@qvac/sdk 0.20.1 → 0.21.1

package/dist/src/schemas/sdk-errors-server.js

@@ -21,6 +21,7 @@ export const SDK_SERVER_ERROR_CODES = {
     LEGACY_PARAKEET_MODEL_DEPRECATED: 52210,
     LEGACY_TTS_MODEL_DEPRECATED: 52211,
     // Model Operations (52,400-52,799)
+    RPC_SERVER_OPERATION_FAILED: 52423,
     MODEL_UNLOAD_FAILED: 52400,
     EMBED_FAILED: 52401,
     EMBED_NO_EMBEDDINGS: 52402,
@@ -178,6 +179,10 @@ const serverErrorDefinitions = {
         message: (legacyFields) => `Legacy ONNX TTS modelConfig fields are no longer supported (${legacyFields ?? 'unknown fields'}). As of @qvac/tts-ggml the addon uses GGUF bundles: supply the primary GGUF via modelSrc, set language in modelConfig, and for Chatterbox add s3genModelSrc (e.g. loadModel({ modelSrc: TTS_T3_TURBO_EN_CHATTERBOX_Q8_0, modelType: "tts", modelConfig: { ttsEngine: "chatterbox", language: "en", s3genModelSrc: TTS_S3GEN_EN_CHATTERBOX } })). Supertonic multilingual mode is selected by the GGUF (e.g. TTS_MULTILINGUAL_SUPERTONIC2_Q8_0) plus language — not ttsSupertonicMultilingual.`
     },
     // Model Operations (52,400-52,799)
+    [SDK_SERVER_ERROR_CODES.RPC_SERVER_OPERATION_FAILED]: {
+        name: 'RPC_SERVER_OPERATION_FAILED',
+        message: (operation, details) => `${operation}: ${details}`
+    },
     [SDK_SERVER_ERROR_CODES.MODEL_UNLOAD_FAILED]: {
         name: 'MODEL_UNLOAD_FAILED',
         message: (modelId) => `Failed to unload model${modelId ? ` "${modelId}"` : ''}`

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

@@ -100,6 +100,21 @@ export declare class BareImportsMapNotFoundError extends QvacErrorBase {
 export declare class BareRuntimeBinaryNotFoundError extends QvacErrorBase {
     constructor(platform: string, arch: string, cause?: unknown);
 }
+/**
+ * Addon platform packages are missing and cannot be installed automatically:
+ * no supported package manager was detected, two installed versions of one
+ * addon need the same package, or an addon names an invalid package.
+ * `dependencies` maps each missing package to the exact version to pin.
+ */
+export declare class HostPrebuildsInstallRefusedError extends QvacErrorBase {
+    readonly dependencies: Record<string, string>;
+    constructor(reason: string, dependencies: Record<string, string>, cause?: unknown);
+}
+/** The package manager ran but the missing addon platform packages are still not installed. */
+export declare class HostPrebuildsInstallFailedError extends QvacErrorBase {
+    readonly dependencies: Record<string, string>;
+    constructor(details: string, dependencies: Record<string, string>, cause?: unknown);
+}
 export declare class ConfigFileNotFoundError extends QvacErrorBase {
     constructor(searchPaths: string, cause?: unknown);
 }

package/dist/src/utils/errors-client.js

@@ -166,6 +166,32 @@ export class BareRuntimeBinaryNotFoundError extends QvacErrorBase {
         super(createErrorOptions(SDK_CLIENT_ERROR_CODES.BARE_RUNTIME_BINARY_NOT_FOUND, [platform, arch], cause));
     }
 }
+/**
+ * Addon platform packages are missing and cannot be installed automatically:
+ * no supported package manager was detected, two installed versions of one
+ * addon need the same package, or an addon names an invalid package.
+ * `dependencies` maps each missing package to the exact version to pin.
+ */
+export class HostPrebuildsInstallRefusedError extends QvacErrorBase {
+    dependencies;
+    constructor(reason, dependencies, cause) {
+        super(createErrorOptions(SDK_CLIENT_ERROR_CODES.HOST_PREBUILDS_INSTALL_REFUSED, [reason, formatDependencyPins(dependencies)], cause));
+        this.dependencies = dependencies;
+    }
+}
+/** The package manager ran but the missing addon platform packages are still not installed. */
+export class HostPrebuildsInstallFailedError extends QvacErrorBase {
+    dependencies;
+    constructor(details, dependencies, cause) {
+        super(createErrorOptions(SDK_CLIENT_ERROR_CODES.HOST_PREBUILDS_INSTALL_FAILED, [details, formatDependencyPins(dependencies)], cause));
+        this.dependencies = dependencies;
+    }
+}
+function formatDependencyPins(dependencies) {
+    return Object.entries(dependencies)
+        .map(([name, version]) => `    "${name}": "${version}"`)
+        .join(',\n');
+}
 export class ConfigFileNotFoundError extends QvacErrorBase {
     constructor(searchPaths, cause) {
         super(createErrorOptions(SDK_CLIENT_ERROR_CODES.CONFIG_FILE_NOT_FOUND, [searchPaths], cause));

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

@@ -389,4 +389,5 @@ export declare class PathTraversalError extends QvacErrorBase {
 export declare class ModelRegistryQueryFailedError extends QvacErrorBase {
     constructor(details?: string, cause?: unknown);
 }
+export { RpcServerOperationError } from '@qvac/inference/surface';
 //# sourceMappingURL=errors-server.d.ts.map

package/dist/src/utils/errors-server.js

@@ -640,3 +640,4 @@ export class ModelRegistryQueryFailedError extends QvacErrorBase {
         super(createErrorOptions(SDK_SERVER_ERROR_CODES.QVAC_MODEL_REGISTRY_QUERY_FAILED, details ? [details] : undefined, cause));
     }
 }
+export { RpcServerOperationError } from '@qvac/inference/surface';

package/dist/src/utils/mcp-adapter.js

@@ -34,7 +34,8 @@ function createMcpToolHandler(client, toolName) {
 export async function getMcpToolsWithHandlers(clients) {
     const allTools = [];
     const handlers = new Map();
-    for (const { client, includeResources } of clients) {
+    for (const { client, includeResources, deferLoading, group } of clients) {
+        const start = allTools.length;
         const { tools: mcpTools } = await client.listTools();
         for (const mcpTool of mcpTools) {
             allTools.push(convertMcpToolToTool(mcpTool));
@@ -87,6 +88,11 @@ export async function getMcpToolsWithHandlers(clients) {
                 });
             }
         }
+        if (deferLoading === true) {
+            for (let i = start; i < allTools.length; i++) {
+                allTools[i] = { ...allTools[i], deferLoading: true, ...(group !== undefined && { group }) };
+            }
+        }
     }
     return { tools: allTools, handlers };
 }

package/dist/src/utils/tool-helpers.d.ts

@@ -7,6 +7,10 @@ export type ToolInput<T extends ZodObjectType = ZodObjectType> = {
     description: string;
     parameters: T;
     handler?: ToolHandler;
+    /** Keep this tool's parameter schema out of the initial prompt; see `Tool.deferLoading`. */
+    deferLoading?: boolean;
+    /** Heading this tool is listed under in the deferred catalog. */
+    group?: string;
 };
 export declare function convertToolInput(input: ToolInput): Tool;
 export declare function convertTools(inputs: ToolInput[]): Tool[];

package/dist/src/utils/tool-helpers.js

@@ -54,6 +54,8 @@ export function convertToolInput(input) {
         type: 'function',
         name: input.name,
         description: input.description,
+        ...(input.deferLoading !== undefined && { deferLoading: input.deferLoading }),
+        ...(input.group !== undefined && { group: input.group }),
         parameters: {
             type: 'object',
             properties,

package/dist/src/worker/ggml-rpc-server-provider.d.ts

@@ -0,0 +1,2 @@
+export { default, ggmlRpcServerProvider } from '@qvac/inference/ggml-rpc-server/provider';
+//# sourceMappingURL=ggml-rpc-server-provider.d.ts.map

package/dist/src/worker/ggml-rpc-server-provider.js

@@ -0,0 +1 @@
+export { default, ggmlRpcServerProvider } from '@qvac/inference/ggml-rpc-server/provider';

package/dist/src/worker/rpc-server-provider.d.ts

@@ -0,0 +1,3 @@
+export { registerRpcServerProvider } from '@qvac/inference/rpc-server-provider';
+export type { RpcServerProvider, RpcServerHandle } from '@qvac/inference/surface';
+//# sourceMappingURL=rpc-server-provider.d.ts.map

package/dist/src/worker/rpc-server-provider.js

@@ -0,0 +1 @@
+export { registerRpcServerProvider } from '@qvac/inference/rpc-server-provider';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@qvac/sdk",
-  "version": "0.20.1",
+  "version": "0.21.1",
   "license": "Apache-2.0",
   "repository": {
     "type": "git",
@@ -68,9 +68,6 @@
       "import": "./dist/src/worker/index.js",
       "require": "./dist/src/worker/index.js"
     },
-    "./dist/server/worker.js": {
-      "import": "./dist/src/worker/legacy-worker-entry.js"
-    },
     "./llamacpp-completion/plugin": {
       "types": "./dist/src/worker/plugins/llamacpp-completion.d.ts",
       "import": "./dist/src/worker/plugins/llamacpp-completion.js"
@@ -142,9 +139,6 @@
     "./worker-lifecycle": {
       "import": "./dist/src/worker/lifecycle.js"
     },
-    "./worker-core": {
-      "import": "./dist/src/worker/worker-core.js"
-    },
     "./plugins": {
       "import": "./dist/src/server/plugins/index.js"
     },
@@ -160,7 +154,15 @@
       "import": "./dist/src/models/registry/index.js",
       "require": "./dist/src/models/registry/index.js"
     },
-    "./pear-pre": "./dist/src/pear/pre.js"
+    "./pear-pre": "./dist/src/pear/pre.js",
+    "./rpc-server-provider": {
+      "types": "./dist/src/worker/rpc-server-provider.d.ts",
+      "import": "./dist/src/worker/rpc-server-provider.js"
+    },
+    "./ggml-rpc-server/provider": {
+      "types": "./dist/src/worker/ggml-rpc-server-provider.d.ts",
+      "import": "./dist/src/worker/ggml-rpc-server-provider.js"
+    }
   },
   "imports": {
     "#rpc": {
@@ -192,9 +194,9 @@
     "test:unit": "bun run scripts/run-unit-tests.ts",
     "format": "prettier --check .",
     "format:fix": "prettier --write .",
-    "lint": "lunte && npm run enforce-inference-peer-dependencies",
+    "lint": "lunte && npm run enforce-inference-versions",
     "lint:fix": "lunte --fix",
-    "enforce-inference-peer-dependencies": "bun scripts/enforce-inference-peer-dependencies.ts",
+    "enforce-inference-versions": "bun scripts/enforce-inference-versions.ts",
     "typecheck": "tsc --noEmit -p tsconfig.json && tsc --noEmit -p test/tsconfig.types.json",
     "postcompile:aliases": "tsc-alias -p tsconfig.alias.json",
     "build": "rm -rf dist && npm run lint && tsc --project tsconfig.json && npm run postcompile:aliases",
@@ -209,29 +211,32 @@
     "contract:export": "tsx ./scripts/export-contract.ts",
     "contract:check": "tsx ./scripts/export-contract.ts --check",
     "check:resource-collectors": "tsx ./scripts/check-resource-collectors-packaging.ts",
+    "bare-kit-runtimes:generate": "tsx ./scripts/generate-bare-kit-runtimes.ts",
     "changelog:generate": "node ../../scripts/sdk/generate-changelog-sdk-pod.cjs --package=sdk && prettier --write changelog"
   },
   "dependencies": {
-    "@qvac/asr-ggml": "^0.5.3",
-    "@qvac/audiogen-ggml": "^0.4.0",
-    "@qvac/bci-whispercpp": "^0.9.1",
-    "@qvac/classification-ggml": "^0.26.1",
-    "@qvac/decoder-audio": "^0.6.0",
-    "@qvac/diffusion-cpp": "^0.24.0",
-    "@qvac/embed-llamacpp": "^0.41.1",
+    "@qvac/asr-ggml": "^0.7.1",
+    "@qvac/audiogen-ggml": "^0.5.2",
+    "@qvac/bci-whispercpp": "^0.10.2",
+    "@qvac/classification-ggml": "^0.28.2",
+    "@qvac/decoder-audio": "^0.7.0",
+    "@qvac/diffusion-cpp": "^0.26.1",
+    "@qvac/embed-llamacpp": "^0.43.2",
     "@qvac/error": "^0.1.1",
-    "@qvac/inference": "^0.20.0",
+    "@qvac/ggml-rpc-server": "^0.1.2",
+    "@qvac/inference": "^0.21.1",
     "@qvac/langdetect-text": "^0.1.2",
-    "@qvac/llm-llamacpp": "^0.53.0",
+    "@qvac/llm-llamacpp": "^0.55.2",
     "@qvac/logging": "^0.1.0",
-    "@qvac/ocr-ggml": "^0.24.1",
-    "@qvac/translation-nmtcpp": "^0.16.1",
-    "@qvac/tts-ggml": "^0.9.1",
-    "@qvac/vla-ggml": "^0.27.1",
+    "@qvac/ocr-ggml": "^0.26.2",
+    "@qvac/translation-nmtcpp": "^0.18.2",
+    "@qvac/tts-ggml": "^0.10.2",
+    "@qvac/vla-ggml": "^0.29.2",
     "bare-abort-controller": "^1.0.0",
     "bare-cpu-info": "0.1.1",
     "bare-crypto": "^1.15.0",
     "bare-env": "^3.0.0",
+    "bare-events": "2.9.1",
     "bare-fs": "^4.5.1",
     "bare-gpu-info": "0.1.1",
     "bare-net": "^2.3.2",
@@ -239,9 +244,9 @@
     "bare-pack": "^2.0.1",
     "bare-path": "^3.0.1",
     "bare-rpc": "^1.3.2",
-    "bare-runtime": "^1.24.2",
+    "bare-runtime": "^1.30.3",
     "bare-signals": "^4.2.0",
-    "bare-stream": "^2.7.0",
+    "bare-stream": "2.13.3",
     "bare-zlib": "^1.3.1",
     "compact-encoding": "^3.0.0",
     "corestore": "^7.4.5",

package/dist/scripts/enforce-inference-peer-dependencies.d.ts

@@ -1,2 +0,0 @@
-export {};
-//# sourceMappingURL=enforce-inference-peer-dependencies.d.ts.map

package/dist/scripts/enforce-inference-peer-dependencies.js

@@ -1,39 +0,0 @@
-// @qvac/inference peerDependencies is the source of truth for addon version
-// ranges. The SDK should use the same version ranges as inference for addons.
-// For every addon p in inference's peerDependencies, this script validates if:
-//   - inference's devDependencies.p == inference's peerDependencies.p,
-//   - SDK's dependencies.p == inference's peerDependencies.p.
-// That is, inference's devDependencies and SDK's dependencies match inference's peerDependencies.
-import { readFileSync } from 'fs';
-import { join, resolve, dirname } from 'path';
-import { fileURLToPath } from 'url';
-const sdkDir = resolve(dirname(fileURLToPath(import.meta.url)), '..');
-const inferenceDir = resolve(sdkDir, '..', 'inference');
-function readManifest(dir) {
-    return JSON.parse(readFileSync(join(dir, 'package.json'), 'utf8'));
-}
-const inferencePkg = readManifest(inferenceDir);
-const sdkPkg = readManifest(sdkDir);
-const peers = inferencePkg.peerDependencies ?? {};
-const inferenceDev = inferencePkg.devDependencies ?? {};
-const sdkDeps = sdkPkg.dependencies ?? {};
-const drifts = [];
-for (const [name, peer] of Object.entries(peers)) {
-    const dev = inferenceDev[name];
-    const dep = sdkDeps[name];
-    const mismatches = [];
-    if (dep !== peer)
-        mismatches.push(dep === undefined ? 'SDK is missing it' : `SDK has ${dep}`);
-    if (dev !== peer) {
-        mismatches.push(dev === undefined
-            ? 'inference devDependencies is missing it'
-            : `inference devDependencies has ${dev}`);
-    }
-    if (mismatches.length > 0) {
-        drifts.push(`${name}: inference has it as ${peer} in peerDependencies, but ${mismatches.join(' and ')}.`);
-    }
-}
-if (drifts.length > 0) {
-    console.error(`${drifts.join('\n')}\n\nPlease ensure the SDK dependencies and inference devDependencies use the same addon versions as inference's peerDependencies.`);
-    process.exit(1);
-}

package/dist/src/deprecate.d.ts

@@ -1,2 +0,0 @@
-export declare function deprecate(key: string, message: string): void;
-//# sourceMappingURL=deprecate.d.ts.map

package/dist/src/deprecate.js

@@ -1,9 +0,0 @@
-import { getServerLogger } from './logging/index.js';
-const warned = new Set();
-export function deprecate(key, message) {
-    if (warned.has(key))
-        return;
-    warned.add(key);
-    // enableConsole so the notice shows even in a worker with no log-stream subscriber.
-    getServerLogger({ enableConsole: true }).warn(`DEPRECATED: ${message}`);
-}

package/dist/src/worker/legacy-worker-entry.d.ts

@@ -1,2 +0,0 @@
-import './index.js';
-//# sourceMappingURL=legacy-worker-entry.d.ts.map

package/dist/src/worker/legacy-worker-entry.js

@@ -1,3 +0,0 @@
-import './index.js';
-import { deprecate } from '../deprecate.js';
-deprecate('worker-entry', "'@qvac/sdk/dist/server/worker.js' is deprecated; use '@qvac/sdk/dist/src/worker/index.js'.");

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

@@ -1,7 +0,0 @@
-import { ensureRPCSetup, cleanupForTerminate, initializeWorker, isWorkerInitialized, shutdownWorker, type WorkerShutdownReason } from './lifecycle.js';
-export { ensureRPCSetup, cleanupForTerminate };
-export declare const initializeWorkerCore: typeof initializeWorker;
-export declare const isCoreInitialized: typeof isWorkerInitialized;
-export declare const shutdownBareDirectWorker: typeof shutdownWorker;
-export type BareDirectShutdownReason = WorkerShutdownReason;
-//# sourceMappingURL=worker-core.d.ts.map

package/dist/src/worker/worker-core.js

@@ -1,8 +0,0 @@
-import { deprecate } from '../deprecate.js';
-import { ensureRPCSetup, cleanupForTerminate, initializeWorker, isWorkerInitialized, shutdownWorker } from './lifecycle.js';
-deprecate('worker-core', "'@qvac/sdk/worker-core' is deprecated; import from '@qvac/sdk/worker-lifecycle' " +
-    '(initializeWorker, isWorkerInitialized, shutdownWorker, WorkerShutdownReason).');
-export { ensureRPCSetup, cleanupForTerminate };
-export const initializeWorkerCore = initializeWorker;
-export const isCoreInitialized = isWorkerInitialized;
-export const shutdownBareDirectWorker = shutdownWorker;