@gmickel/gno 0.18.0 → 0.20.0

package/README.md

@@ -32,7 +32,12 @@ GNO is a local knowledge engine that turns your documents into a searchable, con
 
 ---
 
-## What's New in v0.18
+## What's New in v0.19
+
+- **Exclusion Filters**: explicit `exclude` controls across CLI, API, Web, and MCP to hard-prune unwanted docs by title/path/body text
+- **Ask Query-Mode Parity**: Ask now supports structured `term` / `intent` / `hyde` controls in both API and Web UI
+
+### v0.18
 
 - **Intent Steering**: optional `intent` control for ambiguous queries across CLI, API, Web, and MCP query flows
 - **Rerank Controls**: `candidateLimit` lets you tune rerank cost vs. recall on slower or memory-constrained machines
@@ -162,6 +167,7 @@ gno vsearch "error handling patterns" # Semantic similarity
 gno query "database optimization"    # Full pipeline
 gno query "meeting decisions" --since "last month" --category "meeting,notes" --author "gordon"
 gno query "performance" --intent "web performance and latency"
+gno query "performance" --exclude "reviews,hiring"
 gno ask "what did we decide" --answer # AI synthesis
 ```
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gmickel/gno",
-  "version": "0.18.0",
+  "version": "0.20.0",
   "description": "Local semantic search for your documents. Index Markdown, PDF, and Office files with hybrid BM25 + vector search.",
   "keywords": [
     "embeddings",
@@ -102,7 +102,7 @@
     "markitdown-ts": "^0.0.9",
     "minimatch": "^10.1.1",
     "nanoid": "^5.1.6",
-    "node-llama-cpp": "^3.14.5",
+    "node-llama-cpp": "^3.17.1",
     "officeparser": "^6.0.4",
     "picocolors": "^1.1.1",
     "react": "^19.2.3",
@@ -146,5 +146,8 @@
   },
   "engines": {
     "bun": ">=1.3.0"
-  }
+  },
+  "trustedDependencies": [
+    "node-llama-cpp"
+  ]
 }

package/src/cli/commands/ask.ts

@@ -199,6 +199,7 @@ export async function ask(
       author: options.author,
       tagsAll: options.tagsAll,
       tagsAny: options.tagsAny,
+      exclude: options.exclude,
       noExpand: options.noExpand,
       noRerank: options.noRerank,
       candidateLimit: options.candidateLimit,
@@ -262,6 +263,8 @@ export async function ask(
         vectorsUsed: searchResult.value.meta.vectorsUsed ?? false,
         intent: searchResult.value.meta.intent,
         candidateLimit: searchResult.value.meta.candidateLimit,
+        exclude: searchResult.value.meta.exclude,
+        queryModes: searchResult.value.meta.queryModes,
         answerGenerated,
         totalResults: results.length,
         answerContext,

package/src/cli/program.ts

@@ -225,6 +225,10 @@ function wireSearchCommands(program: Command): void {
     .option("--category <values>", "require category match (comma-separated)")
     .option("--author <text>", "filter by author (case-insensitive contains)")
     .option("--intent <text>", "disambiguating context for ambiguous queries")
+    .option(
+      "--exclude <values>",
+      "exclude docs containing any term (comma-separated)"
+    )
     .option("--tags-all <tags>", "require ALL tags (comma-separated)")
     .option("--tags-any <tags>", "require ANY tag (comma-separated)")
     .option("--full", "include full content")
@@ -270,6 +274,7 @@ function wireSearchCommands(program: Command): void {
         ? parsePositiveInt("limit", cmdOpts.limit)
         : getDefaultLimit(format);
       const categories = parseCsvValues(cmdOpts.category);
+      const exclude = parseCsvValues(cmdOpts.exclude);
 
       const { search, formatSearch } = await import("./commands/search");
       const result = await search(queryText, {
@@ -282,6 +287,7 @@ function wireSearchCommands(program: Command): void {
         categories,
         author: cmdOpts.author as string | undefined,
         intent: cmdOpts.intent as string | undefined,
+        exclude,
         tagsAll,
         tagsAny,
         full: Boolean(cmdOpts.full),
@@ -332,6 +338,10 @@ function wireSearchCommands(program: Command): void {
     .option("--category <values>", "require category match (comma-separated)")
     .option("--author <text>", "filter by author (case-insensitive contains)")
     .option("--intent <text>", "disambiguating context for ambiguous queries")
+    .option(
+      "--exclude <values>",
+      "exclude docs containing any term (comma-separated)"
+    )
     .option("--tags-all <tags>", "require ALL tags (comma-separated)")
     .option("--tags-any <tags>", "require ANY tag (comma-separated)")
     .option("--full", "include full content")
@@ -377,6 +387,7 @@ function wireSearchCommands(program: Command): void {
         ? parsePositiveInt("limit", cmdOpts.limit)
         : getDefaultLimit(format);
       const categories = parseCsvValues(cmdOpts.category);
+      const exclude = parseCsvValues(cmdOpts.exclude);
 
       const { vsearch, formatVsearch } = await import("./commands/vsearch");
       const result = await vsearch(queryText, {
@@ -389,6 +400,7 @@ function wireSearchCommands(program: Command): void {
         categories,
         author: cmdOpts.author as string | undefined,
         intent: cmdOpts.intent as string | undefined,
+        exclude,
         tagsAll,
         tagsAny,
         full: Boolean(cmdOpts.full),
@@ -434,6 +446,10 @@ function wireSearchCommands(program: Command): void {
     .option("--category <values>", "require category match (comma-separated)")
     .option("--author <text>", "filter by author (case-insensitive contains)")
     .option("--intent <text>", "disambiguating context for ambiguous queries")
+    .option(
+      "--exclude <values>",
+      "exclude docs containing any term (comma-separated)"
+    )
     .option("--tags-all <tags>", "require ALL tags (comma-separated)")
     .option("--tags-any <tags>", "require ANY tag (comma-separated)")
     .option("--full", "include full content")
@@ -505,6 +521,7 @@ function wireSearchCommands(program: Command): void {
         ? parsePositiveInt("candidate-limit", cmdOpts.candidateLimit)
         : undefined;
       const categories = parseCsvValues(cmdOpts.category);
+      const exclude = parseCsvValues(cmdOpts.exclude);
 
       // Determine expansion/rerank settings based on flags
       // Priority: --fast > --thorough > --no-expand/--no-rerank > default
@@ -541,6 +558,7 @@ function wireSearchCommands(program: Command): void {
         categories,
         author: cmdOpts.author as string | undefined,
         intent: cmdOpts.intent as string | undefined,
+        exclude,
         tagsAll,
         tagsAny,
         full: Boolean(cmdOpts.full),
@@ -586,6 +604,10 @@ function wireSearchCommands(program: Command): void {
     .option("--category <values>", "require category match (comma-separated)")
     .option("--author <text>", "filter by author (case-insensitive contains)")
     .option("--intent <text>", "disambiguating context for ambiguous queries")
+    .option(
+      "--exclude <values>",
+      "exclude docs containing any term (comma-separated)"
+    )
     .option("--fast", "skip expansion and reranking (fastest)")
     .option("--thorough", "enable query expansion (slower)")
     .option("-C, --candidate-limit <num>", "max candidates passed to reranking")
@@ -616,6 +638,7 @@ function wireSearchCommands(program: Command): void {
         ? parsePositiveInt("max-answer-tokens", cmdOpts.maxAnswerTokens)
         : undefined;
       const categories = parseCsvValues(cmdOpts.category);
+      const exclude = parseCsvValues(cmdOpts.exclude);
 
       // Determine expansion/rerank settings based on flags
       // Default: skip expansion (balanced mode)
@@ -641,6 +664,7 @@ function wireSearchCommands(program: Command): void {
         categories,
         author: cmdOpts.author as string | undefined,
         intent: cmdOpts.intent as string | undefined,
+        exclude,
         noExpand,
         noRerank,
         candidateLimit,

package/src/llm/nodeLlamaCpp/lifecycle.ts

@@ -49,7 +49,10 @@ export class ModelManager {
     if (!this.llama) {
       const { getLlama, LlamaLogLevel } = await import("node-llama-cpp");
       // Suppress model loading warnings (vocab tokens, pooling type)
-      this.llama = await getLlama({ logLevel: LlamaLogLevel.error });
+      this.llama = await getLlama({
+        build: "autoAttempt",
+        logLevel: LlamaLogLevel.error,
+      });
     }
     return this.llama;
   }

package/src/mcp/tools/index.ts

@@ -57,6 +57,7 @@ const searchInputSchema = z.object({
   minScore: z.number().min(0).max(1).optional(),
   lang: z.string().optional(),
   intent: z.string().optional(),
+  exclude: z.array(z.string()).optional(),
   since: z.string().optional(),
   until: z.string().optional(),
   categories: z.array(z.string()).optional(),
@@ -107,6 +108,7 @@ const vsearchInputSchema = z.object({
   minScore: z.number().min(0).max(1).optional(),
   lang: z.string().optional(),
   intent: z.string().optional(),
+  exclude: z.array(z.string()).optional(),
   since: z.string().optional(),
   until: z.string().optional(),
   categories: z.array(z.string()).optional(),
@@ -128,6 +130,7 @@ export const queryInputSchema = z.object({
   lang: z.string().optional(),
   intent: z.string().optional(),
   candidateLimit: z.number().int().min(1).max(100).optional(),
+  exclude: z.array(z.string()).optional(),
   since: z.string().optional(),
   until: z.string().optional(),
   categories: z.array(z.string()).optional(),

package/src/mcp/tools/query.ts

@@ -38,6 +38,7 @@ interface QueryInput {
   lang?: string;
   intent?: string;
   candidateLimit?: number;
+  exclude?: string[];
   since?: string;
   until?: string;
   categories?: string[];
@@ -251,6 +252,7 @@ export function handleQuery(
           queryLanguageHint: args.lang, // Affects expansion prompt, not retrieval
           intent: args.intent,
           candidateLimit: args.candidateLimit,
+          exclude: args.exclude,
           since: args.since,
           until: args.until,
           categories: args.categories,

package/src/mcp/tools/search.ts

@@ -20,6 +20,7 @@ interface SearchInput {
   minScore?: number;
   lang?: string;
   intent?: string;
+  exclude?: string[];
   since?: string;
   until?: string;
   categories?: string[];
@@ -110,6 +111,7 @@ export function handleSearch(
         collection: args.collection,
         lang: args.lang,
         intent: args.intent,
+        exclude: args.exclude,
         since: args.since,
         until: args.until,
         categories: args.categories,

package/src/mcp/tools/vsearch.ts

@@ -29,6 +29,7 @@ interface VsearchInput {
   minScore?: number;
   lang?: string;
   intent?: string;
+  exclude?: string[];
   since?: string;
   until?: string;
   categories?: string[];
@@ -194,6 +195,7 @@ export function handleVsearch(
             minScore: args.minScore,
             collection: args.collection,
             intent: args.intent,
+            exclude: args.exclude,
             since: args.since,
             until: args.until,
             categories: args.categories,

package/src/pipeline/exclude.ts

@@ -0,0 +1,69 @@
+/**
+ * Explicit exclusion helpers for retrieval filters.
+ *
+ * @module src/pipeline/exclude
+ */
+
+import type { ChunkRow } from "../store/types";
+
+export function normalizeExcludeTerms(values: string[]): string[] {
+  const out: string[] = [];
+  const seen = new Set<string>();
+
+  for (const value of values) {
+    for (const part of value.split(",")) {
+      const trimmed = part.trim();
+      if (!trimmed) {
+        continue;
+      }
+      const key = trimmed.toLowerCase();
+      if (seen.has(key)) {
+        continue;
+      }
+      seen.add(key);
+      out.push(trimmed);
+    }
+  }
+
+  return out;
+}
+
+function includesTerm(haystack: string, term: string): boolean {
+  return haystack.toLowerCase().includes(term.toLowerCase());
+}
+
+export function matchesExcludedText(
+  haystacks: string[],
+  excludeTerms: string[] | undefined
+): boolean {
+  if (!excludeTerms?.length) {
+    return false;
+  }
+
+  for (const haystack of haystacks) {
+    if (!haystack) {
+      continue;
+    }
+    for (const term of excludeTerms) {
+      if (includesTerm(haystack, term)) {
+        return true;
+      }
+    }
+  }
+
+  return false;
+}
+
+export function matchesExcludedChunks(
+  chunks: ChunkRow[],
+  excludeTerms: string[] | undefined
+): boolean {
+  if (!excludeTerms?.length || chunks.length === 0) {
+    return false;
+  }
+
+  return matchesExcludedText(
+    chunks.map((chunk) => chunk.text),
+    excludeTerms
+  );
+}

package/src/pipeline/hybrid.ts

@@ -21,6 +21,7 @@ import type {
 import { err, ok } from "../store/types";
 import { createChunkLookup } from "./chunk-lookup";
 import { formatQueryForEmbedding } from "./contextual";
+import { matchesExcludedChunks, matchesExcludedText } from "./exclude";
 import { expandQuery } from "./expansion";
 import {
   buildExplainResults,
@@ -670,6 +671,25 @@ export async function searchHybrid(
       continue;
     }
 
+    const excluded =
+      matchesExcludedText(
+        [
+          doc.title ?? "",
+          doc.relPath,
+          doc.author ?? "",
+          doc.contentType ?? "",
+          ...(doc.categories ?? []),
+        ],
+        options.exclude
+      ) ||
+      matchesExcludedChunks(
+        chunksMap.get(candidate.mirrorHash) ?? [],
+        options.exclude
+      );
+    if (excluded) {
+      continue;
+    }
+
     // For --full mode, de-dupe by docid (keep best scoring candidate per doc)
     if (options.full && seenDocids.has(doc.docid)) {
       continue;
@@ -810,6 +830,7 @@ export async function searchHybrid(
       vectorsUsed: vectorAvailable,
       totalResults: finalResults.length,
       intent: options.intent,
+      exclude: options.exclude,
       collection: options.collection,
       lang: options.lang,
       since: temporalRange.since,

package/src/pipeline/search.ts

@@ -17,6 +17,7 @@ import type {
 
 import { err, ok } from "../store/types";
 import { createChunkLookup } from "./chunk-lookup";
+import { matchesExcludedChunks, matchesExcludedText } from "./exclude";
 import { selectBestChunkForSteering } from "./intent";
 import { detectQueryLanguage } from "./query-language";
 import {
@@ -237,6 +238,21 @@ export async function searchBm25(
           ) ?? rawChunk)
         : rawChunk;
 
+    const excluded =
+      matchesExcludedText(
+        [fts.title ?? "", fts.relPath ?? "", fts.snippet ?? ""],
+        options.exclude
+      ) ||
+      matchesExcludedChunks(
+        chunksMapResult.ok && fts.mirrorHash
+          ? (chunksMapResult.value.get(fts.mirrorHash) ?? [])
+          : [],
+        options.exclude
+      );
+    if (excluded) {
+      continue;
+    }
+
     // For --full, de-dupe by docid (keep best scoring chunk per doc)
     // Raw BM25: smaller (more negative) is better
     if (options.full) {
@@ -309,6 +325,7 @@ export async function searchBm25(
       mode: "bm25",
       totalResults: Math.min(filteredResults.length, limit),
       intent: options.intent,
+      exclude: options.exclude,
       collection: options.collection,
       lang: options.lang,
       since: temporalRange.since,

package/src/pipeline/types.ts

@@ -79,6 +79,8 @@ export interface SearchMeta {
   author?: string;
   /** Rerank candidate limit used */
   candidateLimit?: number;
+  /** Explicit exclusion terms applied */
+  exclude?: string[];
   /** Explain data (when --explain is used) */
   explain?: {
     lines: ExplainLine[];
@@ -124,6 +126,8 @@ export interface SearchOptions {
   author?: string;
   /** Optional disambiguating context that steers scoring/snippets, but is not searched directly */
   intent?: string;
+  /** Explicit exclusion terms for hard candidate pruning */
+  exclude?: string[];
 }
 
 /** Structured query mode identifier */
@@ -317,6 +321,8 @@ export interface AskMeta {
   vectorsUsed: boolean;
   intent?: string;
   candidateLimit?: number;
+  exclude?: string[];
+  queryModes?: QueryModeSummary;
   answerGenerated?: boolean;
   totalResults?: number;
   answerContext?: AnswerContextExplain;

package/src/pipeline/vsearch.ts

@@ -14,6 +14,7 @@ import type { SearchOptions, SearchResult, SearchResults } from "./types";
 import { err, ok } from "../store/types";
 import { createChunkLookup } from "./chunk-lookup";
 import { formatQueryForEmbedding } from "./contextual";
+import { matchesExcludedChunks, matchesExcludedText } from "./exclude";
 import { selectBestChunkForSteering } from "./intent";
 import { detectQueryLanguage } from "./query-language";
 import {
@@ -174,6 +175,25 @@ export async function searchVectorWithEmbedding(
       continue;
     }
 
+    const excluded =
+      matchesExcludedText(
+        [
+          doc.title ?? "",
+          doc.relPath,
+          doc.author ?? "",
+          doc.contentType ?? "",
+          ...(doc.categories ?? []),
+        ],
+        options.exclude
+      ) ||
+      matchesExcludedChunks(
+        chunksMap.get(vec.mirrorHash) ?? [],
+        options.exclude
+      );
+    if (excluded) {
+      continue;
+    }
+
     // For --full, de-dupe by docid (keep best scoring chunk per doc)
     if (options.full) {
       const existing = bestByDocid.get(doc.docid);
@@ -301,6 +321,7 @@ export async function searchVectorWithEmbedding(
       vectorsUsed: true,
       totalResults: finalResults.length,
       intent: options.intent,
+      exclude: options.exclude,
       collection: options.collection,
       lang: options.lang,
       since: temporalRange.since,
@@ -362,6 +383,9 @@ interface DocumentInfo {
   title: string | null;
   collection: string;
   relPath: string;
+  author: string | null;
+  contentType: string | null;
+  categories: string[] | null;
   sourceHash: string;
   sourceMime: string;
   sourceExt: string;
@@ -491,6 +515,9 @@ async function buildDocumentMap(
       title: doc.title,
       collection: doc.collection,
       relPath: doc.relPath,
+      author: doc.author ?? null,
+      contentType: doc.contentType ?? null,
+      categories: doc.categories ?? null,
       sourceHash: doc.sourceHash,
       sourceMime: doc.sourceMime,
       sourceExt: doc.sourceExt,