@qvac/sdk 0.13.5 → 0.14.0

package/dist/server/utils/completion-normalizer.js

@@ -25,14 +25,18 @@ const DIALECT_SPECS = {
         ],
         contentFrames: [
             { open: "<|channel|>final<|message|>", close: "<|return|>" },
+            { open: "<|channel|>assistant<|message|>", close: "<|end|>" },
         ],
-        // Role headers between channel frames + stray `<|return|>` closes.
+        // Role headers and stray Harmony control tokens between frames.
         dropTokens: [
+            "<|message|>",
+            "<|end|>",
             "<|start|>assistant",
             "<|start|>system",
             "<|start|>user",
             "<|return|>",
         ],
+        contentDropTokens: ["<|channel|>"],
     },
     qwen35: {
         // Same <tool_call>…</tool_call> framing as hermes; inner content is XML.
@@ -88,6 +92,7 @@ export function createCompletionNormalizer(config) {
     ];
     const contentFrames = dialectSpec.contentFrames ?? [];
     const dropTokens = dialectSpec.dropTokens ?? [];
+    const contentDropTokens = dialectSpec.contentDropTokens ?? [];
     const markerToEntry = new Map();
     const dropTokenSet = new Set();
     const contentMarkers = [];
@@ -112,10 +117,14 @@ export function createCompletionNormalizer(config) {
         return seq++;
     }
     function emitContent(events, text) {
-        if (!text)
+        let normalized = text;
+        for (const token of contentDropTokens) {
+            normalized = normalized.split(token).join("");
+        }
+        if (!normalized)
             return;
-        contentParts.push(text);
-        events.push({ type: "contentDelta", seq: nextSeq(), text });
+        contentParts.push(normalized);
+        events.push({ type: "contentDelta", seq: nextSeq(), text: normalized });
     }
     // Capture-gated: thinking frames may register even when `captureThinking`
     // is false (e.g. harmony's analysis channel, to strip markers from

package/dist/server/utils/worker-lock.js

@@ -1,6 +1,6 @@
 import fs from "bare-fs";
 import path from "bare-path";
-import process from "bare-process";
+import os from "bare-os";
 import { getServerLogger } from "../../logging/index.js";
 import { getQvacPath } from "./qvac-paths.js";
 const logger = getServerLogger();
@@ -10,7 +10,7 @@ function getLockFilePath() {
 }
 function isProcessAlive(pid) {
     try {
-        process.kill(pid, 0);
+        os.kill(pid, 0);
         return true;
     }
     catch {
@@ -40,13 +40,13 @@ export function acquireWorkerLock() {
         }
     }
     const content = {
-        pid: process.pid,
+        pid: os.pid(),
         startedAt: new Date().toISOString(),
     };
     try {
         fs.mkdirSync(path.dirname(lockPath), { recursive: true });
         fs.writeFileSync(lockPath, JSON.stringify(content));
-        logger.debug(`Worker lock acquired (PID ${process.pid})`);
+        logger.debug(`Worker lock acquired (PID ${os.pid()})`);
     }
     catch (error) {
         logger.error("Failed to write worker lock file:", error instanceof Error ? error.message : String(error));
@@ -56,7 +56,7 @@ export function releaseWorkerLock() {
     const lockPath = getLockFilePath();
     try {
         const existing = readLockFile(lockPath);
-        if (existing && existing.pid === process.pid) {
+        if (existing && existing.pid === os.pid()) {
             fs.unlinkSync(lockPath);
             logger.debug("Worker lock released");
         }

package/dist/server/worker-core.d.ts

@@ -8,7 +8,7 @@ export type BareDirectShutdownReason = "signal" | "rpc-close" | "uncaught-except
  * Pre-terminate cleanup, callable while the worker is still alive.
  *
  * On platforms where the worker lives in the same OS process as the JS host
- * (i.e. mobile via react-native-bare-kit Worklet), `process.exit()` would
+ * (i.e. mobile via react-native-bare-kit Worklet), `Bare.exit()` would
  * kill the entire app. This path runs the same registry/model cleanup as
  * `shutdownBareDirectWorker` but skips the lock release + exit, leaving the
  * caller (typically the SDK client about to call `worklet.terminate()`)

package/dist/server/worker-core.js

@@ -1,4 +1,5 @@
-import process from "bare-process";
+import os from "bare-os";
+import Signal from "bare-signals";
 import { createBareKitRPCServer, createIPCClient, } from "./rpc/create-server.js";
 import { destroySwarm } from "./bare/hyperswarm.js";
 import { initEnv, getValidatedEnv } from "./env.js";
@@ -6,8 +7,8 @@ import { closeAllRagInstances } from "./bare/rag-hyperdb/index.js";
 import { cleanupDownloads } from "./rpc/handlers/load-model/download-manager.js";
 import { unloadAllModels } from "./bare/registry/model-registry.js";
 import { closeRegistryClient } from "./bare/registry/registry-client.js";
-import { clearAllLoggingStreams, startLogBuffering, } from "./bare/registry/logging-stream-registry.js";
-import { clearAllAddonLoggers, getServerLogger, SDK_LOG_ID } from "../logging/index.js";
+import { clearAllLoggingStreams, startLogBuffering, stopLogBufferingWithTimeout, } from "./bare/registry/logging-stream-registry.js";
+import { clearAllAddonLoggers, getServerLogger, SDK_LOG_ID, SDK_ALL_LOG_ID, } from "../logging/index.js";
 import { clearPlugins } from "./plugins/index.js";
 import { acquireWorkerLock, releaseWorkerLock, } from "./utils/worker-lock.js";
 let coreInitialized = false;
@@ -19,11 +20,11 @@ let cleanupRan = false;
 // Set true when shutdownBareDirectWorker is in flight. Distinct from
 // cleanupRan: cleanupForTerminate must NOT set this, otherwise a later
 // SIGTERM/SIGINT/uncaught-exception would early-return at the guard
-// in shutdownBareDirectWorker and skip releaseWorkerLock + process.exit.
+// in shutdownBareDirectWorker and skip releaseWorkerLock + Bare.exit.
 let isShuttingDown = false;
 const logger = getServerLogger();
 // Defense-in-depth grace period for the SIGKILL safety net armed before
-// process.exit() in shutdownBareDirectWorker. If process.exit cannot
+// Bare.exit() in shutdownBareDirectWorker. If Bare.exit cannot
 // terminate the worker within this window — typically because some path
 // holds a non-cancellable native handle (e.g. a libuv worker thread
 // blocked on flock; see QVAC-18197) — we force-kill the OS process to
@@ -31,10 +32,10 @@ const logger = getServerLogger();
 const FORCE_EXIT_GRACE_MS = 3_000;
 function scheduleForceExit() {
     const timer = setTimeout(() => {
-        logger.error(`process.exit did not terminate the worker within ${FORCE_EXIT_GRACE_MS}ms — ` +
+        logger.error(`Bare.exit did not terminate the worker within ${FORCE_EXIT_GRACE_MS}ms — ` +
             `force-killing self (likely blocked native handle)`);
         try {
-            process.kill(process.pid, "SIGKILL");
+            os.kill(os.pid(), "SIGKILL");
         }
         catch {
             // best-effort — if SIGKILL itself fails, there's nothing more to do
@@ -52,12 +53,13 @@ export function initializeWorkerCore() {
         return { hasRPCConfig: !!validatedEnv.QVAC_IPC_SOCKET_PATH };
     }
     startLogBuffering(SDK_LOG_ID);
+    startLogBuffering(SDK_ALL_LOG_ID);
     const { hasRPCConfig } = initEnv();
     acquireWorkerLock();
     setupShutdownHandlers();
     coreInitialized = true;
     logger.debug("Worker core initialized");
-    logger.debug("Arguments to worker:", process.argv);
+    logger.debug("Arguments to worker:", Bare.argv);
     return { hasRPCConfig };
 }
 export function ensureRPCSetup() {
@@ -81,12 +83,18 @@ export function ensureRPCSetup() {
             createBareKitRPCServer();
         }
         logger.info("Bare worker started and listening for RPC requests");
-        logger.debug("Working directory:", process.cwd());
+        logger.debug("Working directory:", os.cwd());
         rpcInitialized = true;
+        // The worker is now reachable, so a `subscribeServerLogs` client can connect.
+        // Bound the global startup buffer the same way model-load buffering is bounded:
+        // if nothing subscribes within the grace window, stop buffering so every server
+        // log doesn't keep churning that buffer for the worker's whole lifetime. A
+        // subscriber that connects in time cancels the timeout and flushes the buffer.
+        stopLogBufferingWithTimeout(SDK_ALL_LOG_ID);
     }
     catch (error) {
         logger.error("Worker error:", error);
-        process.exit(1);
+        Bare.exit(1);
     }
 }
 export function isCoreInitialized() {
@@ -102,7 +110,7 @@ function clearRegistries() {
  * Clears plugin registries (which calls each addon's `releaseLogger` →
  * frees env-bound js_ref_t state), unloads all loaded models (which calls
  * each addon's `destroyInstance`), and closes infra (swarm, rag, downloads,
- * registry client). Does NOT touch the worker lock or call `process.exit`.
+ * registry client). Does NOT touch the worker lock or call `Bare.exit`.
  *
  * Idempotent: subsequent calls are no-ops via the `cleanupRan` flag. The
  * underlying clearPlugins / unloadAllModels / closers are also idempotent
@@ -126,7 +134,7 @@ async function runCleanup() {
  * Pre-terminate cleanup, callable while the worker is still alive.
  *
  * On platforms where the worker lives in the same OS process as the JS host
- * (i.e. mobile via react-native-bare-kit Worklet), `process.exit()` would
+ * (i.e. mobile via react-native-bare-kit Worklet), `Bare.exit()` would
  * kill the entire app. This path runs the same registry/model cleanup as
  * `shutdownBareDirectWorker` but skips the lock release + exit, leaving the
  * caller (typically the SDK client about to call `worklet.terminate()`)
@@ -139,7 +147,7 @@ async function runCleanup() {
 export async function cleanupForTerminate() {
     // Intentionally does NOT set isShuttingDown — that flag is reserved for
     // shutdownBareDirectWorker so a later SIGTERM/SIGINT still gets to run
-    // the lock release + process.exit path. runCleanup is idempotent on its
+    // the lock release + Bare.exit path. runCleanup is idempotent on its
     // own, so a follow-up shutdownBareDirectWorker call won't redo the body.
     if (cleanupRan)
         return;
@@ -175,16 +183,18 @@ export async function shutdownBareDirectWorker(reason) {
     releaseWorkerLock();
     scheduleForceExit();
     const isGraceful = reason === "signal" || reason === "rpc-close";
-    process.exit(isGraceful ? 0 : 1);
+    Bare.exit(isGraceful ? 0 : 1);
 }
 function setupShutdownHandlers() {
-    process.once("SIGTERM", () => void shutdownBareDirectWorker("signal"));
-    process.once("SIGINT", () => void shutdownBareDirectWorker("signal"));
-    process.on("uncaughtException", (err) => {
+    const signals = new Signal.Emitter();
+    signals.unref();
+    signals.once("SIGTERM", () => void shutdownBareDirectWorker("signal"));
+    signals.once("SIGINT", () => void shutdownBareDirectWorker("signal"));
+    Bare.on("uncaughtException", (err) => {
         logger.error("Uncaught exception in worker:", err);
         void shutdownBareDirectWorker("uncaught-exception");
     });
-    process.on("unhandledRejection", (reason) => {
+    Bare.on("unhandledRejection", (reason) => {
         logger.error("Unhandled rejection in worker:", reason);
         void shutdownBareDirectWorker("unhandled-rejection");
     });

package/dist/utils/errors-client.d.ts

@@ -109,6 +109,9 @@ export declare class ConfigFileParseFailedError extends QvacErrorBase {
 export declare class ConfigValidationFailedError extends QvacErrorBase {
     constructor(errors: string, cause?: unknown);
 }
+export declare class RequestValidationFailedError extends QvacErrorBase {
+    constructor(errors: string, cause?: unknown);
+}
 export declare class ModelUnloadFailedError extends QvacErrorBase {
     constructor(modelId?: string, cause?: unknown);
 }

package/dist/utils/errors-client.js

@@ -191,6 +191,11 @@ export class ConfigValidationFailedError extends QvacErrorBase {
         super(createErrorOptions(SDK_CLIENT_ERROR_CODES.CONFIG_VALIDATION_FAILED, [errors], cause));
     }
 }
+export class RequestValidationFailedError extends QvacErrorBase {
+    constructor(errors, cause) {
+        super(createErrorOptions(SDK_CLIENT_ERROR_CODES.REQUEST_VALIDATION_FAILED, [errors], cause));
+    }
+}
 // ============== Operation Errors (Client-side wrappers for server operations) ==============
 // These are used by client API to throw errors based on server responses
 // They reference server error codes but are thrown on client side

package/dist/utils/errors-server.d.ts

@@ -234,6 +234,8 @@ export declare class ChecksumValidationFailedError extends QvacErrorBase {
     constructor(fileName: string, cause?: unknown);
 }
 export declare class HTTPError extends QvacErrorBase {
+    /** HTTP status code; `0` denotes a connection/network failure (no response). */
+    readonly httpStatus: number;
     constructor(status: number, statusText: string, cause?: unknown);
 }
 export declare class NoResponseBodyError extends QvacErrorBase {

package/dist/utils/errors-server.js

@@ -361,8 +361,11 @@ export class ChecksumValidationFailedError extends QvacErrorBase {
     }
 }
 export class HTTPError extends QvacErrorBase {
+    /** HTTP status code; `0` denotes a connection/network failure (no response). */
+    httpStatus;
     constructor(status, statusText, cause) {
         super(createErrorOptions(SDK_SERVER_ERROR_CODES.HTTP_ERROR, [status, statusText], cause));
+        this.httpStatus = status;
     }
 }
 export class NoResponseBodyError extends QvacErrorBase {

package/dist/utils/zod-error.d.ts

@@ -0,0 +1,3 @@
+import { z } from "zod";
+export declare function formatZodError(error: z.ZodError): string;
+//# sourceMappingURL=zod-error.d.ts.map

package/dist/utils/zod-error.js

@@ -0,0 +1,4 @@
+import { z } from "zod";
+export function formatZodError(error) {
+    return z.prettifyError(error);
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@qvac/sdk",
-  "version": "0.13.5",
+  "version": "0.14.0",
   "license": "Apache-2.0",
   "repository": {
     "type": "git",
@@ -8,7 +8,7 @@
     "directory": "packages/sdk"
   },
   "bugs": "https://github.com/tetherto/qvac/issues",
-  "homepage": "https://github.com/tetherto/qvac/tree/main/packages/sdk#readme",
+  "homepage": "https://qvac.tether.io",
   "keywords": [
     "ai",
     "bare",
@@ -100,9 +100,9 @@
       "types": "./dist/server/bare/plugins/tts-ggml/plugin.d.ts",
       "import": "./dist/server/bare/plugins/tts-ggml/plugin.js"
     },
-    "./onnx-ocr/plugin": {
-      "types": "./dist/server/bare/plugins/onnx-ocr/plugin.d.ts",
-      "import": "./dist/server/bare/plugins/onnx-ocr/plugin.js"
+    "./ggml-ocr/plugin": {
+      "types": "./dist/server/bare/plugins/ggml-ocr/plugin.d.ts",
+      "import": "./dist/server/bare/plugins/ggml-ocr/plugin.js"
     },
     "./sdcpp-generation/plugin": {
       "types": "./dist/server/bare/plugins/sdcpp-generation/plugin.d.ts",
@@ -148,12 +148,6 @@
     "./pear-pre": "./dist/pear/pre.js"
   },
   "imports": {
-    "#polyfill-bare-globals": {
-      "bun": "./dist/polyfills/bare-globals.noop.js",
-      "node": "./dist/polyfills/bare-globals.noop.js",
-      "bare": "./dist/polyfills/bare-globals.js",
-      "default": "./dist/polyfills/bare-globals.noop.js"
-    },
     "#rpc": {
       "react-native": "./dist/client/rpc/expo-rpc-client.js",
       "bare": "./dist/client/rpc/bare-client.js",
@@ -206,34 +200,35 @@
     "changelog:generate": "node ../../scripts/sdk/generate-changelog-sdk-pod.cjs --package=sdk && prettier --write changelog"
   },
   "dependencies": {
-    "@qvac/bci-whispercpp": "^0.2.0",
-    "@qvac/classification-ggml": "^0.3.1",
+    "@qvac/bci-whispercpp": "^0.3.2",
+    "@qvac/classification-ggml": "^0.7.0",
     "@qvac/decoder-audio": "^0.5.0",
-    "@qvac/diffusion-cpp": "^0.11.2",
-    "@qvac/embed-llamacpp": "^0.19.1",
+    "@qvac/diffusion-cpp": "^0.12.0",
+    "@qvac/embed-llamacpp": "^0.23.0",
     "@qvac/error": "^0.1.1",
     "@qvac/langdetect-text": "^0.1.2",
-    "@qvac/llm-llamacpp": "^0.24.0",
+    "@qvac/llm-llamacpp": "^0.30.0",
     "@qvac/logging": "^0.1.0",
-    "@qvac/ocr-onnx": "^0.6.0",
+    "@qvac/ocr-ggml": "^0.7.0",
     "@qvac/rag": "^0.6.4",
-    "@qvac/registry-client": "^0.6.0",
-    "@qvac/transcription-parakeet": "^0.7.1",
-    "@qvac/transcription-whispercpp": "^0.9.0",
-    "@qvac/translation-nmtcpp": "^5.0.1",
-    "@qvac/tts-ggml": "^0.2.0",
-    "@qvac/vla-ggml": "^0.3.2",
+    "@qvac/registry-client": "^0.6.1",
+    "@qvac/transcription-parakeet": "^0.8.1",
+    "@qvac/transcription-whispercpp": "^0.10.0",
+    "@qvac/translation-nmtcpp": "^6.3.0",
+    "@qvac/tts-ggml": "^0.3.1",
+    "@qvac/vla-ggml": "^0.7.0",
     "bare-abort-controller": "^1.0.0",
     "bare-crypto": "^1.15.0",
+    "bare-env": "^3.0.0",
     "bare-fetch": "^3.0.1",
     "bare-fs": "^4.5.1",
     "bare-net": "^2.3.2",
     "bare-os": "^3.6.2",
     "bare-pack": "^2.0.1",
     "bare-path": "^3.0.1",
-    "bare-process": "^4.2.2",
     "bare-rpc": "^1.3.2",
     "bare-runtime": "^1.24.2",
+    "bare-signals": "^4.2.0",
     "bare-stream": "^2.7.0",
     "bare-zlib": "^1.3.1",
     "compact-encoding": "^3.0.0",
@@ -256,7 +251,9 @@
     "@types/node": "^24.2.1",
     "@types/semver": "^7.7.1",
     "@types/tar-stream": "^3.1.4",
+    "bare-process": "^4.4.1",
     "bare-readline": "^1.2.0",
+    "bare-stdio": "^1.0.2",
     "bare-subprocess": "^6.1.0",
     "bare-url": "^2.4.3",
     "brittle": "^4.0.0",

package/dist/polyfills/bare-globals.d.ts

@@ -1,2 +0,0 @@
-export {};
-//# sourceMappingURL=bare-globals.d.ts.map

package/dist/polyfills/bare-globals.js

@@ -1,4 +0,0 @@
-import process from "bare-process";
-if (typeof globalThis.process === "undefined") {
-    globalThis.process = process;
-}

package/dist/polyfills/bare-globals.noop.d.ts

@@ -1,2 +0,0 @@
-export {};
-//# sourceMappingURL=bare-globals.noop.d.ts.map

package/dist/polyfills/bare-globals.noop.js

@@ -1 +0,0 @@
-export {};