@qvac/sdk 0.8.2 → 0.9.0

package/dist/server/rpc/profiling/operation-wrappers.js

@@ -2,10 +2,11 @@
  * Generic wrappers for profiling handler execution.
  * Used to wrap dispatch/invoke functions with timing capture.
  */
-import { PROFILING_KEY, } from "../../../schemas/index.js";
+import { PROFILING_KEY, OPERATION_EVENT_KEY, } from "../../../schemas/index.js";
 import { nowMs, generateProfileId } from "../../../profiling/clock.js";
 import { record, shouldProfile } from "../../../profiling/controller.js";
 import { buildOperationEvent } from "./operation-metrics.js";
+import { isTerminalChunk } from "../rpc-utils";
 function getRequestProfilingMeta(request) {
     if (!request || typeof request !== "object") {
         return undefined;
@@ -33,10 +34,10 @@ function resolvePerCallProfiling(options) {
         mode: meta.mode,
     };
 }
-function recordOperationEvent(params) {
+function buildAndRecordOperationEvent(params) {
     const event = buildOperationEvent(params.options.op, params.profileId, params.startTs, params.executionMs, params.options.request, params.finalResponse, params.ttfb);
     if (!event)
-        return;
+        return undefined;
     if (params.errored) {
         event.tags = { ...event.tags, error: "true" };
     }
@@ -44,6 +45,7 @@ function recordOperationEvent(params) {
         event.count = params.count;
     }
     record(event);
+    return event;
 }
 export async function profileReplyHandler(options, handler) {
     const perCall = resolvePerCallProfiling(options);
@@ -55,18 +57,21 @@ export async function profileReplyHandler(options, handler) {
     try {
         const result = await handler();
         const executionMs = nowMs() - startTs;
-        recordOperationEvent({
+        const event = buildAndRecordOperationEvent({
             options,
             profileId,
             startTs,
             executionMs,
             finalResponse: result,
         });
+        if (event) {
+            result[OPERATION_EVENT_KEY] = event;
+        }
         return result;
     }
     catch (error) {
         const executionMs = nowMs() - startTs;
-        recordOperationEvent({
+        buildAndRecordOperationEvent({
             options,
             profileId,
             startTs,
@@ -79,37 +84,61 @@ export async function profileReplyHandler(options, handler) {
 export async function* profileStreamHandler(options, handler) {
     const perCall = resolvePerCallProfiling(options);
     if (!shouldProfile(options.op, perCall)) {
-        yield* handler();
-        return;
+        return yield* handler();
     }
     const profileId = generateProfileId();
     const startTs = nowMs();
     let ttfb;
     let lastChunk;
     let chunkCount = 0;
+    let eventAttached = false;
+    const iterator = handler();
     try {
-        for await (const chunk of handler()) {
+        while (true) {
+            const result = await iterator.next();
+            if (result.done) {
+                if (!eventAttached) {
+                    const executionMs = nowMs() - startTs;
+                    buildAndRecordOperationEvent({
+                        options,
+                        profileId,
+                        startTs,
+                        executionMs,
+                        finalResponse: lastChunk,
+                        ttfb,
+                        count: chunkCount,
+                    });
+                }
+                return result.value;
+            }
+            const chunk = result.value;
             if (ttfb === undefined) {
                 ttfb = nowMs() - startTs;
             }
             chunkCount++;
             lastChunk = chunk;
+            if (!eventAttached && isTerminalChunk(chunk)) {
+                const executionMs = nowMs() - startTs;
+                const event = buildAndRecordOperationEvent({
+                    options,
+                    profileId,
+                    startTs,
+                    executionMs,
+                    finalResponse: chunk,
+                    ttfb,
+                    count: chunkCount,
+                });
+                if (event) {
+                    chunk[OPERATION_EVENT_KEY] = event;
+                }
+                eventAttached = true;
+            }
             yield chunk;
         }
-        const executionMs = nowMs() - startTs;
-        recordOperationEvent({
-            options,
-            profileId,
-            startTs,
-            executionMs,
-            finalResponse: lastChunk,
-            ttfb,
-            count: chunkCount,
-        });
     }
     catch (error) {
         const executionMs = nowMs() - startTs;
-        recordOperationEvent({
+        buildAndRecordOperationEvent({
             options,
             profileId,
             startTs,
@@ -121,4 +150,7 @@ export async function* profileStreamHandler(options, handler) {
         });
         throw error;
     }
+    finally {
+        await iterator.return?.(undefined);
+    }
 }

package/dist/server/rpc/progress-throttle.d.ts

@@ -0,0 +1,8 @@
+export declare const PROGRESS_THROTTLE_MS = 150;
+export declare const PROGRESS_MAX_PENDING = 100;
+export type ProgressThrottle<T> = {
+    push: (update: T) => void;
+    flush: () => void;
+};
+export declare function createProgressThrottle<T>(writeBatch: (updates: T[]) => void, clock?: () => number, throttleMs?: number, maxPending?: number): ProgressThrottle<T>;
+//# sourceMappingURL=progress-throttle.d.ts.map

package/dist/server/rpc/progress-throttle.js

@@ -0,0 +1,47 @@
+export const PROGRESS_THROTTLE_MS = 150;
+export const PROGRESS_MAX_PENDING = 100;
+export function createProgressThrottle(writeBatch, clock = Date.now, throttleMs = PROGRESS_THROTTLE_MS, maxPending = PROGRESS_MAX_PENDING) {
+    let lastWrite = 0;
+    let pending = [];
+    let flushTimer = null;
+    function flushPending() {
+        if (pending.length === 0)
+            return;
+        lastWrite = clock();
+        const batch = pending;
+        pending = [];
+        writeBatch(batch);
+    }
+    function push(update) {
+        const now = clock();
+        if (now - lastWrite >= throttleMs) {
+            lastWrite = now;
+            writeBatch([update]);
+        }
+        else {
+            pending.push(update);
+            if (pending.length >= maxPending) {
+                if (flushTimer) {
+                    clearTimeout(flushTimer);
+                    flushTimer = null;
+                }
+                flushPending();
+                return;
+            }
+            if (!flushTimer) {
+                flushTimer = setTimeout(() => {
+                    flushTimer = null;
+                    flushPending();
+                }, throttleMs - (now - lastWrite));
+            }
+        }
+    }
+    function flush() {
+        if (flushTimer) {
+            clearTimeout(flushTimer);
+            flushTimer = null;
+        }
+        flushPending();
+    }
+    return { push, flush };
+}

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

@@ -0,0 +1,5 @@
+export declare function getNextCommandId(): number;
+export declare function isTerminalChunk(value: unknown): value is {
+    done: true;
+};
+//# sourceMappingURL=rpc-utils.d.ts.map

package/dist/server/rpc/rpc-utils.js

@@ -0,0 +1,11 @@
+let commandCounter = 0;
+export function getNextCommandId() {
+    commandCounter = (commandCounter + 1) % Number.MAX_SAFE_INTEGER;
+    return commandCounter;
+}
+export function isTerminalChunk(value) {
+    return (typeof value === "object" &&
+        value !== null &&
+        "done" in value &&
+        value.done === true);
+}

package/dist/server/utils/path-security.js

@@ -8,7 +8,7 @@ export { sanitizePathComponent } from "../../utils/path-sanitize.js";
  * Both paths are resolved to absolute before comparison.
  */
 export function isPathWithinBase(basePath, targetPath) {
-    return checkPathWithinBase(basePath, targetPath, path.resolve, path.sep || "/");
+    return checkPathWithinBase(basePath, targetPath, (...args) => path.resolve(...args), path.sep || "/");
 }
 /**
  * Sanitize components, join them to a base path, and verify the result

package/dist/server/utils/tool-integration.js

@@ -14,8 +14,18 @@ export function setupToolGrammar(modelConfig, tools) {
     const grammar = convertToolsToGrammar(tools);
     modelConfig["grammar"] = grammar;
 }
+function isInsideThinkBlock(text) {
+    const lastOpen = text.lastIndexOf("<think>");
+    if (lastOpen === -1)
+        return false;
+    const lastClose = text.lastIndexOf("</think>");
+    return lastClose < lastOpen;
+}
 export function checkForToolEvents(accumulatedText, currentToken, tools, emittedToolCallPositions) {
     const events = [];
+    if (isInsideThinkBlock(accumulatedText)) {
+        return events;
+    }
     if (currentToken.includes("</tool_call>") || currentToken.includes("}")) {
         const { toolCalls, errors } = parseToolCalls(accumulatedText, tools);
         for (const call of toolCalls) {

package/dist/server/utils/tool-parser.js

@@ -1,3 +1,6 @@
+function stripThinkingBlocks(text) {
+    return text.replace(/<think>[\s\S]*?<\/think>/gi, "");
+}
 let toolCallSequence = 0;
 function generateStableToolCallId(name, args) {
     const content = `${name}:${JSON.stringify(args)}`;
@@ -196,22 +199,23 @@ export function parseToolCalls(text, tools) {
     if (!tools || tools.length === 0) {
         return { toolCalls: [], errors: [] };
     }
-    let result = parseGemmaFormat(text, tools);
+    const cleaned = stripThinkingBlocks(text);
+    let result = parseGemmaFormat(cleaned, tools);
     if (result.toolCalls.length > 0) {
         return result;
     }
-    result = parseQwenFormat(text, tools);
+    result = parseQwenFormat(cleaned, tools);
     if (result.toolCalls.length > 0) {
         return result;
     }
-    result = parseLlamacppFormat(text, tools);
+    result = parseLlamacppFormat(cleaned, tools);
     if (result.toolCalls.length > 0) {
         return result;
     }
-    if (text.includes("<tool_call>")) {
+    if (cleaned.includes("<tool_call>")) {
         return { toolCalls: [], errors: [] };
     }
-    result = parseGenericFormat(text, tools);
+    result = parseGenericFormat(cleaned, tools);
     return result;
 }
 export function convertToolsToGrammar(tools) {

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

@@ -3,4 +3,6 @@ export declare function initializeWorkerCore(): {
 };
 export declare function ensureRPCSetup(): void;
 export declare function isCoreInitialized(): boolean;
+export type BareDirectShutdownReason = "signal" | "rpc-close";
+export declare function shutdownBareDirectWorker(reason: BareDirectShutdownReason): Promise<void>;
 //# sourceMappingURL=worker-core.d.ts.map

package/dist/server/worker-core.js

@@ -61,28 +61,31 @@ function clearRegistries() {
     clearAllAddonLoggers();
     clearPlugins();
 }
+export async function shutdownBareDirectWorker(reason) {
+    if (isShuttingDown)
+        return;
+    isShuttingDown = true;
+    const message = reason === "signal"
+        ? "🐻 Bare worker shutdown signal received, cleaning up..."
+        : "🧹 Bare direct mode RPC closed, cleaning up...";
+    logger.info(message);
+    try {
+        clearRegistries();
+        await Promise.allSettled([
+            destroySwarm(),
+            closeAllRagInstances(),
+            cleanupDownloads(),
+            unloadAllModels(),
+            closeRegistryClient(),
+        ]);
+        logger.info("✅ Cleanup completed successfully");
+    }
+    catch (error) {
+        logger.error("❌ Error during shutdown cleanup:", error);
+    }
+    process.exit(0);
+}
 function setupShutdownHandlers() {
-    const shutdownHandler = async () => {
-        if (isShuttingDown)
-            return;
-        isShuttingDown = true;
-        logger.info("🐻 Bare worker shutdown signal received, cleaning up...");
-        try {
-            clearRegistries();
-            await Promise.allSettled([
-                destroySwarm(),
-                closeAllRagInstances(),
-                cleanupDownloads(),
-                unloadAllModels(),
-                closeRegistryClient(),
-            ]);
-            logger.info("✅ Cleanup completed successfully");
-        }
-        catch (error) {
-            logger.error("❌ Error during shutdown cleanup:", error);
-        }
-        process.exit(0);
-    };
-    process.once("SIGTERM", () => void shutdownHandler());
-    process.once("SIGINT", () => void shutdownHandler());
+    process.once("SIGTERM", () => void shutdownBareDirectWorker("signal"));
+    process.once("SIGINT", () => void shutdownBareDirectWorker("signal"));
 }

package/dist/server/worker.js

@@ -4,7 +4,7 @@
 import { initializeWorkerCore, ensureRPCSetup } from "./worker-core.js";
 import { registerPlugin } from "./plugins/index.js";
 import { getServerLogger } from "../logging/index.js";
-import { llmPlugin, embeddingsPlugin, whisperPlugin, parakeetPlugin, nmtPlugin, ttsPlugin, ocrPlugin, } from "./bare/plugins/index.js";
+import { llmPlugin, embeddingsPlugin, whisperPlugin, parakeetPlugin, nmtPlugin, ttsPlugin, ocrPlugin, diffusionPlugin, } from "./bare/plugins/index.js";
 const { hasRPCConfig } = initializeWorkerCore();
 const logger = getServerLogger();
 logger.info("🐻 Hello from Bare");
@@ -16,6 +16,7 @@ registerPlugin(parakeetPlugin);
 registerPlugin(nmtPlugin);
 registerPlugin(ttsPlugin);
 registerPlugin(ocrPlugin);
+registerPlugin(diffusionPlugin);
 logger.info(hasRPCConfig
     ? "Parsed RPC configuration from arguments"
     : "Using default configuration (direct mode)");

package/dist/utils/encoding.d.ts

@@ -0,0 +1,9 @@
+/**
+ * Decodes a base64 string to a Uint8Array.
+ *
+ * Uses atob which is available across all runtimes (Node, Bare, React Native).
+ * Buffer is not globally available in React Native (Hermes engine), so this
+ * provides a cross-platform alternative to Buffer.from(str, "base64").
+ */
+export declare function decodeBase64(base64: string): Uint8Array;
+//# sourceMappingURL=encoding.d.ts.map

package/dist/utils/encoding.js

@@ -0,0 +1,14 @@
+/**
+ * Decodes a base64 string to a Uint8Array.
+ *
+ * Uses atob which is available across all runtimes (Node, Bare, React Native).
+ * Buffer is not globally available in React Native (Hermes engine), so this
+ * provides a cross-platform alternative to Buffer.from(str, "base64").
+ */
+export function decodeBase64(base64) {
+    const binary = atob(base64);
+    const bytes = new Uint8Array(binary.length);
+    for (let i = 0; i < binary.length; i++)
+        bytes[i] = binary.charCodeAt(i);
+    return bytes;
+}

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

@@ -227,6 +227,12 @@ export declare class PluginLoggingInvalidError extends QvacErrorBase {
 export declare class PluginDefinitionInvalidError extends QvacErrorBase {
     constructor(modelType: string, details: string, cause?: unknown);
 }
+export declare class LifecycleSuspendFailedError extends QvacErrorBase {
+    constructor(details?: string, cause?: unknown);
+}
+export declare class LifecycleResumeFailedError extends QvacErrorBase {
+    constructor(details?: string, cause?: unknown);
+}
 export declare class PathTraversalError extends QvacErrorBase {
     constructor(component: string, basePath: string, cause?: unknown);
 }

package/dist/utils/errors-server.js

@@ -396,6 +396,17 @@ export class PluginDefinitionInvalidError extends QvacErrorBase {
         super(createErrorOptions(SDK_SERVER_ERROR_CODES.PLUGIN_DEFINITION_INVALID, [modelType, details], cause));
     }
 }
+// ============== Lifecycle Errors ==============
+export class LifecycleSuspendFailedError extends QvacErrorBase {
+    constructor(details, cause) {
+        super(createErrorOptions(SDK_SERVER_ERROR_CODES.LIFECYCLE_SUSPEND_FAILED, details ? [details] : undefined, cause));
+    }
+}
+export class LifecycleResumeFailedError extends QvacErrorBase {
+    constructor(details, cause) {
+        super(createErrorOptions(SDK_SERVER_ERROR_CODES.LIFECYCLE_RESUME_FAILED, details ? [details] : undefined, cause));
+    }
+}
 // ============== Security Errors ==============
 export class PathTraversalError extends QvacErrorBase {
     constructor(component, basePath, cause) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@qvac/sdk",
-  "version": "0.8.2",
+  "version": "0.9.0",
   "license": "Apache-2.0",
   "repository": {
     "type": "git",
@@ -9,6 +9,51 @@
   },
   "bugs": "https://github.com/tetherto/qvac/issues",
   "homepage": "https://github.com/tetherto/qvac/tree/main/packages/sdk#readme",
+  "keywords": [
+    "ai",
+    "bare",
+    "cross-platform",
+    "deep-learning",
+    "diffusion",
+    "edge-ai",
+    "embeddings",
+    "expo",
+    "gguf",
+    "gpu",
+    "holepunch",
+    "hyperswarm",
+    "image-generation",
+    "inference",
+    "large-language-model",
+    "llama-cpp",
+    "llm",
+    "local",
+    "local-ai",
+    "machine-learning",
+    "ml",
+    "model-registry",
+    "node",
+    "ocr",
+    "offline",
+    "on-device",
+    "p2p",
+    "peer-to-peer",
+    "private-ai",
+    "qvac",
+    "rag",
+    "react-native",
+    "sdk",
+    "self-hosted",
+    "speech-to-text",
+    "stt",
+    "tether",
+    "text-to-speech",
+    "transcription",
+    "translation",
+    "tts",
+    "typescript",
+    "whisper"
+  ],
   "type": "module",
   "main": "./dist/index.js",
   "types": "./dist/index.d.ts",
@@ -51,6 +96,10 @@
       "types": "./dist/server/bare/plugins/onnx-ocr/plugin.d.ts",
       "import": "./dist/server/bare/plugins/onnx-ocr/plugin.js"
     },
+    "./sdcpp-generation/plugin": {
+      "types": "./dist/server/bare/plugins/sdcpp-generation/plugin.d.ts",
+      "import": "./dist/server/bare/plugins/sdcpp-generation/plugin.js"
+    },
     "./plugin-utils": {
       "types": "./dist/schemas/plugin.d.ts",
       "import": "./dist/schemas/plugin.js"
@@ -69,6 +118,12 @@
     "./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",
@@ -117,34 +172,35 @@
     "changelog:generate": "node ../../scripts/sdk/generate-changelog-sdk-pod.cjs --package=sdk && prettier --write changelog"
   },
   "dependencies": {
-    "@qvac/decoder-audio": "^0.3.3",
-    "@qvac/dl-filesystem": "^0.2.0",
-    "@qvac/embed-llamacpp": "^0.12.0",
+    "@qvac/decoder-audio": "^0.3.7",
+    "@qvac/diffusion-cpp": "^0.1.3",
+    "@qvac/dl-filesystem": "^0.2.1",
+    "@qvac/embed-llamacpp": "^0.13.4",
     "@qvac/error": "^0.1.1",
-    "@qvac/langdetect-text": "^0.1.0",
-    "@qvac/llm-llamacpp": "^0.12.1",
+    "@qvac/langdetect-text": "^0.1.2",
+    "@qvac/llm-llamacpp": "^0.14.4",
     "@qvac/logging": "^0.1.0",
-    "@qvac/ocr-onnx": "^0.2.0",
-    "@qvac/rag": "^0.4.2",
-    "@qvac/registry-client": "^0.2.1",
-    "@qvac/transcription-parakeet": "^0.2.0",
-    "@qvac/transcription-whispercpp": "^0.5.0",
-    "@qvac/translation-nmtcpp": "^0.6.1",
-    "@qvac/tts-onnx": "^0.6.1",
+    "@qvac/ocr-onnx": "^0.4.2",
+    "@qvac/rag": "^0.4.4",
+    "@qvac/registry-client": "^0.4.0",
+    "@qvac/transcription-parakeet": "^0.3.1",
+    "@qvac/transcription-whispercpp": "^0.6.1",
+    "@qvac/translation-nmtcpp": "^0.6.10",
+    "@qvac/tts-onnx": "^0.8.2",
     "fast-safe-stringify": "2.1.1",
     "which-runtime": "^1.3.2",
     "zod": "^4.0.17"
   },
   "devDependencies": {
     "@eslint/js": "^9.33.0",
-    "@expo/config-plugins": "^54.0.1",
+    "@expo/config-plugins": "^54.0.4",
     "@lancedb/lancedb": "^0.21.3",
     "@modelcontextprotocol/sdk": "^1.12.1",
-    "@qvac/cli": "^0.2.2",
+    "@qvac/cli": "^0.2.3",
     "@sqliteai/sqlite-wasm": "3.50.4-sync.0.8.30-vector.0.9.23",
     "@types/node": "^24.2.1",
     "@types/tar-stream": "^3.1.4",
-    "bare-pack": "^2.0.0",
+    "bare-pack": "^2.0.1",
     "bare-readline": "^1.2.0",
     "brittle": "^3.19.1",
     "chromadb": "^3.0.14",
@@ -152,9 +208,9 @@
     "eslint-import-resolver-typescript": "^4.4.4",
     "eslint-plugin-import": "^2.30.0",
     "eslint-plugin-react": "^7.37.5",
-    "expo": "^54.0.0",
-    "expo-device": "^8.0.0",
-    "expo-file-system": "^19.0.15",
+    "expo": "^54.0.33",
+    "expo-device": "^8.0.10",
+    "expo-file-system": "^19.0.21",
     "globals": "^16.3.0",
     "husky": "^9.1.7",
     "hyperdb": "^4.16.1",
@@ -166,20 +222,20 @@
     "remark-frontmatter": "^5.0.0",
     "remark-mdx": "^3.1.1",
     "tsx": "^4.21.0",
-    "typescript": "^5.9.2",
+    "typescript": "^5.9.3",
     "typescript-eslint": "^8.39.0"
   },
   "peerDependencies": {
     "@modelcontextprotocol/sdk": "^1.0.0",
-    "@qvac/cli": "^0.2.2",
+    "@qvac/cli": "^0.2.3",
     "bare-abort-controller": "^1.0.0",
     "bare-buffer": "^3.4.2",
-    "bare-crypto": "^1.11.2",
+    "bare-crypto": "^1.13.4",
     "bare-dns": "^2.1.4",
     "bare-fetch": "^2.5.1",
     "bare-fs": "^4.5.2",
     "bare-http1": "^4.1.6",
-    "bare-https": "^2.1.1",
+    "bare-https": "^2.1.3",
     "bare-link": "^2.1.10",
     "bare-net": "^2.2.0",
     "bare-os": "^3.6.2",
@@ -188,7 +244,7 @@
     "bare-rpc": "^1.0.0",
     "bare-runtime": "^1.24.2",
     "bare-stream": "^2.7.0",
-    "bare-subprocess": "^5.2.1",
+    "bare-subprocess": "^5.2.3",
     "bare-zlib": "^1.3.1",
     "compact-encoding": "^2.19.0",
     "corestore": "^7.4.5",
@@ -202,7 +258,7 @@
     "llm-splitter": "~0.2.0",
     "pear-pipe": "^1.0.6",
     "react-native-bare-kit": "^0.11.5",
-    "tar-stream": "^3.1.7"
+    "tar-stream": "^3.1.8"
   },
   "peerDependenciesMeta": {
     "@qvac/cli": {

package/dist/server/rpc/delegate-utils.d.ts

@@ -1,2 +0,0 @@
-export declare function getNextCommandId(): number;
-//# sourceMappingURL=delegate-utils.d.ts.map

package/dist/server/rpc/delegate-utils.js

@@ -1,5 +0,0 @@
-let commandCounter = 0;
-export function getNextCommandId() {
-    commandCounter = (commandCounter + 1) % Number.MAX_SAFE_INTEGER;
-    return commandCounter;
-}