Merge branch 'main' into chore/update-claude-fallback-models

Users deploying behind a reverse proxy need a config they can adapt.

The template documents each proxy block and centralizes upstream/subpath values.

It also notes that Nginx location matchers still require literal subpath edits.

docs/nginx-subpath-template.conf

@@ -0,0 +1,218 @@
+# CloudCLI UI Nginx subpath deployment template.
+#
+# Purpose:
+#   Serve CloudCLI UI from a path prefix such as:
+#     http://localhost/ai/
+#     https://example.com/ai/
+#
+# CloudCLI itself still runs at the root of its own HTTP server, for example:
+#     http://127.0.0.1:3001/
+#
+# Nginx receives public requests under /ai, strips that prefix, and forwards the
+# remaining path to CloudCLI. For example:
+#     /ai/                  ->  /
+#     /ai/session/abc       ->  /session/abc
+#     /ai/assets/index.js   ->  /assets/index.js
+#
+# Important Nginx limitation:
+#   Nginx does not allow variables in `location` matchers or `rewrite` regexes.
+#   The configurable variables below are still useful for proxy/filter values,
+#   but if you change /ai to a different subpath, also update every line marked:
+#     [SUBPATH LITERAL]
+#
+# To use a different subpath, replace these literal matchers:
+#     location = /ai
+#     location ^~ /ai/
+#     rewrite ^/ai(?<cloudcli_path>/.*)$ ...
+#
+# Recommended deployment shape:
+#   CloudCLI is the only app using /ai, while root paths /api, /ws, and /shell
+#   are also proxied because the current frontend still calls those endpoints
+#   with root-relative URLs.
+
+worker_processes 1;
+
+events {
+    # Maximum simultaneous connections handled by each worker process.
+    # The default is enough for local testing and small self-hosted deployments.
+    worker_connections 1024;
+}
+
+http {
+    # WebSocket requests include an Upgrade header. Normal HTTP requests do not.
+    # This map gives us the right Connection header for both cases:
+    #   Upgrade present -> "upgrade"
+    #   Upgrade absent  -> "close"
+    map $http_upgrade $connection_upgrade {
+        default upgrade;
+        '' close;
+    }
+
+    server {
+        # For HTTPS deployments, replace this with `listen 443 ssl http2;` and
+        # add ssl_certificate / ssl_certificate_key lines.
+        listen 80 default_server;
+
+        # Use your real hostname in production, for example:
+        #   server_name cloudcli.example.com;
+        server_name localhost 127.0.0.1;
+
+        # ---- User settings -------------------------------------------------
+        #
+        # Public path prefix where users access CloudCLI.
+        # Do not add a trailing slash.
+        #
+        # This variable can be used in redirects and response rewrites. It
+        # cannot be used in `location` matchers, so update the [SUBPATH LITERAL]
+        # lines too if you change it.
+        set $cloudcli_subpath /ai;
+
+        # Private upstream URL where the CloudCLI server is listening.
+        # For a default local server this is usually http://127.0.0.1:3001.
+        set $cloudcli_upstream http://127.0.0.1:3001;
+
+        # Allow larger file uploads through the code editor/project file APIs.
+        client_max_body_size 100m;
+
+        # Redirect /ai to /ai/ so relative browser URL resolution is stable.
+        # [SUBPATH LITERAL] Change `/ai` if you change $cloudcli_subpath.
+        location = /ai {
+            return 301 $cloudcli_subpath/;
+        }
+
+        # Main prefixed CloudCLI UI route.
+        #
+        # [SUBPATH LITERAL] Change `/ai/` and the `^/ai` rewrite if you change
+        # $cloudcli_subpath.
+        location ^~ /ai/ {
+            # Strip the public subpath before proxying. CloudCLI expects to see
+            # root paths such as /, /session/:id, /assets/..., /manifest.json.
+            rewrite ^/ai(?<cloudcli_path>/.*)$ $cloudcli_path break;
+
+            # Forward the rewritten request to the private CloudCLI server.
+            proxy_pass $cloudcli_upstream;
+
+            # Use HTTP/1.1 so WebSocket upgrade requests can pass through if a
+            # browser reaches a socket endpoint under the subpath.
+            proxy_http_version 1.1;
+
+            # Preserve useful request metadata for logs and future app support.
+            proxy_set_header Host $host;
+            proxy_set_header X-Real-IP $remote_addr;
+            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+            proxy_set_header X-Forwarded-Proto $scheme;
+            proxy_set_header X-Forwarded-Prefix $cloudcli_subpath;
+
+            # WebSocket upgrade headers. Harmless for normal HTTP requests.
+            proxy_set_header Upgrade $http_upgrade;
+            proxy_set_header Connection $connection_upgrade;
+
+            # Long-running agent and terminal sessions can stay open for a long
+            # time, so avoid closing idle proxied connections too aggressively.
+            proxy_read_timeout 3600s;
+            proxy_send_timeout 3600s;
+
+            # Disable gzip from the upstream response so sub_filter can inspect
+            # and rewrite HTML/JSON/JS response bodies.
+            proxy_set_header Accept-Encoding "";
+
+            # Rewrite browser-visible root-relative URLs so the runtime can
+            # discover that the app is mounted under the subpath.
+            #
+            # Examples:
+            #   href="/manifest.json" -> href="/ai/manifest.json"
+            #   src="/assets/app.js"  -> src="/ai/assets/app.js"
+            #
+            # These rewrites are important for React Router basename detection.
+            sub_filter_once off;
+            sub_filter_types
+                application/json
+                application/manifest+json
+                application/javascript
+                text/javascript;
+
+            sub_filter 'href="/' 'href="$cloudcli_subpath/';
+            sub_filter 'src="/' 'src="$cloudcli_subpath/';
+
+            # The production HTML and JS register the service worker at /sw.js.
+            # Rewrite that registration so the worker is served from /ai/sw.js.
+            sub_filter "register('/sw.js')" "register('$cloudcli_subpath/sw.js')";
+            sub_filter 'register("/sw.js")' 'register("$cloudcli_subpath/sw.js")';
+
+            # The manifest and service worker contain root-relative paths too.
+            # Rewriting them keeps PWA metadata and cached manifest requests
+            # under the same public subpath.
+            sub_filter '"start_url": "/"' '"start_url": "$cloudcli_subpath/"';
+            sub_filter '"scope": "/"' '"scope": "$cloudcli_subpath/"';
+            sub_filter '"src": "/' '"src": "$cloudcli_subpath/';
+            sub_filter "'/manifest.json'" "'$cloudcli_subpath/manifest.json'";
+            sub_filter '"/manifest.json"' '"$cloudcli_subpath/manifest.json"';
+        }
+
+        # Root API proxy.
+        #
+        # The current CloudCLI frontend calls APIs with root-relative URLs such
+        # as /api/auth/login. Keep this location unless the frontend becomes
+        # fully prefix-aware for API requests.
+        location ^~ /api/ {
+            proxy_pass $cloudcli_upstream;
+
+            proxy_http_version 1.1;
+            proxy_set_header Host $host;
+            proxy_set_header X-Real-IP $remote_addr;
+            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+            proxy_set_header X-Forwarded-Proto $scheme;
+            proxy_set_header X-Forwarded-Prefix $cloudcli_subpath;
+
+            proxy_read_timeout 3600s;
+            proxy_send_timeout 3600s;
+        }
+
+        # Main app WebSocket proxy.
+        #
+        # The frontend opens /ws for realtime chat/session/task updates.
+        location /ws {
+            proxy_pass $cloudcli_upstream;
+
+            proxy_http_version 1.1;
+            proxy_set_header Host $host;
+            proxy_set_header Upgrade $http_upgrade;
+            proxy_set_header Connection $connection_upgrade;
+            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+            proxy_set_header X-Forwarded-Proto $scheme;
+            proxy_set_header X-Forwarded-Prefix $cloudcli_subpath;
+
+            proxy_read_timeout 3600s;
+            proxy_send_timeout 3600s;
+        }
+
+        # Shell WebSocket proxy.
+        #
+        # The browser terminal uses /shell. It requires the same WebSocket
+        # upgrade handling as /ws.
+        location /shell {
+            proxy_pass $cloudcli_upstream;
+
+            proxy_http_version 1.1;
+            proxy_set_header Host $host;
+            proxy_set_header Upgrade $http_upgrade;
+            proxy_set_header Connection $connection_upgrade;
+            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+            proxy_set_header X-Forwarded-Proto $scheme;
+            proxy_set_header X-Forwarded-Prefix $cloudcli_subpath;
+
+            proxy_read_timeout 3600s;
+            proxy_send_timeout 3600s;
+        }
+
+        # Optional health endpoint proxy used by the frontend version checker.
+        location = /health {
+            proxy_pass $cloudcli_upstream;
+
+            proxy_set_header Host $host;
+            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+            proxy_set_header X-Forwarded-Proto $scheme;
+            proxy_set_header X-Forwarded-Prefix $cloudcli_subpath;
+        }
+    }
+}

public/modelConstants.js

@@ -11,7 +11,8 @@ export const CLAUDE_MODELS = {
     {
       value: "default",
       label: "Default (recommended)",
-      description: "Use the default model (currently Opus 4.8 (1M context)) · $5/$25 per Mtok",
+      description:
+        "Use the default model (currently Opus 4.8 (1M context)) · $5/$25 per Mtok",
     },
     {
       value: "sonnet",
@@ -23,6 +24,12 @@ export const CLAUDE_MODELS = {
       label: "Sonnet (1M context)",
       description: "Sonnet 4.6 for long sessions · $3/$15 per Mtok",
     },
+    {
+      value: "opus[1m]",
+      label: "Opus 4.8 (1M context)",
+      description:
+        "Opus 4.8 with 1M context · Most capable for complex work · $5/$25 per Mtok",
+    },
     {
       value: "haiku",
       label: "Haiku",

server/modules/providers/list/claude/claude-models.provider.ts

@@ -18,18 +18,23 @@ export const CLAUDE_FALLBACK_MODELS: ProviderModelsDefinition = {
     {
       value: 'default',
       label: 'Default (recommended)',
-      description: 'Use the default model (currently Opus 4.7 (1M context)) · $5/$25 per Mtok',
+      description: 'Use the default model (currently Opus 4.8 (1M context)) · $5/$25 per Mtok',
     },
     {
-      value: 'sonnet',
-      label: 'Sonnet',
-      description: 'Sonnet 4.6 · Best for everyday tasks · $3/$15 per Mtok',
+      value: "sonnet",
+      label: "Sonnet",
+      description: "Sonnet 4.6 · Best for everyday tasks · $3/$15 per Mtok",
     },
     {
       value: 'sonnet[1m]',
       label: 'Sonnet (1M context)',
       description: 'Sonnet 4.6 for long sessions · $3/$15 per Mtok',
     },
+    {
+      value: 'opus[1m]',
+      label: 'Opus 4.8 (1M context)',
+      description: 'Opus 4.8 with 1M context · Most capable for complex work · $5/$25 per Mtok',
+    },
     {
       value: 'haiku',
       label: 'Haiku',

server/modules/providers/services/provider-models.service.ts

@@ -17,6 +17,7 @@ import { readProviderSessionActiveModelChange } from '@/shared/utils.js';
 
 export const PROVIDER_MODELS_CACHE_TTL_MS = 3 * 24 * 60 * 60 * 1000;
 const PROVIDER_MODELS_CACHE_VERSION = 1;
+const UNCACHED_PROVIDERS = new Set<LLMProvider>(['claude']);
 
 type ProviderModelsServiceDependencies = {
   resolveProvider?: (provider: LLMProvider) => Pick<IProvider, 'models'>;
@@ -232,10 +233,42 @@ export const createProviderModelsService = (dependencies: ProviderModelsServiceD
     return request;
   };
 
+  const loadDirectModels = (
+    provider: LLMProvider,
+  ): Promise<ProviderModelsResult> => {
+    const request = resolveProvider(provider).models.getSupportedModels()
+      .then((models) => {
+        const currentTime = now();
+        return {
+          models,
+          cache: {
+            updatedAt: new Date(currentTime).toISOString(),
+            expiresAt: new Date(currentTime).toISOString(),
+            source: 'fresh' as const,
+          },
+        };
+      })
+      .finally(() => {
+        pendingRequests.delete(provider);
+      });
+
+    pendingRequests.set(provider, request);
+    return request;
+  };
+
   const getProviderModels = async (
     provider: LLMProvider,
     options: ProviderModelsOptions = {},
   ): Promise<ProviderModelsResult> => {
+    if (UNCACHED_PROVIDERS.has(provider)) {
+      const pendingRequest = pendingRequests.get(provider);
+      if (pendingRequest) {
+        return pendingRequest;
+      }
+
+      return loadDirectModels(provider);
+    }
+
     if (options.bypassCache) {
       const pendingRequest = pendingRequests.get(provider);
       if (pendingRequest) {

server/modules/providers/tests/provider-models.service.test.ts

@@ -130,6 +130,37 @@ test('provider models are cached for the three-day ttl', async () => {
   }
 });
 
+test('claude provider models are always loaded directly from the provider', async () => {
+  const tempRoot = await mkdtemp(path.join(os.tmpdir(), 'provider-model-cache-claude-direct-'));
+  let loadCount = 0;
+
+  try {
+    const service = createProviderModelsService({
+      cachePath: path.join(tempRoot, 'models-cache.json'),
+      resolveProvider: (provider) => ({
+        models: {
+          getSupportedModels: async () => {
+            loadCount += 1;
+            return createModels(`${provider}-${loadCount}`);
+          },
+          getCurrentActiveModel: async () => createCurrentActiveModel(`${provider}-active`),
+          changeActiveModel: async (input) => createSessionActiveModelChange(provider, input),
+        },
+      }),
+    });
+
+    const first = await service.getProviderModels('claude');
+    const second = await service.getProviderModels('claude');
+
+    assert.equal(loadCount, 2);
+    assert.equal(first.models.DEFAULT, 'claude-1');
+    assert.equal(second.models.DEFAULT, 'claude-2');
+    assert.equal(second.cache.source, 'fresh');
+  } finally {
+    await rm(tempRoot, { recursive: true, force: true });
+  }
+});
+
 test('provider model cache is persisted across service instances', async () => {
   const tempRoot = await mkdtemp(path.join(os.tmpdir(), 'provider-model-cache-file-'));
   const cachePath = path.join(tempRoot, 'models-cache.json');

server/modules/websocket/services/websocket-auth.service.ts

@@ -20,7 +20,13 @@ export function verifyWebSocketClient(
   dependencies: WebSocketAuthDependencies
 ): boolean {
   const request = info.req as AuthenticatedWebSocketRequest;
-  console.log('WebSocket connection attempt to:', request.url);
+  const upgradeUrl = new URL(request.url ?? '/', 'http://localhost');
+  const loggedUrl = new URL(upgradeUrl);
+  if (loggedUrl.searchParams.has('token')) {
+    loggedUrl.searchParams.set('token', 'REDACTED');
+  }
+
+  console.log('WebSocket connection attempt to:', `${loggedUrl.pathname}${loggedUrl.search}`);
 
   // Platform mode: use the first DB user and skip token checks.
   if (dependencies.isPlatform) {
@@ -36,7 +42,6 @@ export function verifyWebSocketClient(
   }
 
   // OSS mode: read JWT from query string first, then Authorization header.
-  const upgradeUrl = new URL(request.url ?? '/', 'http://localhost');
   const token =
     upgradeUrl.searchParams.get('token') ??
     request.headers.authorization?.split(' ')[1] ??