stroke: string;
 	text: string;
+}
+
+export interface AgentColors {
+	light: AgentColorSet;
+	dark: AgentColorSet;
 	badge: string;
-} {
+}
+
+/**
+ * Maps AgentType enum values to light/dark color tokens for SVG rendering
+ * and Tailwind badge classes.
+ */
+export function agentTypeColor(type: AgentType): AgentColors {
 	switch (type) {
 		case AgentType.ORCHESTRATOR:
 			return {
-				fill: '#dbeafe',
-				stroke: '#3b82f6',
-				text: '#1e40af',
-				badge: 'bg-blue-100 text-blue-700'
+				light: { fill: '#dbeafe', stroke: '#3b82f6', text: '#1e40af' },
+				dark: { fill: '#1e3a5f', stroke: '#60a5fa', text: '#93c5fd' },
+				badge: 'bg-blue-100 dark:bg-blue-900/40 text-blue-700 dark:text-blue-300'
 			};
 		case AgentType.RESEARCHER:
 			return {
-				fill: '#dcfce7',
-				stroke: '#22c55e',
-				text: '#166534',
-				badge: 'bg-green-100 text-green-700'
+				light: { fill: '#dcfce7', stroke: '#22c55e', text: '#166534' },
+				dark: { fill: '#14532d', stroke: '#4ade80', text: '#86efac' },
+				badge: 'bg-green-100 dark:bg-green-900/40 text-green-700 dark:text-green-300'
 			};
 		case AgentType.CODER:
 			return {
-				fill: '#f3e8ff',
-				stroke: '#a855f7',
-				text: '#6b21a8',
-				badge: 'bg-purple-100 text-purple-700'
+				light: { fill: '#f3e8ff', stroke: '#a855f7', text: '#6b21a8' },
+				dark: { fill: '#3b0764', stroke: '#c084fc', text: '#d8b4fe' },
+				badge: 'bg-purple-100 dark:bg-purple-900/40 text-purple-700 dark:text-purple-300'
 			};
 		case AgentType.SYSADMIN:
 			return {
-				fill: '#ffedd5',
-				stroke: '#f97316',
-				text: '#9a3412',
-				badge: 'bg-orange-100 text-orange-700'
+				light: { fill: '#ffedd5', stroke: '#f97316', text: '#9a3412' },
+				dark: { fill: '#431407', stroke: '#fb923c', text: '#fdba74' },
+				badge: 'bg-orange-100 dark:bg-orange-900/40 text-orange-700 dark:text-orange-300'
 			};
 		case AgentType.ASSISTANT:
 			return {
-				fill: '#ccfbf1',
-				stroke: '#14b8a6',
-				text: '#115e59',
-				badge: 'bg-teal-100 text-teal-700'
+				light: { fill: '#ccfbf1', stroke: '#14b8a6', text: '#115e59' },
+				dark: { fill: '#134e4a', stroke: '#2dd4bf', text: '#5eead4' },
+				badge: 'bg-teal-100 dark:bg-teal-900/40 text-teal-700 dark:text-teal-300'
 			};
 		default:
 			return {
-				fill: '#f3f4f6',
-				stroke: '#9ca3af',
-				text: '#374151',
-				badge: 'bg-gray-100 text-gray-700'
+				light: { fill: '#f3f4f6', stroke: '#9ca3af', text: '#374151' },
+				dark: { fill: '#1f2937', stroke: '#6b7280', text: '#9ca3af' },
+				badge: 'bg-gray-100 dark:bg-gray-800 text-gray-700 dark:text-gray-300'
 			};
 	}
 }
@@ -149,52 +151,3 @@ export function buildLineageTree(agents: SimpleAgentIdentifier[]): LineageNode[]
 	return roots;
 }

-/**
- * Returns sample/demo lineage data for visualization development.
- * This will be replaced with real API data when available.
- */
-export function getSampleLineageData(): SimpleAgentIdentifier[] {
-	return [
-		{
-			agentId: 'orch-001',
-			agentType: AgentType.ORCHESTRATOR,
-			spawnDepth: 0
-		},
-		{
-			agentId: 'research-001',
-			agentType: AgentType.RESEARCHER,
-			spawnDepth: 1,
-			parentId: 'orch-001'
-		},
-		{
-			agentId: 'coder-001',
-			agentType: AgentType.CODER,
-			spawnDepth: 1,
-			parentId: 'orch-001'
-		},
-		{
-			agentId: 'sysadmin-001',
-			agentType: AgentType.SYSADMIN,
-			spawnDepth: 1,
-			parentId: 'orch-001'
-		},
-		{
-			agentId: 'assist-001',
-			agentType: AgentType.ASSISTANT,
-			spawnDepth: 2,
-			parentId: 'research-001'
-		},
-		{
-			agentId: 'coder-002',
-			agentType: AgentType.CODER,
-			spawnDepth: 2,
-			parentId: 'coder-001'
-		},
-		{
-			agentId: 'research-002',
-			agentType: AgentType.RESEARCHER,
-			spawnDepth: 3,
-			parentId: 'coder-002'
-		}
-	];
-}
--- a/src/lib/types/resultSource.ts
+++ b/src/lib/types/resultSource.ts
@@ -0,0 +1,23 @@
+import { ResultSource } from '$lib/proto/llm_multiverse/v1/common_pb';
+
+export interface ResultSourceStyle {
+	label: string;
+	bg: string;
+	text: string;
+}
+
+/**
+ * Returns label and badge classes for a ResultSource value.
+ */
+export function resultSourceConfig(source: ResultSource): ResultSourceStyle {
+	switch (source) {
+		case ResultSource.TOOL_OUTPUT:
+			return { label: 'Tool Output', bg: 'bg-blue-100 dark:bg-blue-900/40', text: 'text-blue-800 dark:text-blue-300' };
+		case ResultSource.MODEL_KNOWLEDGE:
+			return { label: 'Model Knowledge', bg: 'bg-purple-100 dark:bg-purple-900/40', text: 'text-purple-800 dark:text-purple-300' };
+		case ResultSource.WEB:
+			return { label: 'Web', bg: 'bg-green-100 dark:bg-green-900/40', text: 'text-green-800 dark:text-green-300' };
+		default:
+			return { label: 'Unspecified', bg: 'bg-gray-100 dark:bg-gray-800', text: 'text-gray-800 dark:text-gray-300' };
+	}
+}
--- a/src/lib/utils/date.ts
+++ b/src/lib/utils/date.ts
@@ -0,0 +1,19 @@
+/**
+ * Formats a date as a relative string: "Today", "Yesterday", "N days ago", or locale date.
+ */
+export function formatRelativeDate(date: Date): string {
+	const now = new Date();
+	const diff = now.getTime() - date.getTime();
+	const days = Math.floor(diff / (1000 * 60 * 60 * 24));
+	if (days === 0) return 'Today';
+	if (days === 1) return 'Yesterday';
+	if (days < 7) return `${days} days ago`;
+	return date.toLocaleDateString();
+}
+
+/**
+ * Formats a date as a short string like "Mar 12, 2026".
+ */
+export function formatShortDate(date: Date): string {
+	return date.toLocaleDateString([], { month: 'short', day: 'numeric', year: 'numeric' });
+}
--- a/src/lib/utils/logger.ts
+++ b/src/lib/utils/logger.ts
@@ -0,0 +1,60 @@
+/**
+ * Lightweight dev-mode logging utility.
+ * All diagnostic logging flows through this module.
+ */
+
+import { ConnectError } from '@connectrpc/connect';
+
+declare global {
+	var __LLM_DEBUG: boolean | undefined;
+}
+
+function isDebugEnabled(): boolean {
+	return import.meta.env.DEV || globalThis.__LLM_DEBUG === true;
+}
+
+function fmt(tag: string, message: string): string {
+	return `[${tag}] ${message}`;
+}
+
+export const logger = {
+	debug(tag: string, message: string, ...data: unknown[]): void {
+		if (isDebugEnabled()) {
+			console.debug(fmt(tag, message), ...data);
+		}
+	},
+
+	info(tag: string, message: string, ...data: unknown[]): void {
+		if (isDebugEnabled()) {
+			console.info(fmt(tag, message), ...data);
+		}
+	},
+
+	warn(tag: string, message: string, ...data: unknown[]): void {
+		if (isDebugEnabled()) {
+			console.warn(fmt(tag, message), ...data);
+		}
+	},
+
+	error(tag: string, message: string, ...data: unknown[]): void {
+		if (isDebugEnabled()) {
+			console.error(fmt(tag, message), ...data);
+		}
+	},
+
+	/** Always logs regardless of debug toggle. Destructures ConnectError fields. */
+	grpcError(tag: string, label: string, err: unknown): void {
+		if (err instanceof ConnectError) {
+			console.error(fmt(tag, label), {
+				code: err.code,
+				rawMessage: err.rawMessage,
+				cause: err.cause,
+				metadata: Object.fromEntries(err.metadata.entries())
+			});
+		} else if (err instanceof Error) {
+			console.error(fmt(tag, label), { message: err.message });
+		} else {
+			console.error(fmt(tag, label), err);
+		}
+	}
+};
--- a/src/lib/utils/sessionConfig.ts
+++ b/src/lib/utils/sessionConfig.ts
@@ -0,0 +1,13 @@
+import { OverrideLevel } from '$lib/proto/llm_multiverse/v1/common_pb';
+import type { SessionConfig } from '$lib/proto/llm_multiverse/v1/orchestrator_pb';
+
+/**
+ * Returns true if the session config differs from defaults.
+ */
+export function isNonDefaultConfig(config: SessionConfig): boolean {
+	return (
+		config.overrideLevel !== OverrideLevel.NONE ||
+		config.disabledTools.length > 0 ||
+		config.grantedPermissions.length > 0
+	);
+}
--- a/src/routes/+layout.svelte
+++ b/src/routes/+layout.svelte
@@ -1,13 +1,15 @@
 <script lang="ts">
 	import favicon from '$lib/assets/favicon.svg';
 	import '../app.css';
+	import { onMount } from 'svelte';
 	import { themeStore } from '$lib/stores/theme.svelte';
 	import ToastContainer from '$lib/components/ToastContainer.svelte';

 	let { children } = $props();

-	$effect(() => {
-		themeStore.init();
+	onMount(() => {
+		const cleanup = themeStore.init();
+		return cleanup;
 	});
 </script>

--- a/src/routes/+page.ts
+++ b/src/routes/+page.ts
@@ -0,0 +1,2 @@
+export const prerender = false;
+export const ssr = false;
--- a/src/routes/audit/+page.svelte
+++ b/src/routes/audit/+page.svelte
@@ -3,7 +3,7 @@
 	import { goto } from '$app/navigation';
 	import { resolveRoute } from '$app/paths';
 	import AuditTimeline from '$lib/components/AuditTimeline.svelte';
-	import ThemeToggle from '$lib/components/ThemeToggle.svelte';
+	import PageHeader from '$lib/components/PageHeader.svelte';
 	import { auditStore } from '$lib/stores/audit.svelte';
 	import type { AuditEventType } from '$lib/stores/audit.svelte';

@@ -40,26 +40,7 @@
 </script>

 <div class="flex h-screen flex-col overflow-hidden bg-gray-50 dark:bg-gray-900">
-	<!-- Header -->
-	<header class="flex items-center justify-between border-b border-gray-200 dark:border-gray-700 bg-white dark:bg-gray-900 px-4 py-3 md:px-6">
-		<div class="flex items-center gap-2 md:gap-4">
-			<!-- eslint-disable svelte/no-navigation-without-resolve -- resolveRoute is resolve; plugin does not recognize the alias -->
-			<a
-				href={chatHref}
-				class="rounded-lg px-2.5 py-1.5 text-sm text-gray-600 dark:text-gray-400 hover:bg-gray-100 dark:hover:bg-gray-700 hover:text-gray-900 dark:hover:text-gray-100"
-			>
-				&larr; Chat
-			</a>
-			<!-- eslint-enable svelte/no-navigation-without-resolve -->
-			<h1 class="text-base md:text-lg font-semibold text-gray-900 dark:text-gray-100">Audit Log</h1>
-		</div>
-		<div class="flex items-center gap-2">
-			<ThemeToggle />
-			<span class="hidden sm:inline rounded-md bg-amber-50 dark:bg-amber-900/30 px-2 py-1 text-xs text-amber-700 dark:text-amber-300">
-				Sample Data
-			</span>
-		</div>
-	</header>
+	<PageHeader title="Audit Log" backHref={chatHref} />

 	<!-- Filters -->
 	<div class="border-b border-gray-200 dark:border-gray-700 bg-white dark:bg-gray-900 px-4 py-3 md:px-6">
--- a/src/routes/chat/+page.svelte
+++ b/src/routes/chat/+page.svelte
@@ -3,7 +3,6 @@
 	import { goto } from '$app/navigation';
 	import { resolveRoute } from '$app/paths';
 	import type { ChatMessage } from '$lib/types';
-	import type { SubagentResult } from '$lib/proto/llm_multiverse/v1/common_pb';
 	import { OverrideLevel } from '$lib/proto/llm_multiverse/v1/common_pb';
 	import type { SessionConfig } from '$lib/proto/llm_multiverse/v1/orchestrator_pb';
 	import { SessionConfigSchema } from '$lib/proto/llm_multiverse/v1/orchestrator_pb';
@@ -13,24 +12,18 @@
 	import OrchestrationProgress from '$lib/components/OrchestrationProgress.svelte';
 	import ThinkingSection from '$lib/components/ThinkingSection.svelte';
 	import FinalResult from '$lib/components/FinalResult.svelte';
+	import InferenceStatsPanel from '$lib/components/InferenceStatsPanel.svelte';
 	import SessionSidebar from '$lib/components/SessionSidebar.svelte';
 	import ConfigSidebar from '$lib/components/ConfigSidebar.svelte';
 	import ThemeToggle from '$lib/components/ThemeToggle.svelte';
 	import ConnectionStatus from '$lib/components/ConnectionStatus.svelte';
-	import { processRequest, OrchestratorError, friendlyMessage } from '$lib/services/orchestrator';
-	import { OrchestrationState } from '$lib/proto/llm_multiverse/v1/orchestrator_pb';
+	import { onMount, untrack } from 'svelte';
 	import { sessionStore } from '$lib/stores/sessions.svelte';
-	import { memoryStore } from '$lib/stores/memory.svelte';
-	import { auditStore } from '$lib/stores/audit.svelte';
-	import { toastStore } from '$lib/stores/toast.svelte';
+	import { isNonDefaultConfig } from '$lib/utils/sessionConfig';
+	import { orchestrationStore as orchestration } from '$lib/composables/useOrchestration.svelte';

 	let messages: ChatMessage[] = $state([]);
-	let isStreaming = $state(false);
-	let error: string | null = $state(null);
-	let lastFailedContent: string | null = $state(null);
-	let orchestrationState: OrchestrationState = $state(OrchestrationState.UNSPECIFIED);
-	let intermediateResult: string = $state('');
-	let finalResult: SubagentResult | null = $state(null);
+	let initialized = $state(false);
 	let sessionConfig: SessionConfig = $state(
 		create(SessionConfigSchema, { overrideLevel: OverrideLevel.NONE })
 	);
@@ -39,12 +32,7 @@
 	const lineageHref = resolveRoute('/lineage');
 	const memoryHref = resolveRoute('/memory');
 	const auditHref = resolveRoute('/audit');
-
-	const isNonDefaultConfig = $derived(
-		sessionConfig.overrideLevel !== OverrideLevel.NONE ||
-			sessionConfig.disabledTools.length > 0 ||
-			sessionConfig.grantedPermissions.length > 0
-	);
+	const hasNonDefaultConfig = $derived(isNonDefaultConfig(sessionConfig));

 	function navigateToSession(sessionId: string, replace = false) {
 		const url = `${resolveRoute('/chat')}?session=${sessionId}`;
@@ -52,20 +40,32 @@
 		goto(url, { replaceState: replace });
 	}

-	$effect(() => {
+	onMount(() => {
 		const sessionParam = $page.url.searchParams.get('session');
 		const session = sessionStore.getOrCreateSession(sessionParam ?? undefined);
 		messages = [...session.messages];
 		if (!sessionParam || sessionParam !== session.id) {
 			navigateToSession(session.id, true);
 		}
+		initialized = true;
+	});
+
+	// Sync messages when session changes via sidebar (after initial mount)
+	$effect(() => {
+		if (!initialized) return;
+		const sessionParam = $page.url.searchParams.get('session');
+		if (sessionParam) {
+			const session = untrack(() => sessionStore.activeSession);
+			if (session && session.id === sessionParam) {
+				messages = [...session.messages];
+			}
+		}
 	});

 	function handleNewChat() {
 		const session = sessionStore.createSession();
 		messages = [];
-		error = null;
-		finalResult = null;
+		orchestration.reset();
 		navigateToSession(session.id);
 	}

@@ -74,109 +74,20 @@
 		const session = sessionStore.activeSession;
 		if (session) {
 			messages = [...session.messages];
-			error = null;
-			finalResult = null;
+			orchestration.reset();
 			navigateToSession(id);
 		}
 	}

 	async function handleSend(content: string) {
-		error = null;
-		lastFailedContent = null;
-		orchestrationState = OrchestrationState.UNSPECIFIED;
-		intermediateResult = '';
-		finalResult = null;
-
 		const sessionId = sessionStore.activeSessionId!;
-		let lastAuditState = OrchestrationState.UNSPECIFIED;
-
-		const userMessage: ChatMessage = {
-			id: crypto.randomUUID(),
-			role: 'user',
-			content,
-			timestamp: new Date()
-		};
-		messages.push(userMessage);
-
-		const assistantMessage: ChatMessage = {
-			id: crypto.randomUUID(),
-			role: 'assistant',
-			content: '',
-			timestamp: new Date()
-		};
-		messages.push(assistantMessage);
-
-		sessionStore.updateMessages(sessionId, messages);
-		isStreaming = true;
-
-		try {
-			for await (const response of processRequest(sessionId, content, sessionConfig)) {
-				orchestrationState = response.state;
-
-				// Capture state changes to the audit log
-				if (response.state !== lastAuditState) {
-					const stateLabel = OrchestrationState[response.state] ?? String(response.state);
-					auditStore.addEvent(sessionId, {
-						eventType: 'state_change',
-						details: response.message || `State changed to ${stateLabel}`,
-						state: stateLabel
-					});
-					lastAuditState = response.state;
-				}
-
-				if (response.intermediateResult) {
-					intermediateResult = response.intermediateResult;
-				}
-				if (response.finalResult) {
-					finalResult = response.finalResult;
-					if (response.finalResult.newMemoryCandidates.length > 0) {
-						memoryStore.addCandidates(
-							sessionId,
-							response.finalResult.newMemoryCandidates.map((mc) => ({
-								content: mc.content,
-								source: mc.source,
-								confidence: mc.confidence
-							}))
-						);
-					}
-				}
-				const idx = messages.length - 1;
-				messages[idx] = {
-					...messages[idx],
-					content: response.message
-				};
-			}
-		} catch (err) {
-			const friendlyMsg =
-				err instanceof OrchestratorError
-					? friendlyMessage(err.code)
-					: 'An unexpected error occurred';
-			error = friendlyMsg;
-			lastFailedContent = content;
-			toastStore.addToast({ message: friendlyMsg, type: 'error' });
-			auditStore.addEvent(sessionId, {
-				eventType: 'error',
-				details: friendlyMsg
-			});
-			const idx = messages.length - 1;
-			messages[idx] = {
-				...messages[idx],
-				content: `\u26A0 ${friendlyMsg}`
-			};
-		} finally {
-			isStreaming = false;
-			sessionStore.updateMessages(sessionId, messages);
-		}
+		messages = await orchestration.send(sessionId, content, sessionConfig, messages);
 	}

 	function handleRetry() {
-		if (!lastFailedContent) return;
-		const content = lastFailedContent;
-		// Remove the failed assistant message before retrying
-		if (messages.length >= 2) {
-			messages = messages.slice(0, -2);
-		}
-		handleSend(content);
+		const sessionId = sessionStore.activeSessionId!;
+		const result = orchestration.retry(sessionId, sessionConfig, messages);
+		messages = result.messages;
 	}
 </script>

@@ -240,7 +151,7 @@
 				class="flex min-h-[44px] min-w-[44px] items-center justify-center gap-1 rounded-lg px-2.5 py-1.5 text-sm md:min-h-0 md:min-w-0
 					{showConfig ? 'bg-blue-100 text-blue-700 dark:bg-blue-900 dark:text-blue-300' : 'bg-gray-100 text-gray-700 hover:bg-gray-200 dark:bg-gray-700 dark:text-gray-300 dark:hover:bg-gray-600'}"
 			>
-				{#if isNonDefaultConfig}
+				{#if hasNonDefaultConfig}
 					<span class="h-2 w-2 rounded-full bg-amber-500"></span>
 				{/if}
 				<span class="hidden sm:inline">Config</span>
@@ -255,12 +166,12 @@

 		<MessageList {messages} />

-		{#if isStreaming}
-			<OrchestrationProgress state={orchestrationState} />
-			<ThinkingSection content={intermediateResult} />
+		{#if orchestration.isStreaming}
+			<OrchestrationProgress state={orchestration.orchestrationState} />
+			<ThinkingSection content={orchestration.intermediateResult} />
 		{/if}

-		{#if isStreaming && messages.length > 0 && messages[messages.length - 1].content === ''}
+		{#if orchestration.isStreaming && messages.length > 0 && messages[messages.length - 1].content === ''}
 			<div class="flex justify-start px-4 pb-2">
 				<div class="flex items-center gap-1.5 rounded-2xl bg-gray-200 dark:bg-gray-700 px-4 py-2.5">
 					<span class="h-2 w-2 animate-bounce rounded-full bg-gray-500 dark:bg-gray-400 [animation-delay:0ms]"
@@ -275,18 +186,22 @@
 			</div>
 		{/if}

-		{#if finalResult && !isStreaming}
-			<FinalResult result={finalResult} />
+		{#if orchestration.finalResult && !orchestration.isStreaming}
+			<FinalResult result={orchestration.finalResult} />
 		{/if}

-		{#if error}
+		{#if orchestration.inferenceStats && !orchestration.isStreaming}
+			<InferenceStatsPanel stats={orchestration.inferenceStats} />
+		{/if}
+
+		{#if orchestration.error}
 			<div class="mx-4 mb-2 flex items-center justify-between gap-3 rounded-lg bg-red-50 dark:bg-red-900/30 px-4 py-2 text-sm text-red-600 dark:text-red-400">
-				<span>{error}</span>
-				{#if lastFailedContent}
+				<span>{orchestration.error}</span>
+				{#if orchestration.lastFailedContent}
 					<button
 						type="button"
 						onclick={handleRetry}
-						disabled={isStreaming}
+						disabled={orchestration.isStreaming}
 						class="shrink-0 rounded-md bg-red-100 px-3 py-1 text-xs font-medium text-red-700 hover:bg-red-200 dark:bg-red-800/50 dark:text-red-300 dark:hover:bg-red-800 disabled:opacity-50"
 					>
 						Retry
@@ -295,7 +210,7 @@
 			</div>
 		{/if}

-		<MessageInput onSend={handleSend} disabled={isStreaming} />
+		<MessageInput onSend={handleSend} disabled={orchestration.isStreaming} />
 	</div>

 	{#if showConfig}
--- a/src/routes/chat/+page.ts
+++ b/src/routes/chat/+page.ts
@@ -0,0 +1,2 @@
+export const prerender = false;
+export const ssr = false;
--- a/src/routes/lineage/+page.svelte
+++ b/src/routes/lineage/+page.svelte
@@ -2,18 +2,19 @@
 	import { resolveRoute } from '$app/paths';
 	import { AgentType } from '$lib/proto/llm_multiverse/v1/common_pb';
 	import LineageTree from '$lib/components/LineageTree.svelte';
-	import ThemeToggle from '$lib/components/ThemeToggle.svelte';
+	import PageHeader from '$lib/components/PageHeader.svelte';
+	import Backdrop from '$lib/components/Backdrop.svelte';
 	import type { LineageNode, SimpleAgentIdentifier } from '$lib/types/lineage';
 	import {
 		buildLineageTree,
-		getSampleLineageData,
 		agentTypeLabel,
 		agentTypeColor
 	} from '$lib/types/lineage';
+	import { themeStore } from '$lib/stores/theme.svelte';

 	const chatHref = resolveRoute('/chat');

-	let agents: SimpleAgentIdentifier[] = $state(getSampleLineageData());
+	let agents: SimpleAgentIdentifier[] = $state([]);
 	let treeNodes: LineageNode[] = $derived(buildLineageTree(agents));
 	let selectedNode: LineageNode | null = $state(null);

@@ -32,26 +33,7 @@
 </script>

 <div class="flex h-screen flex-col overflow-hidden bg-gray-50 dark:bg-gray-900">
-	<!-- Header -->
-	<header class="flex items-center justify-between border-b border-gray-200 dark:border-gray-700 bg-white dark:bg-gray-900 px-4 py-3 md:px-6">
-		<div class="flex items-center gap-2 md:gap-4">
-			<!-- eslint-disable svelte/no-navigation-without-resolve -- resolveRoute is resolve; plugin does not recognize the alias -->
-			<a
-				href={chatHref}
-				class="rounded-lg px-2.5 py-1.5 text-sm text-gray-600 dark:text-gray-400 hover:bg-gray-100 dark:hover:bg-gray-700 hover:text-gray-900 dark:hover:text-gray-100"
-			>
-				&larr; Chat
-			</a>
-			<!-- eslint-enable svelte/no-navigation-without-resolve -->
-			<h1 class="text-base md:text-lg font-semibold text-gray-900 dark:text-gray-100">Agent Lineage</h1>
-		</div>
-		<div class="flex items-center gap-2">
-			<ThemeToggle />
-			<span class="hidden sm:inline rounded-md bg-amber-50 dark:bg-amber-900/30 px-2 py-1 text-xs text-amber-700 dark:text-amber-300">
-				Sample Data
-			</span>
-		</div>
-	</header>
+	<PageHeader title="Agent Lineage" backHref={chatHref} />

 	<div class="flex flex-1 flex-col md:flex-row overflow-hidden">
 		<!-- Main tree area -->
@@ -63,12 +45,13 @@
 				<span class="text-xs font-medium text-gray-500 dark:text-gray-400">Agent Types:</span>
 				{#each agentTypeLegend as type (type)}
 					{@const colors = agentTypeColor(type)}
+					{@const colorSet = themeStore.isDark ? colors.dark : colors.light}
 					<span
 						class="inline-flex items-center gap-1.5 rounded-full px-2.5 py-0.5 text-xs font-medium {colors.badge}"
 					>
 						<span
 							class="h-2 w-2 rounded-full"
-							style="background-color: {colors.stroke}"
+							style="background-color: {colorSet.stroke}"
 						></span>
 						{agentTypeLabel(type)}
 					</span>
@@ -79,15 +62,8 @@
 		<!-- Detail panel: overlay on mobile, side panel on desktop -->
 		{#if selectedNode}
 			{@const colors = agentTypeColor(selectedNode.agentType)}
-			<!-- Mobile backdrop -->
-			<div
-				class="fixed inset-0 z-40 bg-black/50 md:hidden"
-				role="button"
-				tabindex="-1"
-				onclick={() => (selectedNode = null)}
-				onkeydown={(e) => { if (e.key === 'Escape') selectedNode = null; }}
-				aria-label="Close detail panel"
-			></div>
+			{@const colorSet = themeStore.isDark ? colors.dark : colors.light}
+			<Backdrop onClose={() => (selectedNode = null)} />
 			<aside
 				class="fixed inset-y-0 right-0 z-50 w-72 shrink-0 border-l border-gray-200 dark:border-gray-700 bg-white dark:bg-gray-900 p-4 overflow-y-auto
 					md:relative md:z-auto"
@@ -117,7 +93,7 @@
 						>
 							<span
 								class="h-2 w-2 rounded-full"
-								style="background-color: {colors.stroke}"
+								style="background-color: {colorSet.stroke}"
 							></span>
 							{agentTypeLabel(selectedNode.agentType)}
 						</span>
@@ -139,12 +115,13 @@
 							<ul class="mt-1 space-y-1">
 								{#each selectedNode.children as child (child.id)}
 									{@const childColors = agentTypeColor(child.agentType)}
+									{@const childColorSet = themeStore.isDark ? childColors.dark : childColors.light}
 									<li
 										class="flex items-center gap-2 rounded-md border border-gray-100 dark:border-gray-700 px-2 py-1.5"
 									>
 										<span
 											class="h-2 w-2 shrink-0 rounded-full"
-											style="background-color: {childColors.stroke}"
+											style="background-color: {childColorSet.stroke}"
 										></span>
 										<span class="truncate font-mono text-xs text-gray-700 dark:text-gray-300">{child.id}</span>
 									</li>
--- a/src/routes/memory/+page.svelte
+++ b/src/routes/memory/+page.svelte
@@ -2,7 +2,7 @@
 	import { resolveRoute } from '$app/paths';
 	import { ResultSource } from '$lib/proto/llm_multiverse/v1/common_pb';
 	import MemoryCandidateCard from '$lib/components/MemoryCandidateCard.svelte';
-	import ThemeToggle from '$lib/components/ThemeToggle.svelte';
+	import PageHeader from '$lib/components/PageHeader.svelte';
 	import { memoryStore } from '$lib/stores/memory.svelte';

 	const chatHref = resolveRoute('/chat');
@@ -39,26 +39,7 @@
 </script>

 <div class="flex h-screen flex-col overflow-hidden bg-gray-50 dark:bg-gray-900">
-	<!-- Header -->
-	<header class="flex items-center justify-between border-b border-gray-200 dark:border-gray-700 bg-white dark:bg-gray-900 px-4 py-3 md:px-6">
-		<div class="flex items-center gap-2 md:gap-4">
-			<!-- eslint-disable svelte/no-navigation-without-resolve -- resolveRoute is resolve; plugin does not recognize the alias -->
-			<a
-				href={chatHref}
-				class="rounded-lg px-2.5 py-1.5 text-sm text-gray-600 dark:text-gray-400 hover:bg-gray-100 dark:hover:bg-gray-700 hover:text-gray-900 dark:hover:text-gray-100"
-			>
-				&larr; Chat
-			</a>
-			<!-- eslint-enable svelte/no-navigation-without-resolve -->
-			<h1 class="text-base md:text-lg font-semibold text-gray-900 dark:text-gray-100">Memory Candidates</h1>
-		</div>
-		<div class="flex items-center gap-2">
-			<ThemeToggle />
-			<span class="hidden sm:inline rounded-md bg-amber-50 dark:bg-amber-900/30 px-2 py-1 text-xs text-amber-700 dark:text-amber-300">
-				Sample Data
-			</span>
-		</div>
-	</header>
+	<PageHeader title="Memory Candidates" backHref={chatHref} />

 	<!-- Filters -->
 	<div class="border-b border-gray-200 dark:border-gray-700 bg-white dark:bg-gray-900 px-4 py-3 md:px-6">
--- a/vite.config.ts
+++ b/vite.config.ts
@@ -3,5 +3,13 @@ import tailwindcss from '@tailwindcss/vite';
 import { defineConfig } from 'vite';

 export default defineConfig({
-	plugins: [tailwindcss(), sveltekit()]
+	plugins: [tailwindcss(), sveltekit()],
+	server: {
+		proxy: {
+			'/llm_multiverse.v1.OrchestratorService': {
+				target: 'http://localhost:8080',
+				changeOrigin: true
+			}
+		}
+	}
 });
Author	SHA1	Message	Date
shahondin1624	6f83311d94	Merge pull request 'feat: display inference statistics in chat UI' (#45 ) from feature/issue-43-inference-stats into main Merge pull request 'feat: display inference statistics in chat UI' (#45)	2026-03-13 14:50:35 +01:00
shahondin1624	36731aac1d	chore: mark issue #43 plan as COMPLETED Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-03-13 14:46:47 +01:00
shahondin1624	3d85cc6b8c	feat: add aria attributes to context utilization progress bar Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-03-13 14:46:35 +01:00
shahondin1624	52eaf661c4	feat: display inference statistics in chat UI Add InferenceStats proto message and InferenceStatsPanel component that displays token counts, throughput, and context window utilization as a collapsible panel below assistant messages after orchestration completes. Closes #43 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-03-13 14:44:12 +01:00
shahondin1624	e93158f670	Merge pull request 'feat: structured artifact rendering, UX improvements' (#44 ) from feat/structured-artifact-passthrough into main	2026-03-12 23:14:28 +01:00
shahondin1624	2c6c961e08	feat: structured artifact rendering, UX improvements - Render structured artifacts from agent results with type-aware formatting: code blocks with syntax highlighting and copy button, terminal-style command output, search result cards, and text findings - Make FinalResult panel collapsible (default collapsed) with scrollable content (max-h-96) to prevent dominating the chat view - Add clickable URL detection in summaries and artifact content - Fix code block contrast for both light and dark mode - Animate progress bar with pulse ring on active step and gradient shimmer on connecting lines - Fix tab-switching bug: use module-level orchestrationStore singleton so orchestration state survives route navigation - Remove sample/demo data seeding and clean up persisted localStorage entries from previous sample sessions - Remove showSampleBadge prop from PageHeader - Regenerate proto types for new Artifact message and ArtifactType enum - Update README project structure (remove deleted data/ directory) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-03-12 23:13:33 +01:00
shahondin1624	cfd338028a	Merge pull request 'fix: prevent session creation loop without backend' (#42 ) from fix/session-creation-loop into main	2026-03-12 16:41:41 +01:00
shahondin1624	a5bc38cb65	fix: prevent session creation loop when running without backend Make getOrCreateSession idempotent by returning the existing active session when called with no ID, instead of always creating a new one. Use untrack() in the $effect to sever SvelteMap dependency so it only re-runs on $page changes. Disable SSR on client-only routes, reduce preload aggressiveness, and skip retries when already disconnected. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-03-12 16:41:08 +01:00
shahondin1624	d7d1d9fd57	Merge pull request 'refactor: code review improvements' (#41 ) from refactor/code-review-improvements into main	2026-03-12 13:48:57 +01:00
shahondin1624	38f5f31b92	refactor: code review improvements — fix bugs, extract shared utilities, add README - Fix reactivity bug: use SvelteMap instead of Map in sessions store - Fix theme listener memory leak: guard against double-init, return cleanup function - Fix transport singleton ignoring different endpoints - Fix form/button type mismatch in MessageInput - Add safer retry validation in chat page - Extract shared utilities: date formatting, session config check, result source config - Extract shared components: Backdrop, PageHeader - Extract orchestration composable from chat page (310→85 lines of script) - Consolidate AuditTimeline switch functions into config record - Move sample data to dedicated module - Add dark mode support for LineageTree SVG colors - Memoize leaf count computation in LineageTree - Update README with usage guide and project structure Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-03-12 13:48:06 +01:00