@nanogpt/private-mode 0.2.13 → 0.2.14

package/README.md

@@ -81,6 +81,23 @@ Supported private model IDs include:
 
 Both GLM 5.2 variants are deployed with a 393,216-token total context limit, with prompt and output sharing that window. GLM 5.3 Private and GLM 5.3 Flash Private each have a 1,048,576-token total context limit and a 131,072-token output ceiling. Kimi K3 Private has a 256,000-token context limit. DeepSeek V4 Flash 0731 Private has a 1,048,576-token total context limit. The model-list extension `context_length` is the combined prompt-and-output window, while `max_output_tokens` is the output ceiling within that same window. These values are not additive.
 
+### Thinking without leaving Private Mode
+
+Keep the same private model ID when enabling thinking. For example:
+
+```json
+{
+  "model": "private/deepseek-v4-flash",
+  "messages": [{ "role": "user", "content": "What is 17 times 19?" }],
+  "reasoning_effort": "high",
+  "stream": true
+}
+```
+
+DeepSeek V4 Flash uses the 0731 release and supports optional thinking. GLM 5.2 and Gemma 4 also support optional thinking. Use `thinking: true` or `enable_thinking: true` to enable it; `thinking: false` disables it for these optional-thinking models. Kimi K3, GLM 5.3, and GLM 5.3 Flash always generate reasoning. GLM 5.3 models accept `low`, `high`, or `max` effort. `reasoning: { "exclude": true }` controls response visibility and does not disable thinking.
+
+Non-streaming reasoning requests may take more than five minutes. The CLI allows the encrypted request to wait for the hosted route's response, with a 30-minute transport limit. The hosted route stops non-streaming generation after at most 29 minutes to leave time before its platform deadline. Use `stream: true` for earlier output, and configure your API client's timeout to allow long generations. Upgrade the local proxy as well as the hosted service to receive both timeout fixes.
+
 Browser requests are locked down by default. The proxy only accepts same-machine clients and rejects browser `Origin` headers that are not explicitly allowed, so a random website or LAN client cannot spend the local `NANOGPT_API_KEY` while the proxy is running. If a local browser app needs to call the proxy directly, allow that exact origin:
 
 ```bash

package/bin/nanogpt-private-mode.js

@@ -4,6 +4,7 @@ import { parseArgs } from 'node:util';
 import { PRIVATE_MODE_PROXY_VERSION } from '../lib/packageInfo.js';
 import { startPrivateModeProxy } from '../lib/server.js';
 import { verifyPrivateModeReceiptCli } from '../lib/verifyReceipt.js';
+import { installPrivateModeTransportTimeouts } from '../lib/transportTimeouts.js';
 
 const { values, positionals } = parseArgs({
   options: {
@@ -69,6 +70,7 @@ if (!apiKey) {
   process.exit(1);
 }
 
+installPrivateModeTransportTimeouts(String(values['api-base']));
 startPrivateModeProxy({
   apiBase: String(values['api-base']),
   apiKey,

package/lib/requestTransforms.js

@@ -271,8 +271,11 @@ function shouldEnableThinking(body, model) {
   ));
 
   return explicitThinking
+    ?? isThinkingEnabled(body.enable_thinking)
     ?? explicitReasoning
     ?? (body.reasoning_effort !== undefined ? reasoningEffort !== 'none' : undefined)
+    ?? isThinkingEnabled(body.chat_template_kwargs?.thinking)
+    ?? isThinkingEnabled(body.chat_template_kwargs?.enable_thinking)
     ?? hasThinkingSuffix;
 }
 
@@ -526,7 +529,6 @@ export function normalizePrivateModelReasoningControls(body, model) {
 
 export function shouldSuppressPrivateModelReasoning(body, model) {
   const isKimiK3 = isPrivateModeKimiK3Model(model);
-  if (!isKimiK3 && model.thinkingMode !== 'deepseek-v4' && model.thinkingMode !== 'glm-5.3') return false;
   const reasoningVisibilityOptOut = body.reasoningOptOut === true ||
     body.exposeReasoning === false ||
     body.materializeReasoning === false ||
@@ -1071,22 +1073,26 @@ export function applyPrivateModelRequestMutations(body, model) {
       ? true
       : shouldEnableThinking(body, model);
     const requestedReasoningEffort = body.reasoning_effort
-      ?? (isPlainObject(body.reasoning) ? body.reasoning.effort : undefined);
+      ?? (isPlainObject(body.reasoning) ? body.reasoning.effort : undefined)
+      ?? body.chat_template_kwargs?.reasoning_effort;
     body.chat_template_kwargs = {
       ...mergeChatTemplateKwargs(body),
       thinking: thinkingEnabled,
     };
 
-    if (thinkingEnabled) {
-      body.chat_template_kwargs.reasoning_effort = model.thinkingMode === 'glm-5.3'
-        ? normalizeGlm53ReasoningEffort(requestedReasoningEffort)
-        : normalizeDeepSeekV4ReasoningEffort(requestedReasoningEffort);
+    if (model.thinkingMode === 'glm-5.3') {
+      // Match the native GLM API and the browser client: low/high/max belongs
+      // at the top level, including on tool turns.
+      body.reasoning_effort = normalizeGlm53ReasoningEffort(requestedReasoningEffort);
+      delete body.chat_template_kwargs.reasoning_effort;
+    } else if (thinkingEnabled) {
+      body.chat_template_kwargs.reasoning_effort = normalizeDeepSeekV4ReasoningEffort(requestedReasoningEffort);
     } else {
       delete body.chat_template_kwargs.reasoning_effort;
     }
 
     delete body.thinking;
-    delete body.reasoning_effort;
+    if (model.thinkingMode !== 'glm-5.3') delete body.reasoning_effort;
   }
 
   stripUnsupportedPrivateTinfoilFields(body);

package/lib/transportTimeouts.js

@@ -0,0 +1,28 @@
+import { Agent, setGlobalDispatcher } from 'undici';
+import { PRIVATE_MODE_STREAM_ROUTE_MAX_DURATION_MS } from './constants.js';
+
+export class PrivateModeDispatcher extends Agent {
+  constructor(apiBase) {
+    super();
+    this.chatUrl = new URL(`${apiBase.replace(/\/+$/, '')}/api/v1/private/tinfoil/v1/chat/completions`);
+  }
+
+  dispatch(options, handler) {
+    // EHBP calls native fetch after encryption, without retaining a custom
+    // dispatcher. Scope the CLI's override to its configured encrypted endpoint.
+    // Override dispatch options, not Agent defaults: fetch supplies its own
+    // five-minute timeout values on every dispatch.
+    if (String(options.origin) === this.chatUrl.origin && options.path === this.chatUrl.pathname) {
+      options = {
+        ...options,
+        headersTimeout: PRIVATE_MODE_STREAM_ROUTE_MAX_DURATION_MS,
+        bodyTimeout: PRIVATE_MODE_STREAM_ROUTE_MAX_DURATION_MS,
+      };
+    }
+    return super.dispatch(options, handler);
+  }
+}
+
+export function installPrivateModeTransportTimeouts(apiBase) {
+  setGlobalDispatcher(new PrivateModeDispatcher(apiBase));
+}

package/models/private-tee.json

@@ -97,6 +97,9 @@
     "billingModel": "TEE/glm-5.3-flash",
     "providerPricingModel": "TEE/glm-5.3-flash",
     "teeTargetModel": "glm-5-3-flash",
+    "teeTargetAliases": [
+      "glm5-3-flash"
+    ],
     "maxInputTokens": 1048576,
     "maxOutputTokens": 131072,
     "thinkingMode": "glm-5.3",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nanogpt/private-mode",
-  "version": "0.2.13",
+  "version": "0.2.14",
   "description": "OpenAI-compatible localhost proxy for NanoGPT Private Mode.",
   "type": "module",
   "publishConfig": {
@@ -23,6 +23,7 @@
     "lib/serverErrorNormalization.js",
     "lib/sseResponsePump.js",
     "lib/statusContract.js",
+    "lib/transportTimeouts.js",
     "lib/verifyReceipt.js",
     "models",
     "README.md"
@@ -32,7 +33,8 @@
   },
   "dependencies": {
     "ai": "6.0.220",
-    "tinfoil": "1.1.12"
+    "tinfoil": "1.1.12",
+    "undici": "6.28.0"
   },
   "engines": {
     "node": ">=22"