forbocai 0.3.18 → 0.4.0

package/README.md

@@ -57,14 +57,14 @@ git add . && git commit -m "chore: release vX.X.X" && git push
 
 The **ForbocAI SDK** is an **Engine-Agnostic** toolkit for creating autonomous AI-powered NPCs in any game or application. It runs anywhere (Node, Browser, Python, Rust).
 
-- **Local SLM Cortex** — Run **WebLLM** (SmolLM2, Llama3) directly in-browser via WebGPU.
+- **Local SLM Cortex** — Run quantized Small Language Models (SLMs) natively via **node-llama-cpp**.
 - **Autonomous NPCs** — Create agents with persona, memory, and validated actions.
 - **Persistent Memory** — Local vector storage for semantic recall.
 - **Portable Souls** — Permanent storage (Arweave) + Solana NFTs for tradeable agent identities.
 - **Automated QA** — Ghost Agents for headless testing at scale.
 
 **Requirements**:
-- Browser with **WebGPU** support (Chrome 113+, Edge 113+) for Cortex.
+- **Native Environment**: Node.js 18+ (Desktop/Server) or C++11 (Game Engine).
 - Node.js 18+ for CLI/Build.
 
 ---
@@ -164,7 +164,7 @@ console.log(response.dialogue);
 
 | Module | Description | Status |
 |--------|-------------|--------|
-| **Cortex** | Local SLM inference engine (Self-Contained) | ✅ Browser / 🚧 Node |
+| **Cortex** | Local SLM inference engine (Self-Contained) | ✅ Node (Native) / 🚧 C++ (UE) |
 | **Agent** | Autonomous entities with persona, state, memory | ✅ Complete |
 | **Memory** | Local Vector DB (IndexedDB / FS) | ✅ Complete |
 | **Bridge** | Neuro-symbolic action validation | ✅ Complete |

package/dist/index.d.mts

@@ -17,14 +17,12 @@ interface CortexStatus {
     id: string;
     model: string;
     ready: boolean;
-    engine: 'webllm' | 'mock' | 'remote' | 'node-llama-cpp';
+    engine: 'mock' | 'remote' | 'node-llama-cpp';
 }
 interface ICortex {
     init(): Promise<CortexStatus>;
     complete(prompt: string, options?: CompletionOptions): Promise<string>;
     completeStream(prompt: string, options?: CompletionOptions): AsyncGenerator<string>;
-    processObservation(observation: Observation): Promise<Directive>;
-    generateAction(directive: Directive): Promise<AgentAction>;
 }
 interface CompletionOptions {
     temperature?: number;

package/dist/index.d.ts

@@ -17,14 +17,12 @@ interface CortexStatus {
     id: string;
     model: string;
     ready: boolean;
-    engine: 'webllm' | 'mock' | 'remote' | 'node-llama-cpp';
+    engine: 'mock' | 'remote' | 'node-llama-cpp';
 }
 interface ICortex {
     init(): Promise<CortexStatus>;
     complete(prompt: string, options?: CompletionOptions): Promise<string>;
     completeStream(prompt: string, options?: CompletionOptions): AsyncGenerator<string>;
-    processObservation(observation: Observation): Promise<Directive>;
-    generateAction(directive: Directive): Promise<AgentAction>;
 }
 interface CompletionOptions {
     temperature?: number;

package/dist/index.js

@@ -202313,6 +202313,13 @@ var createNativeCortex = (config) => {
   let model;
   let context;
   let session;
+  let embeddingContext;
+  const createEmbeddingContext = async () => {
+    if (!model.embeddingSupported) {
+      throw new Error("Model does not support embeddings");
+    }
+    embeddingContext = await model.createEmbeddingContext();
+  };
   const MODELS_DIR = path.join(process.cwd(), "local_infrastructure", "models");
   const init3 = async () => {
     if (status.ready) return status;
@@ -202341,6 +202348,12 @@ var createNativeCortex = (config) => {
       });
       console.log("> Creating context...");
       context = await model.createContext();
+      if (model.embeddingSupported) {
+        console.log("> Creating embedding context...");
+        embeddingContext = await model.createEmbeddingContext();
+      } else {
+        console.log("> Model does not support embeddings");
+      }
       session = new LlamaChatSession({
         contextSequence: context.getSequence()
       });
@@ -202404,37 +202417,20 @@ var createNativeCortex = (config) => {
   const embed = async (text) => {
     if (!status.ready) await init3();
     try {
-      return new Array(384).fill(0).map(() => Math.random());
+      if (!embeddingContext) {
+        await createEmbeddingContext();
+      }
+      const embeddings = await embeddingContext.getEmbeddingFor(text);
+      return Array.from(embeddings);
     } catch (e2) {
-      return [];
-    }
-  };
-  const processObservation = async (obs) => {
-    const prompt = `System: You are an agent.
-Observation: ${obs.content}
-Task: Generete a JSON directive { "type": "...", "content": "..." }.`;
-    const res = await complete(prompt);
-    try {
-      return JSON.parse(res);
-    } catch {
-      return { type: "thought", content: res };
-    }
-  };
-  const generateAction = async (dir) => {
-    const prompt = `Directive: ${dir.content}. Generate JSON action.`;
-    const res = await complete(prompt);
-    try {
-      return JSON.parse(res);
-    } catch {
-      return { type: "idle", reason: res };
+      console.warn("Embedding generation failed, falling back to random vector:", e2);
+      return new Array(384).fill(0).map(() => Math.random());
     }
   };
   return {
     init: init3,
     complete,
     completeStream,
-    processObservation,
-    generateAction,
     embed
     // Extended interface for private use by Memory
   };
@@ -202831,9 +202827,23 @@ var initVectorEngine = async () => {
   }
 };
 var generateEmbedding = async (text) => {
-  if (!pipeline) await initVectorEngine();
-  const output = await pipeline(text, { pooling: "mean", normalize: true });
-  return Array.from(output.data);
+  const cortexConfig = {
+    model: "smollm2-135m",
+    temperature: 0.7,
+    maxTokens: 128,
+    gpu: true
+  };
+  const cortex = createCortex(cortexConfig);
+  try {
+    await cortex.init();
+    const embedding = await cortex.embed(text);
+    return embedding;
+  } catch (e2) {
+    console.warn("Embedding generation failed, falling back to transformers:", e2);
+    if (!pipeline) await initVectorEngine();
+    const output = await pipeline(text, { pooling: "mean", normalize: true });
+    return Array.from(output.data);
+  }
 };
 var getVectorTable = async (dbPath, tableName = "memories") => {
   if (!lancedb) await initVectorEngine();

package/dist/index.mjs

@@ -51,6 +51,13 @@ var createNativeCortex = (config) => {
   let model;
   let context;
   let session;
+  let embeddingContext;
+  const createEmbeddingContext = async () => {
+    if (!model.embeddingSupported) {
+      throw new Error("Model does not support embeddings");
+    }
+    embeddingContext = await model.createEmbeddingContext();
+  };
   const MODELS_DIR = path.join(process.cwd(), "local_infrastructure", "models");
   const init2 = async () => {
     if (status.ready) return status;
@@ -79,6 +86,12 @@ var createNativeCortex = (config) => {
       });
       console.log("> Creating context...");
       context = await model.createContext();
+      if (model.embeddingSupported) {
+        console.log("> Creating embedding context...");
+        embeddingContext = await model.createEmbeddingContext();
+      } else {
+        console.log("> Model does not support embeddings");
+      }
       session = new LlamaChatSession({
         contextSequence: context.getSequence()
       });
@@ -142,37 +155,20 @@ var createNativeCortex = (config) => {
   const embed = async (text) => {
     if (!status.ready) await init2();
     try {
-      return new Array(384).fill(0).map(() => Math.random());
+      if (!embeddingContext) {
+        await createEmbeddingContext();
+      }
+      const embeddings = await embeddingContext.getEmbeddingFor(text);
+      return Array.from(embeddings);
     } catch (e) {
-      return [];
-    }
-  };
-  const processObservation = async (obs) => {
-    const prompt = `System: You are an agent.
-Observation: ${obs.content}
-Task: Generete a JSON directive { "type": "...", "content": "..." }.`;
-    const res = await complete(prompt);
-    try {
-      return JSON.parse(res);
-    } catch {
-      return { type: "thought", content: res };
-    }
-  };
-  const generateAction = async (dir) => {
-    const prompt = `Directive: ${dir.content}. Generate JSON action.`;
-    const res = await complete(prompt);
-    try {
-      return JSON.parse(res);
-    } catch {
-      return { type: "idle", reason: res };
+      console.warn("Embedding generation failed, falling back to random vector:", e);
+      return new Array(384).fill(0).map(() => Math.random());
     }
   };
   return {
     init: init2,
     complete,
     completeStream,
-    processObservation,
-    generateAction,
     embed
     // Extended interface for private use by Memory
   };
@@ -569,9 +565,23 @@ var initVectorEngine = async () => {
   }
 };
 var generateEmbedding = async (text) => {
-  if (!pipeline) await initVectorEngine();
-  const output = await pipeline(text, { pooling: "mean", normalize: true });
-  return Array.from(output.data);
+  const cortexConfig = {
+    model: "smollm2-135m",
+    temperature: 0.7,
+    maxTokens: 128,
+    gpu: true
+  };
+  const cortex = createCortex(cortexConfig);
+  try {
+    await cortex.init();
+    const embedding = await cortex.embed(text);
+    return embedding;
+  } catch (e) {
+    console.warn("Embedding generation failed, falling back to transformers:", e);
+    if (!pipeline) await initVectorEngine();
+    const output = await pipeline(text, { pooling: "mean", normalize: true });
+    return Array.from(output.data);
+  }
 };
 var getVectorTable = async (dbPath, tableName = "memories") => {
   if (!lancedb) await initVectorEngine();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "forbocai",
-  "version": "0.3.18",
+  "version": "0.4.0",
   "license": "UNLICENSED",
   "description": "The Infrastructure Layer for Autonomous AI Characters",
   "main": "dist/index.js",
@@ -21,11 +21,9 @@
   "dependencies": {
     "@irys/sdk": "^0.1.1",
     "@lancedb/lancedb": "^0.23.0",
-    "@mlc-ai/web-llm": "^0.2.80",
     "@xenova/transformers": "^2.17.2",
     "apache-arrow": "^18.1.0",
     "axios": "^1.6.2",
-    "idb": "^8.0.3",
     "node-llama-cpp": "^3.15.1",
     "onnxruntime-node": "^1.23.2",
     "zod": "^3.22.4"