getmy-ruflo 3.5.46 → 3.5.47

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "getmy-ruflo",
-  "version": "3.5.46",
+  "version": "3.5.47",
   "description": "GetMy Ruflo - AI agent orchestration platform with real terminal execution, 259 MCP tools, 60+ agents, and swarm coordination",
   "main": "bin/ruflo.js",
   "type": "module",

package/v3/@claude-flow/cli/bin/cli.js

@@ -1,4 +1,4 @@
-#!/usr/bin/env node
+#!/usr/bin/env node
 /**
  * @claude-flow/cli - CLI Entry Point
  *
@@ -71,18 +71,12 @@ if (isMCPMode) {
     const params = message.params || {};
 
     switch (message.method) {
-      case 'initialize': {
-        // Negotiate protocol version: echo client's version if supported,
-        // otherwise fall back to the default version.
-        const supportedVersions = ['2024-11-05', '2025-11-05'];
-        const negotiatedVersion = supportedVersions.includes(params.protocolVersion)
-          ? params.protocolVersion
-          : '2024-11-05';
+      case 'initialize':
         return {
           jsonrpc: '2.0',
           id: message.id,
           result: {
-            protocolVersion: negotiatedVersion,
+            protocolVersion: '2024-11-05',
             serverInfo: { name: 'claude-flow', version: VERSION },
             capabilities: {
               tools: { listChanged: true },
@@ -90,7 +84,6 @@ if (isMCPMode) {
             },
           },
         };
-      }
 
       case 'tools/list': {
         const tools = listMCPTools();

package/v3/@claude-flow/cli/bin/mcp-server.js

@@ -1,4 +1,4 @@
-#!/usr/bin/env node
+#!/usr/bin/env node
 /**
  * @claude-flow/cli - MCP Server Entry Point
  *
@@ -107,18 +107,12 @@ async function handleMessage(message) {
 
   try {
     switch (message.method) {
-      case 'initialize': {
-        // Negotiate protocol version: echo client's version if supported,
-        // otherwise fall back to the default version.
-        const supportedVersions = ['2024-11-05', '2025-11-05'];
-        const negotiatedVersion = supportedVersions.includes(params.protocolVersion)
-          ? params.protocolVersion
-          : '2024-11-05';
+      case 'initialize':
         return {
           jsonrpc: '2.0',
           id: message.id,
           result: {
-            protocolVersion: negotiatedVersion,
+            protocolVersion: '2024-11-05',
             serverInfo: { name: 'claude-flow', version: VERSION },
             capabilities: {
               tools: { listChanged: true },
@@ -126,7 +120,6 @@ async function handleMessage(message) {
             },
           },
         };
-      }
 
       case 'tools/list': {
         const tools = listMCPTools();

package/v3/@claude-flow/cli/bin/mcp-sse-server.js

@@ -15,8 +15,7 @@ import { randomUUID, timingSafeEqual } from 'crypto';
 import { listMCPTools, callMCPTool, hasTool, initializePluginTools } from '../dist/src/mcp-client.js';
 
 const VERSION = '3.0.0';
-const SUPPORTED_PROTOCOL_VERSIONS = ['2024-11-05', '2025-11-05'];
-const DEFAULT_PROTOCOL_VERSION = '2024-11-05';
+const PROTOCOL_VERSION = '2025-11-05';
 const KEEPALIVE_INTERVAL_MS = 30_000;
 
 // ---------------------------------------------------------------------------
@@ -120,20 +119,12 @@ async function handleMessage(message, sessionId) {
         const sid = sessionId || `mcp-${Date.now()}-${randomUUID().slice(0, 8)}`;
         sessions.set(sid, { createdAt: Date.now() });
         metrics.sessionsCreated++;
-
-        // Negotiate protocol version: echo client's version if supported,
-        // otherwise fall back to the default version the client expects.
-        const requestedVersion = params.protocolVersion;
-        const negotiatedVersion = SUPPORTED_PROTOCOL_VERSIONS.includes(requestedVersion)
-          ? requestedVersion
-          : DEFAULT_PROTOCOL_VERSION;
-
         return {
           _sessionId: sid,
           jsonrpc: '2.0',
           id: message.id,
           result: {
-            protocolVersion: negotiatedVersion,
+            protocolVersion: PROTOCOL_VERSION,
             serverInfo: { name: 'claude-flow', version: VERSION },
             capabilities: {
               tools: { listChanged: true },
@@ -238,7 +229,7 @@ async function handleMessage(message, sessionId) {
 function broadcastSSE(sessionId, data) {
   const conns = sseConnections.get(sessionId);
   if (!conns) return;
-  const payload = `data: ${JSON.stringify(data)}\n\n`;
+  const payload = `event: message\ndata: ${JSON.stringify(data)}\n\n`;
   for (const conn of conns) {
     try {
       conn.write(payload);
@@ -381,7 +372,7 @@ const server = createServer(async (req, res) => {
       tools: tools.length,
       uptime: uptimeSeconds,
       sessions: sessions.size,
-      protocolVersion: DEFAULT_PROTOCOL_VERSION,
+      protocolVersion: PROTOCOL_VERSION,
     }));
     return;
   }
@@ -502,7 +493,7 @@ const server = createServer(async (req, res) => {
       headers['Cache-Control'] = 'no-cache';
       headers['Connection'] = 'keep-alive';
       res.writeHead(200, headers);
-      res.write(`data: ${JSON.stringify(response)}\n\n`);
+      res.write(`event: message\ndata: ${JSON.stringify(response)}\n\n`);
       res.end();
     } else {
       // Respond as plain JSON
@@ -602,7 +593,7 @@ server.listen(port, host, () => {
     platform: process.platform,
     port,
     protocol: 'streamable-http',
-    protocolVersion: DEFAULT_PROTOCOL_VERSION,
+    protocolVersion: PROTOCOL_VERSION,
     tools: tools.length,
     version: VERSION,
   }));

package/v3/@claude-flow/cli/dist/src/appliance/gguf-engine.d.ts

@@ -1,91 +0,0 @@
-/**
- * ruvLLM GGUF Inference Engine -- Pure Node.js GGUF Model Interface
- *
- * Provides:
- *   1. GGUF binary header parsing (metadata without loading weights)
- *   2. Model loading abstraction (node-llama-cpp when available, metadata-only fallback)
- *   3. Token generation interface with async iterator streaming
- *   4. KV-cache persistence to RVF-compatible binary format
- *
- * Zero external dependencies. node-llama-cpp is an optional peer.
- *
- * @module @claude-flow/cli/appliance/gguf-engine
- */
-export interface GgufMetadata {
-    magic: string;
-    version: number;
-    tensorCount: number;
-    kvCount: number;
-    architecture?: string;
-    name?: string;
-    contextLength?: number;
-    embeddingLength?: number;
-    blockCount?: number;
-    vocabSize?: number;
-    quantization?: string;
-    fileSize: number;
-    metadata: Record<string, unknown>;
-}
-export interface GgufEngineConfig {
-    contextSize?: number;
-    maxTokens?: number;
-    temperature?: number;
-    kvCachePath?: string;
-    verbose?: boolean;
-}
-export interface GenerateRequest {
-    prompt: string;
-    model?: string;
-    maxTokens?: number;
-    temperature?: number;
-    stream?: boolean;
-    stopSequences?: string[];
-}
-export interface GenerateResponse {
-    text: string;
-    model: string;
-    tokensUsed: number;
-    latencyMs: number;
-    metadataOnly: boolean;
-}
-/**
- * Parse the header and metadata from a GGUF file without loading tensors.
- * Reads only the first 256 KB of the file.
- */
-export declare function parseGgufHeader(path: string): Promise<GgufMetadata>;
-export declare class GgufEngine {
-    private config;
-    private llamaCpp;
-    private llamaModel;
-    private llamaContext;
-    private loadedModels;
-    private activeModelPath;
-    private kvCache;
-    constructor(config: GgufEngineConfig);
-    /** Probe for node-llama-cpp availability. */
-    initialize(): Promise<void>;
-    /** Parse GGUF header and optionally load the model for inference. */
-    loadModel(path: string): Promise<GgufMetadata>;
-    /** Generate text. Delegates to node-llama-cpp or returns a metadata-only stub. */
-    generate(request: GenerateRequest): Promise<GenerateResponse>;
-    /** Stream tokens via async iterator. Falls back to yielding full response. */
-    stream(request: GenerateRequest): AsyncGenerator<string>;
-    /**
-     * Persist the KV cache to an RVF-compatible binary file.
-     * Format: RVKV magic | version u32 | model SHA-256 (32B) | entry count u32
-     *         entries: [key_len u32, key, val_len u32, val] | footer SHA-256 (32B)
-     */
-    persistKvCache(outputPath: string): Promise<void>;
-    /** Restore KV cache from an RVF-compatible binary file. */
-    loadKvCache(inputPath: string): Promise<void>;
-    /** Return metadata for all loaded models. */
-    getLoadedModels(): GgufMetadata[];
-    /** Store a key-value pair in the in-memory KV cache. */
-    setKvEntry(key: string, value: Buffer): void;
-    /** Retrieve a key-value pair from the in-memory KV cache. */
-    getKvEntry(key: string): Buffer | undefined;
-    /** Release resources, unload models, and optionally persist the KV cache. */
-    shutdown(): Promise<void>;
-    private tryLoadLlamaCpp;
-}
-//# sourceMappingURL=gguf-engine.d.ts.map

package/v3/@claude-flow/cli/dist/src/appliance/gguf-engine.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"gguf-engine.d.ts","sourceRoot":"","sources":["../../../src/appliance/gguf-engine.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAmBH,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACnC;AAED,MAAM,WAAW,gBAAgB;IAC/B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,OAAO,CAAC,EAAE,OAAO,CAAC;CACnB;AAED,MAAM,WAAW,eAAe;IAC9B,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;CAC1B;AAED,MAAM,WAAW,gBAAgB;IAC/B,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,YAAY,EAAE,OAAO,CAAC;CACvB;AA6ED;;;GAGG;AACH,wBAAsB,eAAe,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,YAAY,CAAC,CAWzE;AAyED,qBAAa,UAAU;IACrB,OAAO,CAAC,MAAM,CAA6B;IAC3C,OAAO,CAAC,QAAQ,CAAa;IAC7B,OAAO,CAAC,UAAU,CAAa;IAC/B,OAAO,CAAC,YAAY,CAAa;IACjC,OAAO,CAAC,YAAY,CAAwC;IAC5D,OAAO,CAAC,eAAe,CAAuB;IAC9C,OAAO,CAAC,OAAO,CAAkC;gBAErC,MAAM,EAAE,gBAAgB;IAUpC,6CAA6C;IACvC,UAAU,IAAI,OAAO,CAAC,IAAI,CAAC;IAOjC,qEAAqE;IAC/D,SAAS,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,YAAY,CAAC;IAqBpD,kFAAkF;IAC5E,QAAQ,CAAC,OAAO,EAAE,eAAe,GAAG,OAAO,CAAC,gBAAgB,CAAC;IA6CnE,8EAA8E;IACvE,MAAM,CAAC,OAAO,EAAE,eAAe,GAAG,cAAc,CAAC,MAAM,CAAC;IAuB/D;;;;OAIG;IACG,cAAc,CAAC,UAAU,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;IA0BvD,2DAA2D;IACrD,WAAW,CAAC,SAAS,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;IAmCnD,6CAA6C;IAC7C,eAAe,IAAI,YAAY,EAAE;IAEjC,wDAAwD;IACxD,UAAU,CAAC,GAAG,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,IAAI;IAE5C,6DAA6D;IAC7D,UAAU,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS;IAE3C,6EAA6E;IACvE,QAAQ,IAAI,OAAO,CAAC,IAAI,CAAC;YAiBjB,eAAe;CAI9B"}

package/v3/@claude-flow/cli/dist/src/appliance/gguf-engine.js

@@ -1,425 +0,0 @@
-/**
- * ruvLLM GGUF Inference Engine -- Pure Node.js GGUF Model Interface
- *
- * Provides:
- *   1. GGUF binary header parsing (metadata without loading weights)
- *   2. Model loading abstraction (node-llama-cpp when available, metadata-only fallback)
- *   3. Token generation interface with async iterator streaming
- *   4. KV-cache persistence to RVF-compatible binary format
- *
- * Zero external dependencies. node-llama-cpp is an optional peer.
- *
- * @module @claude-flow/cli/appliance/gguf-engine
- */
-import { open, readFile, writeFile, stat as fsStat } from 'node:fs/promises';
-import { createHash } from 'node:crypto';
-import { basename } from 'node:path';
-// ── GGUF Metadata Value Types ───────────────────────────────
-var GgufValueType;
-(function (GgufValueType) {
-    GgufValueType[GgufValueType["UINT8"] = 0] = "UINT8";
-    GgufValueType[GgufValueType["INT8"] = 1] = "INT8";
-    GgufValueType[GgufValueType["UINT16"] = 2] = "UINT16";
-    GgufValueType[GgufValueType["INT16"] = 3] = "INT16";
-    GgufValueType[GgufValueType["UINT32"] = 4] = "UINT32";
-    GgufValueType[GgufValueType["INT32"] = 5] = "INT32";
-    GgufValueType[GgufValueType["FLOAT32"] = 6] = "FLOAT32";
-    GgufValueType[GgufValueType["BOOL"] = 7] = "BOOL";
-    GgufValueType[GgufValueType["STRING"] = 8] = "STRING";
-    GgufValueType[GgufValueType["ARRAY"] = 9] = "ARRAY";
-    GgufValueType[GgufValueType["UINT64"] = 10] = "UINT64";
-    GgufValueType[GgufValueType["INT64"] = 11] = "INT64";
-    GgufValueType[GgufValueType["FLOAT64"] = 12] = "FLOAT64";
-})(GgufValueType || (GgufValueType = {}));
-const GGUF_MAGIC = 0x46554747; // "GGUF" in little-endian
-const RVKV_MAGIC = 0x564B5652; // "RVKV" in little-endian
-const RVKV_VERSION = 1;
-// ── Internal Buffer Reader ──────────────────────────────────
-/** Stateful cursor over a Buffer for sequential binary reads. */
-class BufferReader {
-    buf;
-    offset = 0;
-    constructor(buf) {
-        this.buf = buf;
-    }
-    get remaining() { return this.buf.length - this.offset; }
-    readU8() { const v = this.buf.readUInt8(this.offset); this.offset += 1; return v; }
-    readI8() { const v = this.buf.readInt8(this.offset); this.offset += 1; return v; }
-    readU16() { const v = this.buf.readUInt16LE(this.offset); this.offset += 2; return v; }
-    readI16() { const v = this.buf.readInt16LE(this.offset); this.offset += 2; return v; }
-    readU32() { const v = this.buf.readUInt32LE(this.offset); this.offset += 4; return v; }
-    readI32() { const v = this.buf.readInt32LE(this.offset); this.offset += 4; return v; }
-    readF32() { const v = this.buf.readFloatLE(this.offset); this.offset += 4; return v; }
-    readF64() { const v = this.buf.readDoubleLE(this.offset); this.offset += 8; return v; }
-    readU64() { const v = this.buf.readBigUInt64LE(this.offset); this.offset += 8; return v; }
-    readI64() { const v = this.buf.readBigInt64LE(this.offset); this.offset += 8; return v; }
-    /** Safe for values up to 2^53. Real GGUF files never exceed this for tensor/kv counts. */
-    readU64AsNumber() { return Number(this.readU64()); }
-    readBool() { return this.readU8() !== 0; }
-    /** GGUF string: [length u64 LE][utf-8 bytes]. */
-    readString() {
-        const len = this.readU64AsNumber();
-        if (len === 0)
-            return '';
-        if (len > this.remaining)
-            throw new Error(`String length ${len} exceeds remaining buffer`);
-        const s = this.buf.toString('utf-8', this.offset, this.offset + len);
-        this.offset += len;
-        return s;
-    }
-}
-// ── GGUF Value Reading ──────────────────────────────────────
-/** Read a typed scalar from the buffer (shared by value and array-element readers). */
-function readScalar(reader, t) {
-    switch (t) {
-        case GgufValueType.UINT8: return reader.readU8();
-        case GgufValueType.INT8: return reader.readI8();
-        case GgufValueType.UINT16: return reader.readU16();
-        case GgufValueType.INT16: return reader.readI16();
-        case GgufValueType.UINT32: return reader.readU32();
-        case GgufValueType.INT32: return reader.readI32();
-        case GgufValueType.FLOAT32: return reader.readF32();
-        case GgufValueType.BOOL: return reader.readBool();
-        case GgufValueType.STRING: return reader.readString();
-        case GgufValueType.UINT64: return Number(reader.readU64());
-        case GgufValueType.INT64: return Number(reader.readI64());
-        case GgufValueType.FLOAT64: return reader.readF64();
-        default: return undefined;
-    }
-}
-/** Read a single GGUF typed value (scalar or array) from the buffer. */
-function readGgufValue(reader) {
-    const valueType = reader.readU32();
-    if (valueType === GgufValueType.ARRAY) {
-        const elemType = reader.readU32();
-        const len = reader.readU64AsNumber();
-        const arr = [];
-        for (let i = 0; i < len; i++) {
-            const v = readScalar(reader, elemType);
-            if (v === undefined)
-                throw new Error(`Unknown GGUF array element type: ${elemType}`);
-            arr.push(v);
-        }
-        return arr;
-    }
-    const v = readScalar(reader, valueType);
-    if (v === undefined)
-        throw new Error(`Unknown GGUF value type: ${valueType}`);
-    return v;
-}
-// ── GGUF Header Parsing ─────────────────────────────────────
-/**
- * Parse the header and metadata from a GGUF file without loading tensors.
- * Reads only the first 256 KB of the file.
- */
-export async function parseGgufHeader(path) {
-    const fileInfo = await fsStat(path);
-    const readSize = Math.min(fileInfo.size, 256 * 1024);
-    const fh = await open(path, 'r');
-    try {
-        const buf = Buffer.alloc(readSize);
-        await fh.read(buf, 0, readSize, 0);
-        return parseGgufBuffer(buf, fileInfo.size, path);
-    }
-    finally {
-        await fh.close();
-    }
-}
-function parseGgufBuffer(buf, fileSize, filePath) {
-    const reader = new BufferReader(buf);
-    const magic = reader.readU32();
-    if (magic !== GGUF_MAGIC) {
-        throw new Error(`Invalid GGUF magic: 0x${magic.toString(16)} (expected 0x${GGUF_MAGIC.toString(16)})`);
-    }
-    const version = reader.readU32();
-    if (version < 2 || version > 3) {
-        throw new Error(`Unsupported GGUF version: ${version} (expected 2 or 3)`);
-    }
-    const tensorCount = reader.readU64AsNumber();
-    const kvCount = reader.readU64AsNumber();
-    const metadata = {};
-    for (let i = 0; i < kvCount; i++) {
-        if (reader.remaining < 12)
-            break;
-        try {
-            const key = reader.readString();
-            metadata[key] = readGgufValue(reader);
-        }
-        catch {
-            break; // reached end of read window
-        }
-    }
-    const arch = asString(metadata['general.architecture']);
-    const pfx = arch || 'llama'; // fallback prefix for well-known keys
-    return {
-        magic: 'GGUF', version, tensorCount, kvCount,
-        architecture: arch,
-        name: asString(metadata['general.name']),
-        contextLength: asNumber(metadata[`${pfx}.context_length`]),
-        embeddingLength: asNumber(metadata[`${pfx}.embedding_length`]),
-        blockCount: asNumber(metadata[`${pfx}.block_count`]),
-        vocabSize: inferVocabSize(metadata),
-        quantization: inferQuantFromMetadata(metadata, filePath),
-        fileSize, metadata,
-    };
-}
-// ── Metadata Helpers ────────────────────────────────────────
-function asString(v) { return typeof v === 'string' ? v : undefined; }
-function asNumber(v) { return typeof v === 'number' ? v : undefined; }
-const QUANT_RE = [
-    [/q2_k/i, 'Q2_K'], [/q3_k_s/i, 'Q3_K_S'], [/q3_k_m/i, 'Q3_K_M'], [/q3_k_l/i, 'Q3_K_L'],
-    [/q4_k_s/i, 'Q4_K_S'], [/q4_k_m/i, 'Q4_K_M'], [/q4_0/i, 'Q4_0'], [/q4_1/i, 'Q4_1'],
-    [/q5_k_s/i, 'Q5_K_S'], [/q5_k_m/i, 'Q5_K_M'], [/q5_0/i, 'Q5_0'], [/q5_1/i, 'Q5_1'],
-    [/q6_k/i, 'Q6_K'], [/q8_0/i, 'Q8_0'], [/f16/i, 'F16'], [/f32/i, 'F32'],
-];
-function inferQuantFromMetadata(meta, filePath) {
-    const ft = meta['general.file_type'];
-    if (typeof ft === 'number')
-        return `file_type_${ft}`;
-    const name = basename(filePath);
-    for (const [re, label] of QUANT_RE)
-        if (re.test(name))
-            return label;
-    return 'unknown';
-}
-function inferVocabSize(meta) {
-    const tokens = meta['tokenizer.ggml.tokens'];
-    if (Array.isArray(tokens))
-        return tokens.length;
-    return asNumber(meta['tokenizer.ggml.vocab_size']);
-}
-// ── GGUF Engine ─────────────────────────────────────────────
-export class GgufEngine {
-    config;
-    llamaCpp = null;
-    llamaModel = null;
-    llamaContext = null;
-    loadedModels = new Map();
-    activeModelPath = null;
-    kvCache = new Map();
-    constructor(config) {
-        this.config = {
-            contextSize: config.contextSize ?? 4096,
-            maxTokens: config.maxTokens ?? 512,
-            temperature: config.temperature ?? 0.7,
-            kvCachePath: config.kvCachePath ?? '',
-            verbose: config.verbose ?? false,
-        };
-    }
-    /** Probe for node-llama-cpp availability. */
-    async initialize() {
-        this.llamaCpp = await this.tryLoadLlamaCpp();
-        if (this.config.verbose) {
-            console.log(`[gguf-engine] node-llama-cpp: ${this.llamaCpp ? 'available' : 'not found (metadata-only mode)'}`);
-        }
-    }
-    /** Parse GGUF header and optionally load the model for inference. */
-    async loadModel(path) {
-        const meta = await parseGgufHeader(path);
-        this.loadedModels.set(path, meta);
-        this.activeModelPath = path;
-        if (this.llamaCpp) {
-            try {
-                const { getLlama } = this.llamaCpp;
-                const llama = await getLlama();
-                this.llamaModel = await llama.loadModel({ modelPath: path });
-                this.llamaContext = await this.llamaModel.createContext({ contextSize: this.config.contextSize });
-                if (this.config.verbose)
-                    console.log(`[gguf-engine] Model loaded: ${basename(path)}`);
-            }
-            catch (err) {
-                if (this.config.verbose)
-                    console.warn('[gguf-engine] node-llama-cpp load failed:', err);
-                this.llamaModel = null;
-                this.llamaContext = null;
-            }
-        }
-        return meta;
-    }
-    /** Generate text. Delegates to node-llama-cpp or returns a metadata-only stub. */
-    async generate(request) {
-        const start = performance.now();
-        const modelPath = request.model ?? this.activeModelPath;
-        const modelName = modelPath ? basename(modelPath) : 'none';
-        if (this.llamaContext && this.llamaModel) {
-            try {
-                const session = new this.llamaCpp.LlamaChatSession({
-                    contextSequence: this.llamaContext.getSequence(),
-                });
-                const text = await session.prompt(request.prompt, {
-                    maxTokens: request.maxTokens ?? this.config.maxTokens,
-                    temperature: request.temperature ?? this.config.temperature,
-                    stopGenerationTrigger: request.stopSequences
-                        ? request.stopSequences.map((s) => new this.llamaCpp.LlamaText([s]))
-                        : undefined,
-                });
-                // Use llama.cpp tokenizer for accurate count when available, else estimate
-                let tokensUsed;
-                try {
-                    const seq = this.llamaContext.getSequence();
-                    tokensUsed = seq.tokenCount ?? Math.ceil(text.length / 4);
-                }
-                catch {
-                    tokensUsed = Math.ceil(text.length / 4); // ~4 chars per token heuristic
-                }
-                return {
-                    text, model: modelName, tokensUsed,
-                    latencyMs: performance.now() - start, metadataOnly: false,
-                };
-            }
-            catch (err) {
-                if (this.config.verbose)
-                    console.warn('[gguf-engine] Generation failed:', err);
-            }
-        }
-        // Metadata-only fallback
-        const meta = modelPath ? this.loadedModels.get(modelPath) : undefined;
-        return {
-            text: meta
-                ? `[metadata-only] Model: ${meta.name ?? modelName}, arch: ${meta.architecture ?? 'unknown'}, ctx: ${meta.contextLength ?? 'unknown'}`
-                : '[metadata-only] No model loaded',
-            model: modelName, tokensUsed: 0,
-            latencyMs: performance.now() - start, metadataOnly: true,
-        };
-    }
-    /** Stream tokens via async iterator. Falls back to yielding full response. */
-    async *stream(request) {
-        if (this.llamaContext && this.llamaModel) {
-            try {
-                const session = new this.llamaCpp.LlamaChatSession({
-                    contextSequence: this.llamaContext.getSequence(),
-                });
-                const it = session.promptWithMeta(request.prompt, {
-                    maxTokens: request.maxTokens ?? this.config.maxTokens,
-                    temperature: request.temperature ?? this.config.temperature,
-                });
-                if (it && typeof it[Symbol.asyncIterator] === 'function') {
-                    for await (const chunk of it) {
-                        if (typeof chunk === 'string')
-                            yield chunk;
-                        else if (chunk?.text)
-                            yield chunk.text;
-                    }
-                    return;
-                }
-            }
-            catch { /* fall through to single-chunk fallback */ }
-        }
-        const response = await this.generate(request);
-        yield response.text;
-    }
-    /**
-     * Persist the KV cache to an RVF-compatible binary file.
-     * Format: RVKV magic | version u32 | model SHA-256 (32B) | entry count u32
-     *         entries: [key_len u32, key, val_len u32, val] | footer SHA-256 (32B)
-     */
-    async persistKvCache(outputPath) {
-        const path = outputPath || this.config.kvCachePath;
-        if (!path)
-            throw new Error('No KV cache output path specified');
-        const modelHash = createHash('sha256').update(this.activeModelPath ?? 'no-model').digest();
-        const entryBufs = [];
-        for (const [key, value] of this.kvCache) {
-            const keyBuf = Buffer.from(key, 'utf-8');
-            const hdr = Buffer.alloc(8);
-            hdr.writeUInt32LE(keyBuf.length, 0);
-            hdr.writeUInt32LE(value.length, 4);
-            entryBufs.push(hdr, keyBuf, value);
-        }
-        const entryData = Buffer.concat(entryBufs);
-        const footer = createHash('sha256').update(entryData).digest();
-        const header = Buffer.alloc(44);
-        header.writeUInt32LE(RVKV_MAGIC, 0);
-        header.writeUInt32LE(RVKV_VERSION, 4);
-        modelHash.copy(header, 8);
-        header.writeUInt32LE(this.kvCache.size, 40);
-        await writeFile(path, Buffer.concat([header, entryData, footer]));
-        if (this.config.verbose)
-            console.log(`[gguf-engine] KV cache persisted: ${this.kvCache.size} entries`);
-    }
-    /** Restore KV cache from an RVF-compatible binary file. */
-    async loadKvCache(inputPath) {
-        const data = await readFile(inputPath);
-        if (data.length < 44)
-            throw new Error('KV cache file too small');
-        const magic = data.readUInt32LE(0);
-        if (magic !== RVKV_MAGIC)
-            throw new Error(`Invalid KV cache magic: 0x${magic.toString(16)}`);
-        const version = data.readUInt32LE(4);
-        if (version !== RVKV_VERSION)
-            throw new Error(`Unsupported KV cache version: ${version}`);
-        const entryCount = data.readUInt32LE(40);
-        let offset = 44;
-        const entries = new Map();
-        for (let i = 0; i < entryCount; i++) {
-            if (offset + 8 > data.length)
-                throw new Error('KV cache file truncated');
-            const keyLen = data.readUInt32LE(offset);
-            const valLen = data.readUInt32LE(offset + 4);
-            offset += 8;
-            if (offset + keyLen + valLen > data.length)
-                throw new Error('KV cache file truncated');
-            entries.set(data.toString('utf-8', offset, offset + keyLen), Buffer.from(data.subarray(offset + keyLen, offset + keyLen + valLen)));
-            offset += keyLen + valLen;
-        }
-        // Verify footer hash (mandatory)
-        if (offset + 32 > data.length) {
-            throw new Error('KV cache file missing SHA256 footer');
-        }
-        const stored = data.subarray(offset, offset + 32);
-        const computed = createHash('sha256').update(data.subarray(44, offset)).digest();
-        if (!stored.equals(computed))
-            throw new Error('KV cache integrity check failed: hash mismatch');
-        this.kvCache = entries;
-        if (this.config.verbose)
-            console.log(`[gguf-engine] KV cache loaded: ${entries.size} entries`);
-    }
-    /** Return metadata for all loaded models. */
-    getLoadedModels() { return Array.from(this.loadedModels.values()); }
-    /** Store a key-value pair in the in-memory KV cache. */
-    setKvEntry(key, value) { this.kvCache.set(key, value); }
-    /** Retrieve a key-value pair from the in-memory KV cache. */
-    getKvEntry(key) { return this.kvCache.get(key); }
-    /** Release resources, unload models, and optionally persist the KV cache. */
-    async shutdown() {
-        if (this.config.kvCachePath && this.kvCache.size > 0) {
-            try {
-                await this.persistKvCache(this.config.kvCachePath);
-            }
-            catch (err) {
-                if (this.config.verbose)
-                    console.warn('[gguf-engine] KV persist failed:', err);
-            }
-        }
-        if (this.llamaContext?.dispose) {
-            try {
-                await this.llamaContext.dispose();
-            }
-            catch { /* ignore */ }
-        }
-        if (this.llamaModel?.dispose) {
-            try {
-                await this.llamaModel.dispose();
-            }
-            catch { /* ignore */ }
-        }
-        this.llamaContext = null;
-        this.llamaModel = null;
-        this.activeModelPath = null;
-        this.loadedModels.clear();
-        this.kvCache.clear();
-        if (this.config.verbose)
-            console.log('[gguf-engine] Shutdown complete');
-    }
-    // ── Private ───────────────────────────────────────────────
-    async tryLoadLlamaCpp() {
-        // @ts-ignore -- optional peer dependency, may not be installed
-        try {
-            return await import('node-llama-cpp');
-        }
-        catch {
-            return null;
-        }
-    }
-}
-//# sourceMappingURL=gguf-engine.js.map