@gmickel/gno 0.15.0 → 0.16.0

package/README.md

@@ -32,7 +32,14 @@ GNO is a local knowledge engine that turns your documents into a searchable, con
 
 ---
 
-## What's New in v0.13
+## What's New in v0.15
+
+- **HTTP Backends**: Offload embedding, reranking, and generation to remote GPU servers
+- Simple URI config: `http://host:port/path#modelname`
+- Works with llama-server, Ollama, LocalAI, vLLM
+- Run GNO on lightweight machines while GPU inference runs on your network
+
+### v0.13
 
 - **Knowledge Graph**: Interactive force-directed visualization of document connections
 - **Graph with Similarity**: See semantic similarity as golden edges (not just wiki/markdown links)
@@ -343,28 +350,29 @@ graph TD
 
 ## Features
 
-| Feature             | Description                                                                    |
-| :------------------ | :----------------------------------------------------------------------------- |
-| **Hybrid Search**   | BM25 + vector + RRF fusion + cross-encoder reranking                           |
-| **Document Editor** | Create, edit, delete docs with live markdown preview                           |
-| **Web UI**          | Visual dashboard for search, browse, edit, and AI Q&A                          |
-| **REST API**        | HTTP API for custom tools and integrations                                     |
-| **Multi-Format**    | Markdown, PDF, DOCX, XLSX, PPTX, plain text                                    |
-| **Local LLM**       | AI answers via llama.cpp, no API keys                                          |
-| **Privacy First**   | 100% offline, zero telemetry, your data stays yours                            |
-| **MCP Server**      | Works with Claude Desktop, Cursor, Zed, + 8 more                               |
-| **Collections**     | Organize sources with patterns, excludes, contexts                             |
-| **Tag Filtering**   | Frontmatter tags with hierarchical paths, filter via `--tags-any`/`--tags-all` |
-| **Note Linking**    | Wiki links, backlinks, related notes, cross-collection navigation              |
-| **Multilingual**    | 30+ languages, auto-detection, cross-lingual search                            |
-| **Incremental**     | SHA-256 tracking, only changed files re-indexed                                |
-| **Keyboard First**  | ⌘N capture, ⌘K search, ⌘/ shortcuts, ⌘S save                                   |
+| Feature              | Description                                                                    |
+| :------------------- | :----------------------------------------------------------------------------- |
+| **Hybrid Search**    | BM25 + vector + RRF fusion + cross-encoder reranking                           |
+| **Document Editor**  | Create, edit, delete docs with live markdown preview                           |
+| **Web UI**           | Visual dashboard for search, browse, edit, and AI Q&A                          |
+| **REST API**         | HTTP API for custom tools and integrations                                     |
+| **Multi-Format**     | Markdown, PDF, DOCX, XLSX, PPTX, plain text                                    |
+| **Local LLM**        | AI answers via llama.cpp, no API keys                                          |
+| **Remote Inference** | Offload to GPU servers via HTTP (llama-server, Ollama, LocalAI)                |
+| **Privacy First**    | 100% offline, zero telemetry, your data stays yours                            |
+| **MCP Server**       | Works with Claude Desktop, Cursor, Zed, + 8 more                               |
+| **Collections**      | Organize sources with patterns, excludes, contexts                             |
+| **Tag Filtering**    | Frontmatter tags with hierarchical paths, filter via `--tags-any`/`--tags-all` |
+| **Note Linking**     | Wiki links, backlinks, related notes, cross-collection navigation              |
+| **Multilingual**     | 30+ languages, auto-detection, cross-lingual search                            |
+| **Incremental**      | SHA-256 tracking, only changed files re-indexed                                |
+| **Keyboard First**   | ⌘N capture, ⌘K search, ⌘/ shortcuts, ⌘S save                                   |
 
 ---
 
 ## Local Models
 
-Models auto-download on first use to `~/.cache/gno/models/`.
+Models auto-download on first use to `~/.cache/gno/models/`. Alternatively, offload to a GPU server on your network using HTTP backends.
 
 | Model               | Purpose                               | Size         |
 | :------------------ | :------------------------------------ | :----------- |
@@ -385,6 +393,24 @@ gno models use slim
 gno models pull --all  # Optional: pre-download models (auto-downloads on first use)
 ```
 
+### HTTP Backends (Remote GPU)
+
+Offload inference to a GPU server on your network:
+
+```yaml
+# ~/.config/gno/config.yaml
+models:
+  activePreset: remote-gpu
+  presets:
+    - id: remote-gpu
+      name: Remote GPU Server
+      embed: "http://192.168.1.100:8081/v1/embeddings#bge-m3"
+      rerank: "http://192.168.1.100:8082/v1/completions#reranker"
+      gen: "http://192.168.1.100:8083/v1/chat/completions#qwen3-4b"
+```
+
+Works with llama-server, Ollama, LocalAI, vLLM, or any OpenAI-compatible server.
+
 > **Configuration**: [Model Setup](https://gno.sh/docs/CONFIGURATION/)
 
 ---

package/assets/skill/examples.md

@@ -249,6 +249,7 @@ gno similar gno://notes/auth.md --cross-collection
 
 ```markdown
 # In your documents:
+
 See [[API Design]] for details.
 Check [[work:Project Plan]] for cross-collection link.
 Read [[Security#OAuth]] for specific section.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gmickel/gno",
-  "version": "0.15.0",
+  "version": "0.16.0",
   "description": "Local semantic search for your documents. Index Markdown, PDF, and Office files with hybrid BM25 + vector search.",
   "keywords": [
     "embeddings",
@@ -95,12 +95,12 @@
     "commander": "^14.0.2",
     "embla-carousel-react": "^8.6.0",
     "franc": "^6.2.0",
-    "lucide-react": "^0.562.0",
-    "markitdown-ts": "^0.0.8",
+    "lucide-react": "^0.563.0",
+    "markitdown-ts": "^0.0.9",
     "minimatch": "^10.1.1",
     "nanoid": "^5.1.6",
     "node-llama-cpp": "^3.14.5",
-    "officeparser": "^5.2.2",
+    "officeparser": "^6.0.4",
     "picocolors": "^1.1.1",
     "react": "^19.2.3",
     "react-dom": "^19.2.3",
@@ -118,7 +118,7 @@
   },
   "devDependencies": {
     "@ai-sdk/openai": "^3.0.2",
-    "@biomejs/biome": "2.3.10",
+    "@biomejs/biome": "2.3.13",
     "@tailwindcss/cli": "^4.1.18",
     "@types/bun": "latest",
     "@types/react": "^19.2.7",
@@ -129,13 +129,13 @@
     "evalite": "^1.0.0-beta.15",
     "exceljs": "^4.4.0",
     "lefthook": "^2.0.13",
-    "oxfmt": "^0.21.0",
-    "oxlint": "^1.36.0",
-    "oxlint-tsgolint": "^0.10.1",
+    "oxfmt": "^0.28.0",
+    "oxlint": "^1.42.0",
+    "oxlint-tsgolint": "^0.11.4",
     "pdf-lib": "^1.17.1",
     "playwright": "^1.52.0",
     "pptxgenjs": "^4.0.1",
-    "ultracite": "7.0.4",
+    "ultracite": "7.1.3",
     "vitest": "^4.0.16"
   },
   "peerDependencies": {

package/src/cli/AGENTS.md

@@ -54,9 +54,9 @@ program
 
 ```typescript
 export const EXIT = {
-  SUCCESS: 0,    // Command completed successfully
+  SUCCESS: 0, // Command completed successfully
   VALIDATION: 1, // Bad args, missing params
-  RUNTIME: 2,    // IO, DB, model, network errors
+  RUNTIME: 2, // IO, DB, model, network errors
 } as const;
 ```
 

package/src/cli/CLAUDE.md

@@ -54,9 +54,9 @@ program
 
 ```typescript
 export const EXIT = {
-  SUCCESS: 0,    // Command completed successfully
+  SUCCESS: 0, // Command completed successfully
   VALIDATION: 1, // Bad args, missing params
-  RUNTIME: 2,    // IO, DB, model, network errors
+  RUNTIME: 2, // IO, DB, model, network errors
 } as const;
 ```
 

package/src/converters/adapters/officeparser/adapter.ts

@@ -1,9 +1,9 @@
 /**
  * officeparser adapter for PPTX conversion.
- * Uses parseOfficeAsync() with Buffer for in-memory extraction.
+ * Uses parseOffice() v6 API with Buffer for in-memory extraction.
  */
 
-import { parseOfficeAsync } from "officeparser";
+import { parseOffice } from "officeparser";
 
 import type {
   Converter,
@@ -76,10 +76,12 @@ export const officeparserAdapter: Converter = {
     try {
       // Zero-copy Buffer view (input.bytes is immutable by contract)
       const buffer = toBuffer(input.bytes);
-      const text = await parseOfficeAsync(buffer, {
+      // v6 API: parseOffice returns AST, use .toText() for plain text
+      const ast = await parseOffice(buffer, {
         newlineDelimiter: "\n",
         ignoreNotes: false, // Include speaker notes
       });
+      const text = ast.toText();
 
       if (!text || text.trim().length === 0) {
         return {

package/src/converters/versions.ts

@@ -20,5 +20,5 @@ export const NATIVE_VERSIONS = {
  */
 export const ADAPTER_VERSIONS = {
   "markitdown-ts": "0.0.8",
-  officeparser: "5.2.0",
+  officeparser: "6.0.4",
 } as const;

package/src/llm/nodeLlamaCpp/embedding.ts

@@ -24,6 +24,15 @@ type LlamaEmbeddingContext = Awaited<
   ReturnType<LlamaModel["createEmbeddingContext"]>
 >;
 
+// ─────────────────────────────────────────────────────────────────────────────
+// Constants
+// ─────────────────────────────────────────────────────────────────────────────
+
+// Max concurrent embedding operations per batch to avoid overwhelming the context.
+// node-llama-cpp contexts may not handle high concurrency well; this provides
+// a safe default while still allowing parallelism within chunks.
+const MAX_CONCURRENT_EMBEDDINGS = 16;
+
 // ─────────────────────────────────────────────────────────────────────────────
 // Implementation
 // ─────────────────────────────────────────────────────────────────────────────
@@ -78,19 +87,53 @@ export class NodeLlamaCppEmbedding implements EmbeddingPort {
       return ctx;
     }
 
+    if (texts.length === 0) {
+      return { ok: true, value: [] };
+    }
+
     try {
-      const results: number[][] = [];
-      for (const text of texts) {
-        const embedding = await ctx.value.getEmbeddingFor(text);
-        const vector = Array.from(embedding.vector) as number[];
-        results.push(vector);
-
-        // Cache dimensions on first call
-        if (this.dims === null) {
-          this.dims = vector.length;
+      // Process in chunks to avoid overwhelming the embedding context.
+      // node-llama-cpp v3.x only exposes getEmbeddingFor (single text), not a native
+      // batch method. We use allSettled within chunks to ensure all in-flight ops
+      // complete before returning (prevents orphaned operations on early failure).
+      const allResults: number[][] = [];
+
+      for (let i = 0; i < texts.length; i += MAX_CONCURRENT_EMBEDDINGS) {
+        const chunk = texts.slice(i, i + MAX_CONCURRENT_EMBEDDINGS);
+        const settled = await Promise.allSettled(
+          chunk.map((text) => ctx.value.getEmbeddingFor(text))
+        );
+
+        // Check for any failures in this chunk
+        const firstRejection = settled.find(
+          (r): r is PromiseRejectedResult => r.status === "rejected"
+        );
+        if (firstRejection) {
+          return {
+            ok: false,
+            error: inferenceFailedError(this.modelUri, firstRejection.reason),
+          };
         }
+
+        // Extract results from this chunk (cast safe after rejection check)
+        const chunkResults = (
+          settled as Array<
+            PromiseFulfilledResult<
+              Awaited<ReturnType<typeof ctx.value.getEmbeddingFor>>
+            >
+          >
+        ).map((r) => Array.from(r.value.vector) as number[]);
+
+        allResults.push(...chunkResults);
       }
-      return { ok: true, value: results };
+
+      // Cache dimensions from first result
+      const firstResult = allResults[0];
+      if (this.dims === null && firstResult !== undefined) {
+        this.dims = firstResult.length;
+      }
+
+      return { ok: true, value: allResults };
     } catch (e) {
       return { ok: false, error: inferenceFailedError(this.modelUri, e) };
     }

package/src/mcp/AGENTS.md

@@ -40,14 +40,16 @@ export const toolName: Tool = {
   description: "What this tool does",
   inputSchema: {
     type: "object",
-    properties: { /* ... */ },
+    properties: {
+      /* ... */
+    },
     required: ["query"],
   },
 };
 
 export async function handleToolName(
   args: ToolArgs,
-  store: SqliteAdapter,
+  store: SqliteAdapter
   // ... other ports
 ): Promise<CallToolResult> {
   // 1. Validate args

package/src/mcp/CLAUDE.md

@@ -40,14 +40,16 @@ export const toolName: Tool = {
   description: "What this tool does",
   inputSchema: {
     type: "object",
-    properties: { /* ... */ },
+    properties: {
+      /* ... */
+    },
     required: ["query"],
   },
 };
 
 export async function handleToolName(
   args: ToolArgs,
-  store: SqliteAdapter,
+  store: SqliteAdapter
   // ... other ports
 ): Promise<CallToolResult> {
   // 1. Validate args

package/src/serve/AGENTS.md

@@ -52,7 +52,7 @@ interface ServerContext {
   embedPort: EmbeddingPort | null;
   genPort: GenerationPort | null;
   rerankPort: RerankPort | null;
-  capabilities: { bm25, vector, hybrid, answer };
+  capabilities: { bm25; vector; hybrid; answer };
 }
 ```
 
@@ -113,7 +113,7 @@ Use HTML imports with `Bun.serve()`. Don't use `vite`. HTML imports fully suppor
 Server example:
 
 ```ts
-import index from "./index.html"
+import index from "./index.html";
 
 Bun.serve({
   routes: {
@@ -134,13 +134,13 @@ Bun.serve({
     },
     close: (ws) => {
       // handle close
-    }
+    },
   },
   development: {
     hmr: true,
     console: true,
-  }
-})
+  },
+});
 ```
 
 HTML files can import .tsx, .jsx or .js files directly and Bun's bundler will transpile & bundle automatically. `<link>` tags can point to stylesheets and Bun's CSS bundler will bundle.
@@ -160,7 +160,7 @@ With the following `frontend.tsx`:
 import React from "react";
 
 // import .css files directly and it works
-import './index.css';
+import "./index.css";
 
 import { createRoot } from "react-dom/client";
 

package/src/serve/CLAUDE.md

@@ -52,7 +52,7 @@ interface ServerContext {
   embedPort: EmbeddingPort | null;
   genPort: GenerationPort | null;
   rerankPort: RerankPort | null;
-  capabilities: { bm25, vector, hybrid, answer };
+  capabilities: { bm25; vector; hybrid; answer };
 }
 ```
 
@@ -113,7 +113,7 @@ Use HTML imports with `Bun.serve()`. Don't use `vite`. HTML imports fully suppor
 Server example:
 
 ```ts
-import index from "./index.html"
+import index from "./index.html";
 
 Bun.serve({
   routes: {
@@ -134,13 +134,13 @@ Bun.serve({
     },
     close: (ws) => {
       // handle close
-    }
+    },
   },
   development: {
     hmr: true,
     console: true,
-  }
-})
+  },
+});
 ```
 
 HTML files can import .tsx, .jsx or .js files directly and Bun's bundler will transpile & bundle automatically. `<link>` tags can point to stylesheets and Bun's CSS bundler will bundle.
@@ -160,7 +160,7 @@ With the following `frontend.tsx`:
 import React from "react";
 
 // import .css files directly and it works
-import './index.css';
+import "./index.css";
 
 import { createRoot } from "react-dom/client";
 

package/vendor/fts5-snowball/README.md

@@ -24,13 +24,15 @@ The Snowball stemmer supports: Arabic, Basque, Catalan, Danish, Dutch, English,
 ## Usage
 
 ```typescript
-import { Database } from 'bun:sqlite';
+import { Database } from "bun:sqlite";
 
 // Load extension
-db.loadExtension('vendor/fts5-snowball/darwin-arm64/fts5stemmer.dylib');
+db.loadExtension("vendor/fts5-snowball/darwin-arm64/fts5stemmer.dylib");
 
 // Create FTS table with snowball tokenizer
-db.exec(`CREATE VIRTUAL TABLE docs USING fts5(content, tokenize='snowball english')`);
+db.exec(
+  `CREATE VIRTUAL TABLE docs USING fts5(content, tokenize='snowball english')`
+);
 ```
 
 ## License