llm-switcher 1.2.9 → 1.2.10

package/CHANGELOG.md

@@ -1,5 +1,10 @@
 # Changelog — LLM Switcher
 
+## Release 1.2.10
+
+- **Bifrost in the log:** The log line and the request inspector showed `anthropic -> openai-chat` for a Bifrost request, because the gateway logged the profile's `outFormat` before it checked Bifrost. Now both show `bifrost`.
+- **Bifrost and the contract lab:** A sampled Bifrost exchange went to intact labeled with the profile's `outFormat`. intact then compared an Anthropic request with an OpenAI conversion that never ran. Now the label is `anthropic`.
+
 ## Release 1.2.9
 
 - **Bifrost:** Claude Code that reaches a Claude Code account on intact now goes through unchanged. Only the key changes. Before, a `convert` profile, or a `hybrid` profile with a `claude/...` model, changed the request to OpenAI Chat. The `direct` route also ran the healer and `thinkingMode`. Now the gateway asks intact for `bifrost_ua` of the mapped model. If the `User-Agent` of the client starts with this value, the gateway sends every client header, the body bytes, and the query string. There is no setting. intact 0.1.10 or newer gives `bifrost_ua`.

package/README.md

@@ -134,6 +134,7 @@ LLM Switcher acts as a transparent man-in-the-middle without ever touching clien
 - **Bi-Directional Protocol Conversion:**
   - **4 Client Inbound Formats:** Anthropic Messages, OpenAI Chat Completions, Codex Responses API, Vertex `generateContent`.
   - **3 Upstream Outbound Formats:** OpenAI Chat, Anthropic Native, Vertex Native.
+- **Bifrost:** Claude Code on a Claude Code account of intact goes through unchanged. Only the key changes. No conversion, no healer, no filter. The gateway turns it on by itself. [Bifrost](#bifrost-claude-code-to-a-claude-code-account-on-intact)
 - **Multi-Active CLI Routing:** Run Claude Code on Profile A, Codex on Profile B, and Cursor on Profile C simultaneously on a single gateway instance.
 - **Deep Thinking & Reasoning Extraction:** Tested on 48 live response combinations. Accurately extracts `reasoning_content`, `<think>` tags, Vertex `thought` parts, and signatures into native `thinking_delta` blocks.
 - **Edge Healer Engine (Anti-Collision for Token Optimizers):**

package/README.vi.md

@@ -134,6 +134,7 @@ LLM Switcher hoạt động như một lớp trung gian mạng trong suốt (tra
 - **Chuyển đổi Giao thức 2 Chiều Đối xứng:**
   - **4 Định dạng đầu vào (Client):** Anthropic Messages, OpenAI Chat Completions, Codex Responses API, Vertex `generateContent`.
   - **3 Định dạng đầu ra (Upstream):** OpenAI Chat, Anthropic Native, Vertex Native.
+- **Bifrost:** Claude Code tới tài khoản Claude Code trên intact đi nguyên vẹn. Chỉ key thay đổi. Không chuyển định dạng, không healer, không lọc. Gateway tự bật. [Bifrost](#bifrost-claude-code-tới-tài-khoản-claude-code-trên-intact)
 - **Multi-Active CLI Routing:** Kích hoạt cùng lúc Claude Code dùng Profile A, Codex dùng Profile B, Cursor dùng Profile C trên cùng 1 gateway mà không tranh chấp cấu hình.
 - **Trích xuất Thinking & Reasoning Chuyên sâu:** Kiểm chứng thực tế qua 48 tổ hợp mẫu response live. Tự động bóc tách `reasoning_content`, thẻ `<think>`, các block `thought` của Vertex và thought signature thành các `thinking_delta` chuẩn của Anthropic.
 - **Edge Healer Engine (Tự chữa lành tin nhắn):**

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "llm-switcher",
-  "version": "1.2.9",
+  "version": "1.2.10",
   "description": "Zero-dependency multi-protocol edge gateway & provider switcher for Claude Code and Codex, with OpenAI-compatible, Anthropic and Vertex upstreams",
   "keywords": [
     "llm",

package/proxy.mjs

@@ -775,10 +775,12 @@ async function handleConvert(clientFormat, req, res, bodyBuffer, opts = {}) {
   const requestPreview = previewOf(ir);
   const requestedModel = ir.model || payload.model || '';
   const mappedModel = mapModel(requestedModel, profile, clientFormat);
+  const bifrost = clientFormat === 'anthropic' && await crossesBifrost(req, profile, mappedModel);
   const outFormat = resolveOutFormat(profile, mappedModel);
-  console.log(`[llm-switcher] ${clientFormat} -> ${outFormat} "${requestedModel}" -> "${mappedModel}" [${profile.name || profileKey}]`);
+  const route = bifrost ? 'bifrost' : outFormat;
+  console.log(`[llm-switcher] ${clientFormat} -> ${route} "${requestedModel}" -> "${mappedModel}" [${profile.name || profileKey}]`);
 
-  const logBase = { clientFormat, outFormat, profile: profileKey, model: mappedModel, stream: ir.stream, requestPreview };
+  const logBase = { clientFormat, outFormat: route, profile: profileKey, model: mappedModel, stream: ir.stream, requestPreview };
   const log = (extra) => logInspection({ ...logBase, duration: Date.now() - reqStartTime, tokens: { prompt: 0, completion: 0 }, ...extra });
 
   // Contract lab: a sampled request carries a trace id to intact, and the bytes this gateway
@@ -801,14 +803,14 @@ async function handleConvert(clientFormat, req, res, bodyBuffer, opts = {}) {
   try {
     // Bifrost: the provider's own client reaching its own account through intact. Bytes and headers
     // go as sent, only the key changes: no healer, no thinkingMode, no conversion.
-    if (clientFormat === 'anthropic' && await crossesBifrost(req, profile, mappedModel)) {
+    if (bifrost) {
       const { url } = upstreamEndpoint(profile, 'anthropic', mappedModel, ir.stream, req);
       const headers = bifrostHeaders(req, profile);
       if (traceId) headers['x-intact-trace'] = traceId;
       try {
         const r = await forwardAnthropicDirect(res, payload, bodyBuffer, withClientQuery(url, req), headers, mappedModel, ac.signal, profile, true);
         answered = !r.error && r.status >= 200 && r.status < 300;
-        log({ outFormat: 'anthropic', status: r.status, tokens: r.tokens, responsePreview: '(bifrost)', error: r.error || undefined });
+        log({ status: r.status, tokens: r.tokens, responsePreview: '(bifrost)', error: r.error || undefined });
       } catch (err) {
         if (ac.signal.aborted) return log({ status: 499, error: 'client disconnected' });
         console.error(`[${profileKey}] Bifrost forward error:`, err.message);
@@ -936,7 +938,7 @@ async function handleConvert(clientFormat, req, res, bodyBuffer, opts = {}) {
         toolResponse: halfTap?.text() || '',
         toolVersion: toolVersionFromUA(req.headers['user-agent']),
         inFormat: clientFormat,
-        outFormat,
+        outFormat: bifrost ? 'anthropic' : outFormat, // Bifrost converts nothing
         openerKey: profile.apiKey || ''
       });
     }

package/tests/gateway.e2e.test.mjs

@@ -611,6 +611,9 @@ test('Bifrost: Claude Code to a Claude Code account on intact changes only the k
   assert.equal(up.headers['anthropic-dangerous-direct-browser-access'], 'true');
   assert.equal(up.headers['x-app'], 'cli');
   assert.equal(up.raw, JSON.stringify({ ...body, model: 'claude/claude-opus-5' }));
+  // The inspector and the log name the route that ran, not the profile's outFormat.
+  const logs = (await (await fetch(url('/api/logs'), { headers: withToken('/api/logs', {}) })).json()).logs;
+  assert.equal(logs[0].outFormat, 'bifrost');
 });
 
 test('Bifrost stays off for a model that is not a Claude Code account', async () => {