@nanogpt/private-mode 0.1.1 → 0.1.3

package/README.md

@@ -1,6 +1,6 @@
 # NanoGPT Private Mode
 
-OpenAI-compatible localhost proxy for NanoGPT Private Mode with Tinfoil-backed models.
+OpenAI-compatible localhost proxy for NanoGPT Private Mode with supported TEE models.
 
 ```bash
 NANOGPT_API_KEY=sk-your-key npx @nanogpt/private-mode
@@ -28,17 +28,17 @@ const response = await client.chat.completions.create({
 });
 ```
 
-The local proxy verifies Tinfoil attestation, encrypts request bodies with EHBP, sends ciphertext through NanoGPT, decrypts encrypted responses locally, and returns normal OpenAI JSON to the calling app.
+The local proxy verifies TEE attestation, encrypts request bodies with EHBP, sends ciphertext through NanoGPT, decrypts encrypted responses locally, and returns normal OpenAI JSON to the calling app.
 
-NanoGPT can see account identity, selected private model, selected Tinfoil enclave metadata, timing, sizes, status, and usage metadata. NanoGPT cannot read the prompt or completion body for supported private models.
+NanoGPT can see account identity, selected private model, selected TEE target metadata, timing, sizes, status, and usage metadata. NanoGPT cannot read the prompt or completion body for supported private models.
 
-NanoGPT's web app can also use these models without running this local proxy. Select an eligible Tinfoil-backed `TEE/*` model and use the Private Mode control in the model picker. This package is for API clients, CLIs, agents, and other OpenAI-compatible tools.
+NanoGPT's web app can also use these models without running this local proxy. Select an eligible Private Mode model and use the Private Mode control in the model picker. This package is for API clients, CLIs, agents, and other OpenAI-compatible tools.
 
 The web-app toggle is narrower than the proxy path in v1. It supports text chat and model settings, and disables attachments, web search, URL-scraped content, project tools, multi-model chat, Context Memory injection, quick replies, and automatic title generation for private turns.
 
 In the hosted web app, decrypted Private Mode turns remain in local browser history. Cloud conversation sync is blocked for Private Mode chats unless password-based end-to-end sync is enabled; the recoverable default sync mode is not used for those chats.
 
-Private Mode also requires enough NanoGPT balance and API-key spend-limit headroom before dispatch. NanoGPT cannot read encrypted prompts to count tokens, so the reserve uses the encrypted request size when available and otherwise falls back to a conservative model estimate capped at 32,768 output tokens by default (`NANOGPT_PRIVATE_TINFOIL_RESERVE_MAX_OUTPUT_TOKENS`). The final charge is still based on Tinfoil usage metrics.
+Private Mode also requires enough NanoGPT balance and API-key spend-limit headroom before dispatch. For streaming calls, NanoGPT may precharge a conservative reserve before dispatch and refund unused balance after verified usage metadata is available.
 
 Useful local checks:
 

package/lib/requestTransforms.js

@@ -6,6 +6,32 @@ const ASSISTANT_REASONING_MESSAGE_FIELDS = [
   'reasoning_content',
   'reasoning_details',
 ];
+const PRIVATE_TINFOIL_CHAT_COMPLETION_BODY_FIELDS = new Set([
+  'chat_template_kwargs',
+  'frequency_penalty',
+  'function_call',
+  'functions',
+  'logit_bias',
+  'logprobs',
+  'max_tokens',
+  'messages',
+  'model',
+  'n',
+  'parallel_tool_calls',
+  'presence_penalty',
+  'reasoning_effort',
+  'response_format',
+  'seed',
+  'stop',
+  'stream',
+  'stream_options',
+  'temperature',
+  'tool_choice',
+  'tools',
+  'top_logprobs',
+  'top_p',
+  'user',
+]);
 
 // Keep request mutation behavior in sync with lib/privateMode/tinfoilBrowserClient.ts
 // for explicit fields. The local proxy preserves OpenAI's non-streaming default.
@@ -79,6 +105,14 @@ function mergeChatTemplateKwargs(body) {
   return isPlainObject(body.chat_template_kwargs) ? body.chat_template_kwargs : {};
 }
 
+function stripUnsupportedPrivateTinfoilFields(body) {
+  for (const key of Object.keys(body)) {
+    if (!PRIVATE_TINFOIL_CHAT_COMPLETION_BODY_FIELDS.has(key)) {
+      delete body[key];
+    }
+  }
+}
+
 function stripPrivateModeReasoningBlocks(value) {
   return value
     .replace(/◁think▷/g, '<think>')
@@ -178,9 +212,17 @@ function applyTinfoilCompatibilityMutations(body, model) {
 export function applyPrivateModelRequestMutations(body, model) {
   stripPrivateModeReasoningFromMessages(body);
 
+  if (body.max_tokens === undefined && body.max_completion_tokens !== undefined) {
+    body.max_tokens = body.max_completion_tokens;
+  }
+  if (body.max_tokens === undefined && body.maxCompletionTokens !== undefined) {
+    body.max_tokens = body.maxCompletionTokens;
+  }
   if (body.max_tokens === undefined && body.maxTokens !== undefined) {
     body.max_tokens = body.maxTokens;
   }
+  delete body.max_completion_tokens;
+  delete body.maxCompletionTokens;
   delete body.maxTokens;
 
   body.model = model.upstreamModel;
@@ -192,6 +234,8 @@ export function applyPrivateModelRequestMutations(body, model) {
       ...(body.stream_options && typeof body.stream_options === 'object' ? body.stream_options : {}),
       include_usage: true,
     };
+  } else {
+    delete body.stream_options;
   }
   applyTinfoilCompatibilityMutations(body, model);
 
@@ -202,10 +246,7 @@ export function applyPrivateModelRequestMutations(body, model) {
     };
     delete body.thinking;
     delete body.reasoning_effort;
-    return body;
-  }
-
-  if (model.thinkingMode === 'deepseek-v4-pro') {
+  } else if (model.thinkingMode === 'deepseek-v4-pro') {
     const thinkingEnabled = shouldEnableThinking(body, model);
     body.chat_template_kwargs = {
       ...mergeChatTemplateKwargs(body),
@@ -223,5 +264,6 @@ export function applyPrivateModelRequestMutations(body, model) {
     delete body.reasoning_effort;
   }
 
+  stripUnsupportedPrivateTinfoilFields(body);
   return body;
 }

package/lib/server.js

@@ -6,10 +6,14 @@ import {
   getCorsHeadersForRequest,
 } from './originPolicy.js';
 import { applyPrivateModelRequestMutations } from './requestTransforms.js';
+import {
+  normalizePrivateModeUpstreamErrorMessage,
+  readErrorMessage,
+} from './serverErrorNormalization.js';
 import { buildPrivateModeStatusContract } from './statusContract.js';
 
 const MODELS = JSON.parse(
-  readFileSync(new URL('../models/tinfoil.json', import.meta.url), 'utf8'),
+  readFileSync(new URL('../models/private-tee.json', import.meta.url), 'utf8'),
 );
 
 function readMaxBodyBytes() {
@@ -66,7 +70,7 @@ function openAIModelList() {
 export function privateModeStatus(apiBase, secureState, localBase, originPolicy) {
   return {
     ok: true,
-    mode: 'tinfoil',
+    mode: 'private_tee',
     apiBase,
     local_base_url: localBase,
     models_path: '/v1/models',
@@ -302,9 +306,10 @@ async function handleChatCompletion({ apiBase, apiKey, secureState, req, res, co
   } catch (error) {
     responseComplete = true;
     if (res.destroyed) return;
+    const message = error instanceof Error ? error.message : String(error);
     jsonResponse(res, 502, {
       error: {
-        message: `Private Mode request failed: ${error instanceof Error ? error.message : String(error)}`,
+        message: `Private Mode request failed: ${normalizePrivateModeUpstreamErrorMessage(message, message)}`,
         type: 'api_error',
         code: 'private_mode_request_failed',
       },
@@ -321,6 +326,20 @@ async function handleChatCompletion({ apiBase, apiKey, secureState, req, res, co
   const privateMode = response.headers.get('x-nanogpt-private-mode');
   if (privateMode) headers['x-nanogpt-private-mode'] = privateMode;
 
+  if (!response.ok) {
+    responseComplete = true;
+    const errorHeaders = { ...headers };
+    delete errorHeaders['content-type'];
+    jsonResponse(res, response.status, {
+      error: {
+        message: await readErrorMessage(response, `Private Mode request failed with HTTP ${response.status}`),
+        type: 'api_error',
+        code: 'private_mode_upstream_failed',
+      },
+    }, errorHeaders);
+    return;
+  }
+
   if (privateStreamRequested && response.body) {
     res.writeHead(response.status, headers);
     const reader = response.body.getReader();
@@ -484,7 +503,7 @@ Allowed browser origins: ${originPolicy.allowedOrigins.join(', ')}
 
 Request and response bodies are encrypted after they leave this local proxy.
 NanoGPT can see your account, private model header, timing, sizes, status, and usage metadata.
-The verified Tinfoil enclave and this local proxy can see plaintext.
+The verified TEE target and this local proxy can see plaintext.
 `);
   }
 

package/lib/serverErrorNormalization.js

@@ -0,0 +1,45 @@
+// Keep upstream error normalization in sync with lib/privateMode/tinfoilBrowserClient.ts.
+const UPSTREAM_OBJECT_SHAPE_ERROR_PATTERN = /object has no attribute ['"]get['"]/i;
+const MISSING_EHBP_RESPONSE_NONCE_PATTERN = /missing\s+ehbp-response-nonce\s+header/i;
+
+export function normalizePrivateModeUpstreamErrorMessage(message, fallback) {
+  if (UPSTREAM_OBJECT_SHAPE_ERROR_PATTERN.test(message)) {
+    return 'Private Mode upstream returned an internal error while processing request metadata. Retry or choose another Private Mode model.';
+  }
+  if (MISSING_EHBP_RESPONSE_NONCE_PATTERN.test(message)) {
+    return 'Private Mode did not receive an encrypted response from Tinfoil. The private request may have been rejected before completion.';
+  }
+  return message || fallback;
+}
+
+export async function readErrorMessage(response, fallback) {
+  let message = fallback;
+  try {
+    const data = await response.clone().json();
+    if (data && typeof data === 'object' && !Array.isArray(data)) {
+      const error = data.error && typeof data.error === 'object' && !Array.isArray(data.error)
+        ? data.error
+        : {};
+      message = (
+        typeof error.message === 'string' && error.message.trim()
+          ? error.message
+          : typeof data.message === 'string' && data.message.trim()
+            ? data.message
+            : fallback
+      );
+    } else if (typeof data === 'string' && data.trim()) {
+      message = data;
+    } else if (typeof data === 'number' || typeof data === 'boolean') {
+      message = String(data);
+    }
+  } catch {
+    try {
+      const text = await response.text();
+      message = text || fallback;
+    } catch {
+      message = fallback;
+    }
+  }
+
+  return normalizePrivateModeUpstreamErrorMessage(message, fallback);
+}

package/lib/statusContract.js

@@ -22,7 +22,7 @@ export const PRIVATE_MODE_FRONTEND_DISABLED_FEATURES = Object.freeze([
 export const PRIVATE_MODE_NANOGPT_VISIBLE = Object.freeze([
   'account',
   'selected_private_model',
-  'selected_tinfoil_enclave',
+  'selected_tee_target',
   'request_timing',
   'request_size',
   'response_size',
@@ -40,7 +40,7 @@ export const PRIVATE_MODE_ENCRYPTED_FROM_NANOGPT = Object.freeze([
 
 export const PRIVATE_MODE_PLAINTEXT_VISIBLE_TO = Object.freeze([
   'local_proxy',
-  'verified_tinfoil_enclave',
+  'verified_tee_target',
 ]);
 
 export function buildPrivateModeStatusContract() {

package/models/{tinfoil.json → private-tee.json}

@@ -4,9 +4,9 @@
     "name": "Kimi K2.6 Private",
     "upstreamModel": "kimi-k2-6",
     "billingModel": "TEE/kimi-k2-6",
-    "tinfoilEnclaveModel": "kimi-k2-6",
+    "teeTargetModel": "kimi-k2-6",
     "created": 1764547200,
-    "ownedBy": "nanogpt-tinfoil",
+    "ownedBy": "nanogpt-private-mode",
     "aliases": ["private/kimi-k2.6", "TEE/kimi-k2-6", "TEE/kimi-k2.6"]
   },
   {
@@ -14,9 +14,9 @@
     "name": "GPT OSS 120B Private",
     "upstreamModel": "gpt-oss-120b",
     "billingModel": "TEE/gpt-oss-120b",
-    "tinfoilEnclaveModel": "gpt-oss-120b",
+    "teeTargetModel": "gpt-oss-120b",
     "created": 1764547200,
-    "ownedBy": "nanogpt-tinfoil",
+    "ownedBy": "nanogpt-private-mode",
     "aliases": ["TEE/gpt-oss-120b", "phala/gpt-oss-120b"]
   },
   {
@@ -24,40 +24,30 @@
     "name": "Llama 3.3 70B Private",
     "upstreamModel": "llama3-3-70b",
     "billingModel": "TEE/llama3-3-70b",
-    "tinfoilEnclaveModel": "llama3-3-70b",
+    "teeTargetModel": "llama3-3-70b",
     "created": 1764547200,
-    "ownedBy": "nanogpt-tinfoil",
+    "ownedBy": "nanogpt-private-mode",
     "aliases": ["private/llama-3.3-70b", "TEE/llama3-3-70b"]
   },
   {
-    "id": "private/glm-5-1",
-    "name": "GLM 5.1 Private",
-    "upstreamModel": "glm-5-1",
-    "billingModel": "TEE/glm-5-1",
-    "tinfoilEnclaveModel": "glm-5-1",
-    "created": 1764547200,
-    "ownedBy": "nanogpt-tinfoil",
-    "aliases": ["private/glm-5.1", "TEE/glm-5-1", "TEE/glm-5.1"]
-  },
-  {
-    "id": "private/glm-5-1-thinking",
-    "name": "GLM 5.1 Thinking Private",
-    "upstreamModel": "glm-5-1",
-    "billingModel": "TEE/glm-5-1-thinking",
-    "tinfoilEnclaveModel": "glm-5-1",
-    "created": 1764547200,
-    "ownedBy": "nanogpt-tinfoil",
-    "aliases": ["private/glm-5.1-thinking", "TEE/glm-5-1-thinking", "TEE/glm-5.1-thinking"]
+    "id": "private/glm-5-2",
+    "name": "GLM 5.2 Private",
+    "upstreamModel": "glm-5-2",
+    "billingModel": "TEE/glm-5-2",
+    "teeTargetModel": "glm-5-2",
+    "created": 1781827200,
+    "ownedBy": "nanogpt-private-mode",
+    "aliases": ["private/glm-5.2", "TEE/glm-5-2", "TEE/glm-5.2"]
   },
   {
     "id": "private/gemma4-31b",
     "name": "Gemma 4 31B Private",
     "upstreamModel": "gemma4-31b",
     "billingModel": "TEE/gemma4-31b",
-    "tinfoilEnclaveModel": "gemma4-31b",
+    "teeTargetModel": "gemma4-31b",
     "thinkingMode": "gemma",
     "created": 1764547200,
-    "ownedBy": "nanogpt-tinfoil",
+    "ownedBy": "nanogpt-private-mode",
     "aliases": ["TEE/gemma4-31b", "gemma4-31b"]
   },
   {
@@ -65,10 +55,10 @@
     "name": "Gemma 4 31B Thinking Private",
     "upstreamModel": "gemma4-31b",
     "billingModel": "TEE/gemma4-31b:thinking",
-    "tinfoilEnclaveModel": "gemma4-31b",
+    "teeTargetModel": "gemma4-31b",
     "thinkingMode": "gemma",
     "created": 1764547200,
-    "ownedBy": "nanogpt-tinfoil",
+    "ownedBy": "nanogpt-private-mode",
     "aliases": ["TEE/gemma4-31b:thinking", "gemma4-31b:thinking"]
   },
   {
@@ -76,10 +66,10 @@
     "name": "DeepSeek V4 Pro Private",
     "upstreamModel": "deepseek-v4-pro",
     "billingModel": "TEE/deepseek-v4-pro",
-    "tinfoilEnclaveModel": "deepseek-v4-pro",
+    "teeTargetModel": "deepseek-v4-pro",
     "thinkingMode": "deepseek-v4-pro",
     "created": 1764547200,
-    "ownedBy": "nanogpt-tinfoil",
+    "ownedBy": "nanogpt-private-mode",
     "aliases": ["TEE/deepseek-v4-pro", "deepseek-v4-pro"]
   },
   {
@@ -87,10 +77,10 @@
     "name": "DeepSeek V4 Pro Thinking Private",
     "upstreamModel": "deepseek-v4-pro",
     "billingModel": "TEE/deepseek-v4-pro:thinking",
-    "tinfoilEnclaveModel": "deepseek-v4-pro",
+    "teeTargetModel": "deepseek-v4-pro",
     "thinkingMode": "deepseek-v4-pro",
     "created": 1764547200,
-    "ownedBy": "nanogpt-tinfoil",
+    "ownedBy": "nanogpt-private-mode",
     "aliases": ["TEE/deepseek-v4-pro:thinking", "deepseek-v4-pro:thinking"]
   }
 ]

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nanogpt/private-mode",
-  "version": "0.1.1",
+  "version": "0.1.3",
   "description": "OpenAI-compatible localhost proxy for NanoGPT Private Mode.",
   "type": "module",
   "publishConfig": {
@@ -15,6 +15,7 @@
     "lib/originPolicy.js",
     "lib/requestTransforms.js",
     "lib/server.js",
+    "lib/serverErrorNormalization.js",
     "lib/statusContract.js",
     "lib/verifyReceipt.js",
     "models",