teleton 0.8.1 → 0.8.2

package/dist/{chunk-QBHRXLZS.js → chunk-H7MFXJZK.js}

@@ -1,9 +1,9 @@
 import {
   getErrorMessage
-} from "./chunk-XBE4JB7C.js";
+} from "./chunk-3UFPFWYP.js";
 import {
   createLogger
-} from "./chunk-RCMD3U65.js";
+} from "./chunk-NQ6FZKCE.js";
 
 // src/agent/tools/telegram/gifts/get-my-gifts.ts
 import { Type } from "@sinclair/typebox";

package/dist/{chunk-3S4GGLLR.js → chunk-HEDJCLA6.js}

@@ -7,7 +7,7 @@ import {
   EMBEDDING_CACHE_MAX_ENTRIES,
   EMBEDDING_CACHE_TTL_DAYS,
   VOYAGE_BATCH_SIZE
-} from "./chunk-UP55PXFH.js";
+} from "./chunk-C4NKJT2Z.js";
 import {
   fetchWithTimeout
 } from "./chunk-XQUHC3JZ.js";
@@ -16,7 +16,7 @@ import {
 } from "./chunk-EYWNOHMJ.js";
 import {
   createLogger
-} from "./chunk-RCMD3U65.js";
+} from "./chunk-NQ6FZKCE.js";
 
 // src/memory/embeddings/provider.ts
 var NoopEmbeddingProvider = class {
@@ -52,7 +52,7 @@ var AnthropicEmbeddingProvider = class {
     this.dimensions = dims[this.model] ?? 512;
   }
   async embedQuery(text) {
-    const result = await this.embed([text]);
+    const result = await this.embed([text], "query");
     return result[0] ?? [];
   }
   async embedBatch(texts) {
@@ -66,7 +66,7 @@ var AnthropicEmbeddingProvider = class {
     }
     return results;
   }
-  async embed(texts) {
+  async embed(texts, inputType = "document") {
     const response = await fetchWithTimeout(`${this.baseUrl}/embeddings`, {
       method: "POST",
       headers: {
@@ -76,7 +76,7 @@ var AnthropicEmbeddingProvider = class {
       body: JSON.stringify({
         input: texts,
         model: this.model,
-        input_type: "document"
+        input_type: inputType
       })
     });
     if (!response.ok) {
@@ -90,8 +90,8 @@ var AnthropicEmbeddingProvider = class {
 
 // src/memory/embeddings/local.ts
 import { pipeline, env } from "@huggingface/transformers";
-import { join } from "path";
-import { mkdirSync } from "fs";
+import { join, dirname } from "path";
+import { mkdirSync, writeFileSync, renameSync, statSync, unlinkSync } from "fs";
 var log = createLogger("Memory");
 var modelCacheDir = join(TELETON_ROOT, "models");
 try {
@@ -99,6 +99,36 @@ try {
 } catch {
 }
 env.cacheDir = modelCacheDir;
+var MIN_FILE_SIZES = { "onnx/model.onnx": 1e6 };
+function isCacheFileValid(filePath, fileName) {
+  try {
+    return statSync(filePath).size >= (MIN_FILE_SIZES[fileName] ?? 1);
+  } catch {
+    return false;
+  }
+}
+async function ensureModelCached(model) {
+  const files = ["config.json", "tokenizer_config.json", "tokenizer.json", "onnx/model.onnx"];
+  const baseUrl = `https://huggingface.co/${model}/resolve/main`;
+  for (const file of files) {
+    const localPath = join(modelCacheDir, model, file);
+    if (isCacheFileValid(localPath, file)) continue;
+    try {
+      unlinkSync(localPath);
+    } catch {
+    }
+    log.info(`Downloading ${model}/${file}...`);
+    mkdirSync(dirname(localPath), { recursive: true });
+    const res = await fetch(`${baseUrl}/${file}`, { redirect: "follow" });
+    if (!res.ok) {
+      throw new Error(`Failed to download ${model}/${file}: ${res.status} ${res.statusText}`);
+    }
+    const buffer = Buffer.from(await res.arrayBuffer());
+    const tmpPath = localPath + ".tmp";
+    writeFileSync(tmpPath, buffer);
+    renameSync(tmpPath, localPath);
+  }
+}
 var extractorPromise = null;
 function getExtractor(model) {
   if (!extractorPromise) {
@@ -106,7 +136,10 @@ function getExtractor(model) {
     extractorPromise = pipeline("feature-extraction", model, {
       dtype: "fp32",
       // Explicit cache_dir to avoid any env race condition
-      cache_dir: modelCacheDir
+      cache_dir: modelCacheDir,
+      // Prevent pthread_setaffinity_np EINVAL on VPS/containers with restricted CPU sets.
+      // ONNX Runtime skips thread affinity when thread counts are explicit.
+      session_options: { intraOpNumThreads: 1, interOpNumThreads: 1 }
     }).then((ext) => {
       log.info(`Local embedding model ready`);
       return ext;
@@ -131,11 +164,11 @@ var LocalEmbeddingProvider = class {
    * Pre-download and load the model at startup.
    * If loading fails, retries once then marks provider as disabled (FTS5-only).
    * Call this once during app init — avoids retry spam on every message.
-   * @returns true if model loaded successfully, false if fallback to noop
    */
   async warmup() {
     for (let attempt = 1; attempt <= 2; attempt++) {
       try {
+        await ensureModelCached(this.model);
         await getExtractor(this.model);
         return true;
       } catch {
@@ -202,6 +235,16 @@ var CachedEmbeddingProvider = class _CachedEmbeddingProvider {
     this.id = inner.id;
     this.model = inner.model;
     this.dimensions = inner.dimensions;
+    this.stmtCacheGet = db.prepare(
+      `SELECT embedding FROM embedding_cache WHERE hash = ? AND model = ? AND provider = ?`
+    );
+    this.stmtCachePut = db.prepare(
+      `INSERT OR REPLACE INTO embedding_cache (hash, embedding, model, provider, dims, created_at, accessed_at)
+       VALUES (?, ?, ?, ?, ?, unixepoch(), unixepoch())`
+    );
+    this.stmtCacheTouch = db.prepare(
+      `UPDATE embedding_cache SET accessed_at = unixepoch() WHERE hash = ? AND model = ? AND provider = ?`
+    );
   }
   id;
   model;
@@ -210,21 +253,17 @@ var CachedEmbeddingProvider = class _CachedEmbeddingProvider {
   hits = 0;
   misses = 0;
   ops = 0;
+  stmtCacheGet;
+  stmtCachePut;
+  stmtCacheTouch;
   cacheGet(hash) {
-    return this.db.prepare(
-      `SELECT embedding FROM embedding_cache WHERE hash = ? AND model = ? AND provider = ?`
-    ).get(hash, this.model, this.id);
+    return this.stmtCacheGet.get(hash, this.model, this.id);
   }
   cachePut(hash, blob) {
-    this.db.prepare(
-      `INSERT OR REPLACE INTO embedding_cache (hash, embedding, model, provider, dims, created_at, accessed_at)
-         VALUES (?, ?, ?, ?, ?, unixepoch(), unixepoch())`
-    ).run(hash, blob, this.model, this.id, this.dimensions);
+    this.stmtCachePut.run(hash, blob, this.model, this.id, this.dimensions);
   }
   cacheTouch(hash) {
-    this.db.prepare(
-      `UPDATE embedding_cache SET accessed_at = unixepoch() WHERE hash = ? AND model = ? AND provider = ?`
-    ).run(hash, this.model, this.id);
+    this.stmtCacheTouch.run(hash, this.model, this.id);
   }
   async warmup() {
     return this.inner.warmup?.() ?? true;

package/dist/{chunk-AYWEJCDB.js → chunk-J73TA3UM.js}

@@ -2,17 +2,17 @@ import {
   getClaudeCodeApiKey,
   getProviderMetadata,
   refreshClaudeCodeApiKey
-} from "./chunk-PHSAHTK4.js";
+} from "./chunk-YOSUPUAJ.js";
 import {
   appendToTranscript,
   readTranscript
-} from "./chunk-QV2GLOTK.js";
+} from "./chunk-LC4TV3KL.js";
 import {
   fetchWithTimeout
 } from "./chunk-XQUHC3JZ.js";
 import {
   createLogger
-} from "./chunk-RCMD3U65.js";
+} from "./chunk-NQ6FZKCE.js";
 
 // src/agent/client.ts
 import {
@@ -262,7 +262,7 @@ async function chatWithContext(config, options) {
     systemPrompt = "/no_think\n" + systemPrompt;
     if (tools && tools.length > 0) {
       cocoonAllowedTools = new Set(tools.map((t) => t.name));
-      const { injectToolsIntoSystemPrompt } = await import("./tool-adapter-IH5VGBOO.js");
+      const { injectToolsIntoSystemPrompt } = await import("./tool-adapter-IVX2XQJE.js");
       systemPrompt = injectToolsIntoSystemPrompt(systemPrompt, tools);
       tools = void 0;
     }
@@ -281,13 +281,13 @@ async function chatWithContext(config, options) {
     cacheRetention: "long"
   };
   if (isCocoon) {
-    const { stripCocoonPayload } = await import("./tool-adapter-IH5VGBOO.js");
+    const { stripCocoonPayload } = await import("./tool-adapter-IVX2XQJE.js");
     completeOptions.onPayload = stripCocoonPayload;
   }
   let response = await complete(model, context, completeOptions);
   if (provider === "claude-code" && response.stopReason === "error" && response.errorMessage && (response.errorMessage.includes("401") || response.errorMessage.toLowerCase().includes("unauthorized"))) {
     log.warn("Claude Code token rejected (401), refreshing credentials and retrying...");
-    const refreshedKey = refreshClaudeCodeApiKey();
+    const refreshedKey = await refreshClaudeCodeApiKey();
     if (refreshedKey) {
       completeOptions.apiKey = refreshedKey;
       response = await complete(model, context, completeOptions);
@@ -296,7 +296,7 @@ async function chatWithContext(config, options) {
   if (isCocoon) {
     const textBlock = response.content.find((b) => b.type === "text");
     if (textBlock?.type === "text" && textBlock.text.includes("<tool_call>")) {
-      const { parseToolCallsFromText, extractPlainText } = await import("./tool-adapter-IH5VGBOO.js");
+      const { parseToolCallsFromText, extractPlainText } = await import("./tool-adapter-IVX2XQJE.js");
       const syntheticCalls = parseToolCallsFromText(textBlock.text, cocoonAllowedTools);
       if (syntheticCalls.length > 0) {
         const plainText = extractPlainText(textBlock.text);

package/dist/{chunk-QV2GLOTK.js → chunk-LC4TV3KL.js}

@@ -3,7 +3,7 @@ import {
 } from "./chunk-EYWNOHMJ.js";
 import {
   createLogger
-} from "./chunk-RCMD3U65.js";
+} from "./chunk-NQ6FZKCE.js";
 
 // src/session/transcript.ts
 import {

package/dist/{chunk-RCMD3U65.js → chunk-NQ6FZKCE.js}

@@ -88,12 +88,20 @@ var rootLogger = pino(
       paths: [
         "apiKey",
         "api_key",
+        "api_hash",
+        "accessToken",
+        "access_token",
+        "refresh_token",
         "password",
         "secret",
         "token",
         "mnemonic",
         "*.apiKey",
         "*.api_key",
+        "*.api_hash",
+        "*.accessToken",
+        "*.access_token",
+        "*.refresh_token",
         "*.password",
         "*.secret",
         "*.token",
@@ -120,6 +128,11 @@ function setLogLevel(level) {
   const streams = multiStream.streams;
   if (Array.isArray(streams) && streams[0]) {
     streams[0].level = pino.levels.values[level] ?? 30;
+  } else {
+    process.stderr.write(
+      `[Logger] setLogLevel: pino multistream internal API changed, stdout level not updated
+`
+    );
   }
   _verbose = level === "debug" || level === "trace";
 }