Merge pull request #730 from diegosouzapw/release/v3.2.3

chore(release): v3.2.3 — Enhancements and Bugfixes

.npmignore

@@ -47,3 +47,12 @@ AGENTS.md
 # Build artifacts (pre-built goes inside app/)
 .next/
 node_modules/
+
+# Ignore large binary files and other build directories
+*.tgz
+*.AppImage
+*.deb
+*.rpm
+electron/
+app/electron/
+app/vscode-extension/

CHANGELOG.md

@@ -4,6 +4,20 @@
 
 ---
 
+## [3.2.3] — 2026-03-29
+
+### ✨ Enhancements & Refactoring
+
+- **Provider Limits Quota UI (#728)** — Normalized quota limit logic and data labeling inside the Limits interface.
+
+### 🐛 Bug Fixes
+
+- **Core Routing Schemas & Leaks** — Expanded `comboStrategySchema` to natively support `fill-first` and `p2c` strategies to unblock complex combo editing natively.
+- **Thinking Tags Extraction (CLI)** — Restructured CLI token responses sanitizer RegEx capturing model reasoning structures inside streams avoiding broken `<thinking>` extractions breaking response text output format.
+- **Strict Format Enforcements** — Hardened pipeline sanitization execution making it universally apply to translation mode targets.
+
+---
+
 ## [3.2.2] — 2026-03-29
 
 ### ✨ New Features

docs/openapi.yaml

@@ -1,7 +1,7 @@
 openapi: 3.1.0
 info:
   title: OmniRoute API
-  version: 3.2.2
+  version: 3.2.3
   description: |
     OmniRoute is a local-first AI API proxy router. It provides an OpenAI-compatible
     endpoint that routes requests to multiple AI providers with load balancing,

open-sse/handlers/chatCore.ts

@@ -531,10 +531,10 @@ export async function handleChatCore({
       connectionId,
       duration: Date.now() - startTime,
       tokens: tokens || {},
-      requestBody: attachLogMeta(body, {
+      requestBody: attachLogMeta((body as Record<string, unknown>) ?? undefined, {
         claudePromptCache: claudeCacheMeta,
       }),
-      responseBody: attachLogMeta(responseBody ?? undefined, {
+      responseBody: attachLogMeta((responseBody as Record<string, unknown>) ?? undefined, {
         claudePromptCache: claudeCacheMeta
           ? {
               applied: claudeCacheMeta.applied,
@@ -1464,8 +1464,9 @@ export async function handleChatCore({
 
     // Sanitize response for OpenAI SDK compatibility
     // Strips non-standard fields (x_groq, usage_breakdown, service_tier, etc.)
-    // Extracts <think> tags into reasoning_content
-    if (sourceFormat === FORMATS.OPENAI) {
+    // Extracts <think> and <thinking> tags into reasoning_content
+    // Target format determines output shape. If we are outputting OpenAI shape or pseudo-OpenAI shape, sanitize.
+    if (targetFormat === FORMATS.OPENAI || targetFormat === FORMATS.OPENAI_RESPONSES) {
       translatedResponse = sanitizeOpenAIResponse(translatedResponse);
     }
 

open-sse/handlers/responseSanitizer.ts

@@ -32,13 +32,12 @@ function toNumber(value: unknown): number | undefined {
   return typeof value === "number" && Number.isFinite(value) ? value : undefined;
 }
 
-// ── Think tag regex ────────────────────────────────────────────────────────
-// Matches <think>...</think> blocks (greedy, dotAll)
-const THINK_TAG_REGEX = /<think>([\s\S]*?)<\/think>/gi;
+// Matches <think>...</think> blocks and <thinking>...</thinking> (greedy, dotAll)
+const THINK_TAG_REGEX = /<(?:think|thinking)>([\s\S]*?)<\/(?:think|thinking)>/gi;
 
-// #638: Collapse runs of 3+ consecutive newlines into \n\n
+// #638, #727: Collapse runs of 2+ consecutive newlines into \n\n
 // Tool call responses from thinking models often accumulate excessive newlines
-const EXCESSIVE_NEWLINES = /\n{3,}/g;
+const EXCESSIVE_NEWLINES = /\n{2,}/g;
 function collapseExcessiveNewlines(text: string): string {
   return text.replace(EXCESSIVE_NEWLINES, "\n\n");
 }

open-sse/utils/stream.ts

@@ -528,31 +528,47 @@ export function createSSEStream(options: StreamOptions = {}) {
               // Content for call log is accumulated only from parsed (above) to avoid double-counting;
               // do not add again from item here.
 
+              // #723, #727: Sanitize intermediate stream chunks if target is OpenAI format loop
+              let itemSanitized: Record<string, unknown> = item;
+              if (targetFormat === FORMATS.OPENAI || targetFormat === FORMATS.OPENAI_RESPONSES) {
+                itemSanitized = sanitizeStreamingChunk(itemSanitized) as Record<string, unknown>;
+
+                // Extract reasoning tags from content if translation generated them
+                const delta = itemSanitized?.choices?.[0]?.delta;
+                if (delta?.content && typeof delta.content === "string") {
+                  const { content, thinking } = extractThinkingFromContent(delta.content);
+                  delta.content = content;
+                  if (thinking && !delta.reasoning_content) {
+                    delta.reasoning_content = thinking;
+                  }
+                }
+              }
+
               // Filter empty chunks
-              if (!hasValuableContent(item, sourceFormat)) {
+              if (!hasValuableContent(itemSanitized, sourceFormat)) {
                 continue; // Skip this empty chunk
               }
 
               // Inject estimated usage if finish chunk has no valid usage
               const isFinishChunk =
-                item.type === "message_delta" || item.choices?.[0]?.finish_reason;
+                itemSanitized.type === "message_delta" || itemSanitized.choices?.[0]?.finish_reason;
               if (
                 state.finishReason &&
                 isFinishChunk &&
-                !hasValidUsage(item.usage) &&
+                !hasValidUsage(itemSanitized.usage) &&
                 totalContentLength > 0
               ) {
                 const estimated = estimateUsage(body, totalContentLength, sourceFormat);
-                item.usage = filterUsageForFormat(estimated, sourceFormat); // Filter + already has buffer
+                itemSanitized.usage = filterUsageForFormat(estimated, sourceFormat); // Filter + already has buffer
                 state.usage = estimated;
               } else if (state.finishReason && isFinishChunk && state.usage) {
                 // Add buffer and filter usage for client (but keep original in state.usage for logging)
                 const buffered = addBufferToUsage(state.usage);
-                item.usage = filterUsageForFormat(buffered, sourceFormat);
+                itemSanitized.usage = filterUsageForFormat(buffered, sourceFormat);
               }
 
-              const output = formatSSE(item, sourceFormat);
-              clientPayloadCollector.push(item);
+              const output = formatSSE(itemSanitized, sourceFormat);
+              clientPayloadCollector.push(itemSanitized);
               reqLogger?.appendConvertedChunk?.(output);
               controller.enqueue(encoder.encode(output));
             }

package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "omniroute",
-  "version": "3.2.2",
+  "version": "3.2.3",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omniroute",
-      "version": "3.2.2",
+      "version": "3.2.3",
       "hasInstallScript": true,
       "license": "MIT",
       "workspaces": [

package.json

@@ -1,6 +1,6 @@
 {
   "name": "omniroute",
-  "version": "3.2.2",
+  "version": "3.2.3",
   "description": "Smart AI Router with auto fallback — route to FREE & cheap models, zero downtime. Works with Cursor, Cline, Claude Desktop, Codex, and any OpenAI-compatible tool.",
   "type": "module",
   "bin": {

src/app/globals.css

@@ -337,3 +337,16 @@ button .material-symbols-outlined,
 .traffic-light.green {
   background: var(--color-traffic-green);
 }
+
+/* ── Mobile Layout Fixes (Issue #659) ── */
+@media (max-width: 768px) {
+  .ant-table-wrapper {
+    overflow-x: auto;
+    -webkit-overflow-scrolling: touch;
+    max-width: 100vw;
+  }
+
+  .ant-table {
+    min-width: 600px; /* Prevent columns from crushing together */
+  }
+}

src/shared/components/layouts/DashboardLayout.tsx

@@ -57,8 +57,8 @@ export default function DashboardLayout({ children }) {
       >
         <Header onMenuClick={() => setSidebarOpen(true)} />
         <MaintenanceBanner />
-        <div className="flex-1 overflow-y-auto custom-scrollbar p-6 lg:p-10">
-          <div className="max-w-7xl mx-auto">
+        <div className="flex-1 overflow-y-auto overflow-x-hidden custom-scrollbar p-4 sm:p-6 lg:p-10">
+          <div className="max-w-7xl mx-auto w-full">
             <Breadcrumbs />
             {children}
           </div>

src/shared/validation/schemas.ts

@@ -78,6 +78,9 @@ const comboStrategySchema = z.enum([
   "cost-optimized",
   "strict-random",
   "auto",
+  "fill-first",
+  // #729 schema fixes for combo edit/save
+  "p2c",
 ]);
 
 const comboRuntimeConfigSchema = z
@@ -884,6 +887,7 @@ export const updateComboSchema = z
     system_message: z.string().max(50000).optional(),
     tool_filter_regex: z.string().max(1000).optional(),
     context_cache_protection: z.boolean().optional(),
+    context_length: z.number().int().min(1000).max(2000000).optional(),
   })
   .superRefine((value, ctx) => {
     if (
@@ -895,7 +899,8 @@ export const updateComboSchema = z
       value.allowedProviders === undefined &&
       value.system_message === undefined &&
       value.tool_filter_regex === undefined &&
-      value.context_cache_protection === undefined
+      value.context_cache_protection === undefined &&
+      value.context_length === undefined
     ) {
       ctx.addIssue({
         code: z.ZodIssueCode.custom,

tests/unit/claude-cache-control-passthrough.test.mjs

@@ -0,0 +1,184 @@
+import { describe, test } from "node:test";
+import assert from "node:assert/strict";
+import { prepareClaudeRequest } from "../../open-sse/translator/helpers/claudeHelper.ts";
+
+describe("Claude cache_control passthrough", () => {
+  test("preserveCacheControl=true preserves cache_control in system blocks", () => {
+    const body = {
+      system: [
+        { type: "text", text: "System prompt 1" },
+        { type: "text", text: "System prompt 2", cache_control: { type: "ephemeral", ttl: "5m" } },
+      ],
+      messages: [],
+    };
+
+    const result = prepareClaudeRequest(body, "claude", true);
+
+    assert.equal(result.system.length, 2);
+    assert.equal(result.system[0].cache_control, undefined);
+    assert.deepEqual(result.system[1].cache_control, { type: "ephemeral", ttl: "5m" });
+  });
+
+  test("preserveCacheControl=false replaces cache_control in system blocks", () => {
+    const body = {
+      system: [
+        { type: "text", text: "System prompt 1" },
+        { type: "text", text: "System prompt 2", cache_control: { type: "ephemeral", ttl: "5m" } },
+      ],
+      messages: [],
+    };
+
+    const result = prepareClaudeRequest(body, "claude", false);
+
+    assert.equal(result.system.length, 2);
+    assert.equal(result.system[0].cache_control, undefined);
+    assert.deepEqual(result.system[1].cache_control, { type: "ephemeral", ttl: "1h" });
+  });
+
+  test("preserveCacheControl=true preserves cache_control in message content blocks", () => {
+    const body = {
+      messages: [
+        {
+          role: "user",
+          content: [
+            { type: "text", text: "User message 1" },
+            { type: "text", text: "User message 2", cache_control: { type: "ephemeral" } },
+          ],
+        },
+        {
+          role: "assistant",
+          content: [
+            {
+              type: "text",
+              text: "Assistant response",
+              cache_control: { type: "ephemeral", ttl: "10m" },
+            },
+          ],
+        },
+      ],
+    };
+
+    const result = prepareClaudeRequest(body, "claude", true);
+
+    assert.equal(result.messages.length, 2);
+    assert.equal(result.messages[0].content[0].cache_control, undefined);
+    assert.deepEqual(result.messages[0].content[1].cache_control, { type: "ephemeral" });
+    assert.deepEqual(result.messages[1].content[0].cache_control, {
+      type: "ephemeral",
+      ttl: "10m",
+    });
+  });
+
+  test("preserveCacheControl=false strips and re-adds cache_control in messages", () => {
+    const body = {
+      messages: [
+        {
+          role: "user",
+          content: [
+            { type: "text", text: "User message 1" },
+            { type: "text", text: "User message 2", cache_control: { type: "ephemeral" } },
+          ],
+        },
+        {
+          role: "assistant",
+          content: [
+            {
+              type: "text",
+              text: "Assistant response",
+              cache_control: { type: "ephemeral", ttl: "10m" },
+            },
+          ],
+        },
+      ],
+    };
+
+    const result = prepareClaudeRequest(body, "claude", false);
+
+    // Original cache_control should be stripped and OmniRoute's strategy applied
+    assert.equal(result.messages.length, 2);
+    // User message should not have cache_control (only second-to-last user gets it)
+    assert.equal(result.messages[0].content[0].cache_control, undefined);
+    assert.equal(result.messages[0].content[1].cache_control, undefined);
+    // Last assistant should have cache_control added by OmniRoute
+    assert.deepEqual(result.messages[1].content[0].cache_control, { type: "ephemeral" });
+  });
+
+  test("preserveCacheControl=true preserves cache_control in tools", () => {
+    const body = {
+      messages: [],
+      tools: [
+        { name: "tool1", description: "Tool 1", input_schema: { type: "object" } },
+        {
+          name: "tool2",
+          description: "Tool 2",
+          input_schema: { type: "object" },
+          cache_control: { type: "ephemeral", ttl: "30m" },
+        },
+      ],
+    };
+
+    const result = prepareClaudeRequest(body, "claude", true);
+
+    assert.equal(result.tools.length, 2);
+    assert.equal(result.tools[0].cache_control, undefined);
+    assert.deepEqual(result.tools[1].cache_control, { type: "ephemeral", ttl: "30m" });
+  });
+
+  test("preserveCacheControl=false replaces cache_control in tools", () => {
+    const body = {
+      messages: [],
+      tools: [
+        { name: "tool1", description: "Tool 1", input_schema: { type: "object" } },
+        {
+          name: "tool2",
+          description: "Tool 2",
+          input_schema: { type: "object" },
+          cache_control: { type: "ephemeral", ttl: "30m" },
+        },
+      ],
+    };
+
+    const result = prepareClaudeRequest(body, "claude", false);
+
+    assert.equal(result.tools.length, 2);
+    assert.equal(result.tools[0].cache_control, undefined);
+    assert.deepEqual(result.tools[1].cache_control, { type: "ephemeral", ttl: "1h" });
+  });
+
+  test("preserveCacheControl=true with Claude Code-style caching", () => {
+    const body = {
+      system: [{ type: "text", text: "System", cache_control: { type: "ephemeral", ttl: "5m" } }],
+      messages: [
+        {
+          role: "user",
+          content: [{ type: "text", text: "Turn 1", cache_control: { type: "ephemeral" } }],
+        },
+        {
+          role: "assistant",
+          content: [{ type: "text", text: "Response 1" }],
+        },
+        {
+          role: "user",
+          content: [{ type: "text", text: "Turn 2" }],
+        },
+      ],
+      tools: [
+        {
+          name: "bash",
+          description: "Execute bash",
+          input_schema: { type: "object" },
+          cache_control: { type: "ephemeral", ttl: "5m" },
+        },
+      ],
+    };
+
+    const result = prepareClaudeRequest(body, "claude", true);
+
+    // All original cache_control should be preserved
+    assert.deepEqual(result.system[0].cache_control, { type: "ephemeral", ttl: "5m" });
+    assert.deepEqual(result.messages[0].content[0].cache_control, { type: "ephemeral" });
+    assert.equal(result.messages[1].content[0].cache_control, undefined);
+    assert.equal(result.messages[2].content[0].cache_control, undefined);
+    assert.deepEqual(result.tools[0].cache_control, { type: "ephemeral", ttl: "5m" });
+  });
+});