@gmickel/gno 0.39.0 → 0.40.0

package/README.md

@@ -87,7 +87,7 @@ gno daemon
 
 ## What's New
 
-> Latest release: [v0.37.0](./CHANGELOG.md#0370---2026-04-06)  
+> Latest release: [v0.39.1](./CHANGELOG.md#0391---2026-04-06)  
 > Full release history: [CHANGELOG.md](./CHANGELOG.md)
 
 - **Retrieval Quality Upgrade**: stronger BM25 lexical handling, code-aware chunking, terminal result hyperlinks, and per-collection model overrides
@@ -95,6 +95,19 @@ gno daemon
 - **Default Embed Model**: built-in presets now use `Qwen3-Embedding-0.6B-GGUF` after it beat `bge-m3` on both code and multilingual prose benchmark lanes
 - **Regression Fixes**: tightened phrase/negation/hyphen/underscore BM25 behavior, cleaned non-TTY hyperlink output, improved `gno doctor` chunking visibility, and fixed the embedding autoresearch harness
 
+### Upgrading Existing Collections
+
+If you already had collections indexed before the default embed-model switch to
+`Qwen3-Embedding-0.6B-GGUF`, run:
+
+```bash
+gno models pull --embed
+gno embed
+```
+
+That regenerates embeddings for the new default model. Old vectors are kept
+until you explicitly clear stale embeddings.
+
 ### Fine-Tuned Model Quick Use
 
 ```yaml

package/assets/skill/SKILL.md

@@ -173,6 +173,9 @@ If you edit/create files that should be searchable via vector search:
 gno index              # Full re-index (sync + embed)
 # or
 gno embed              # Embed only (if already synced)
+gno embed travel       # Embed one collection only
+# or
+gno embed --collection travel
 ```
 
 MCP `gno.sync` and `gno.capture` do NOT auto-embed. Use CLI for embedding.
@@ -206,6 +209,10 @@ gno collection clear-embeddings gno-code        # stale models only
 gno collection clear-embeddings gno-code --all  # remove everything, then re-embed
 ```
 
+MCP-equivalent write tool:
+
+- `gno_clear_collection_embeddings`
+
 ## Reference Documentation
 
 | Topic                                                 | File                                 |

package/assets/skill/cli-reference.md

@@ -72,6 +72,12 @@ gno collection rename <old> <new>
 gno collection clear-embeddings <name> [--all] [--json]
 ```
 
+### gno embed
+
+```bash
+gno embed [collection] [--collection <name>] [--force] [--model <uri>] [--batch-size <n>] [--dry-run]
+```
+
 ## Indexing
 
 ### gno update

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gmickel/gno",
-  "version": "0.39.0",
+  "version": "0.40.0",
   "description": "Local semantic search for your documents. Index Markdown, PDF, and Office files with hybrid BM25 + vector search.",
   "keywords": [
     "embeddings",

package/src/cli/program.ts

@@ -1495,34 +1495,43 @@ function wireManagementCommands(program: Command): void {
 
   // embed - Generate embeddings
   program
-    .command("embed")
+    .command("embed [collection]")
     .description("Generate embeddings for indexed documents")
+    .option("--collection <name>", "restrict to one collection")
     .option("--model <uri>", "embedding model URI")
     .option("--batch-size <num>", "batch size", "32")
     .option("--force", "regenerate all embeddings")
     .option("--dry-run", "show what would be done")
     .option("--json", "JSON output")
-    .action(async (cmdOpts: Record<string, unknown>) => {
-      const globals = getGlobals();
-      const format = getFormat(cmdOpts);
+    .action(
+      async (
+        collectionArg: string | undefined,
+        cmdOpts: Record<string, unknown>
+      ) => {
+        const globals = getGlobals();
+        const format = getFormat(cmdOpts);
 
-      const { embed, formatEmbed } = await import("./commands/embed");
-      const opts = {
-        model: cmdOpts.model as string | undefined,
-        batchSize: parsePositiveInt("batch-size", cmdOpts.batchSize),
-        force: Boolean(cmdOpts.force),
-        dryRun: Boolean(cmdOpts.dryRun),
-        yes: globals.yes,
-        json: format === "json",
-        verbose: globals.verbose,
-      };
-      const result = await embed(opts);
+        const { embed, formatEmbed } = await import("./commands/embed");
+        const collection =
+          collectionArg ?? (cmdOpts.collection as string | undefined);
+        const opts = {
+          collection,
+          model: cmdOpts.model as string | undefined,
+          batchSize: parsePositiveInt("batch-size", cmdOpts.batchSize),
+          force: Boolean(cmdOpts.force),
+          dryRun: Boolean(cmdOpts.dryRun),
+          yes: globals.yes,
+          json: format === "json",
+          verbose: globals.verbose,
+        };
+        const result = await embed(opts);
 
-      if (!result.success) {
-        throw new CliError("RUNTIME", result.error ?? "Embed failed");
+        if (!result.success) {
+          throw new CliError("RUNTIME", result.error ?? "Embed failed");
+        }
+        process.stdout.write(`${formatEmbed(result, opts)}\n`);
       }
-      process.stdout.write(`${formatEmbed(result, opts)}\n`);
-    });
+    );
 
   // cleanup - Clean stale data
   program

package/src/embed/backlog.ts

@@ -25,6 +25,7 @@ export interface EmbedBacklogDeps {
   statsPort: VectorStatsPort;
   embedPort: EmbeddingPort;
   vectorIndex: VectorIndexPort;
+  collection?: string;
   modelUri: string;
   batchSize?: number;
 }
@@ -52,7 +53,7 @@ interface Cursor {
 export async function embedBacklog(
   deps: EmbedBacklogDeps
 ): Promise<StoreResult<EmbedBacklogResult>> {
-  const { statsPort, embedPort, vectorIndex, modelUri } = deps;
+  const { statsPort, embedPort, vectorIndex, modelUri, collection } = deps;
   const batchSize = deps.batchSize ?? 32;
 
   let embedded = 0;
@@ -65,6 +66,7 @@ export async function embedBacklog(
       const batchResult = await statsPort.getBacklog(modelUri, {
         limit: batchSize,
         after: cursor,
+        collection,
       });
 
       if (!batchResult.ok) {

package/src/mcp/tools/clear-collection-embeddings.ts

@@ -0,0 +1,96 @@
+/**
+ * MCP gno_clear_collection_embeddings tool.
+ *
+ * @module src/mcp/tools/clear-collection-embeddings
+ */
+
+import type { ToolContext } from "../server";
+
+import { MCP_ERRORS } from "../../core/errors";
+import { withWriteLock } from "../../core/file-lock";
+import { resolveModelUri } from "../../llm/registry";
+import { runTool, type ToolResult } from "./index";
+
+interface ClearCollectionEmbeddingsInput {
+  collection: string;
+  mode?: "stale" | "all";
+}
+
+interface ClearCollectionEmbeddingsResult {
+  collection: string;
+  deletedVectors: number;
+  deletedModels: string[];
+  mode: "stale" | "all";
+  protectedSharedVectors: number;
+  note?: string;
+}
+
+function formatResult(result: ClearCollectionEmbeddingsResult): string {
+  const lines = [
+    `Collection: ${result.collection}`,
+    `Mode: ${result.mode}`,
+    `Deleted vectors: ${result.deletedVectors}`,
+  ];
+  if (result.deletedModels.length > 0) {
+    lines.push(`Models: ${result.deletedModels.join(", ")}`);
+  }
+  if (result.protectedSharedVectors > 0) {
+    lines.push(`Protected shared vectors: ${result.protectedSharedVectors}`);
+  }
+  if (result.note) {
+    lines.push(result.note);
+  }
+  return lines.join("\n");
+}
+
+export function handleClearCollectionEmbeddings(
+  args: ClearCollectionEmbeddingsInput,
+  ctx: ToolContext
+): Promise<ToolResult> {
+  return runTool(
+    ctx,
+    "gno_clear_collection_embeddings",
+    async () => {
+      if (!ctx.enableWrite) {
+        throw new Error("Write tools disabled. Start MCP with --enable-write.");
+      }
+
+      const collection = ctx.collections.find(
+        (item) => item.name === args.collection
+      );
+      if (!collection) {
+        throw new Error(
+          `${MCP_ERRORS.NOT_FOUND.code}: Collection not found: ${args.collection}`
+        );
+      }
+
+      const mode = args.mode ?? "stale";
+      return withWriteLock(ctx.writeLockPath, async () => {
+        const result = await ctx.store.clearEmbeddingsForCollection(
+          collection.name,
+          {
+            mode,
+            activeModel: resolveModelUri(
+              ctx.config,
+              "embed",
+              undefined,
+              collection.name
+            ),
+          }
+        );
+        if (!result.ok) {
+          throw new Error(`${result.error.code}: ${result.error.message}`);
+        }
+
+        return {
+          ...result.value,
+          note:
+            mode === "all"
+              ? `Run gno_embed or gno_index for ${collection.name} to rebuild embeddings.`
+              : undefined,
+        };
+      });
+    },
+    formatResult
+  );
+}

package/src/mcp/tools/embed.ts

@@ -11,14 +11,16 @@ import { acquireWriteLock, type WriteLockHandle } from "../../core/file-lock";
 import { JobError } from "../../core/job-manager";
 import { embedBacklog } from "../../embed";
 import { LlmAdapter } from "../../llm/nodeLlamaCpp/adapter";
-import { getActivePreset } from "../../llm/registry";
+import { resolveModelUri } from "../../llm/registry";
 import {
   createVectorIndexPort,
   createVectorStatsPort,
 } from "../../store/vector";
 import { runTool, type ToolResult } from "./index";
 
-type EmbedInput = Record<string, never>;
+interface EmbedInput {
+  collection?: string;
+}
 
 interface EmbedResultOutput {
   jobId: string;
@@ -57,9 +59,23 @@ export function handleEmbed(
           );
         }
 
-        // Get model from active preset
-        const preset = getActivePreset(ctx.config);
-        const modelUri = preset.embed;
+        const requestedCollection = args.collection?.trim();
+        const collection = requestedCollection
+          ? ctx.collections.find((item) => item.name === requestedCollection)
+          : null;
+
+        if (requestedCollection && !collection) {
+          throw new Error(
+            `${MCP_ERRORS.NOT_FOUND.code}: Collection not found: ${requestedCollection}`
+          );
+        }
+
+        const modelUri = resolveModelUri(
+          ctx.config,
+          "embed",
+          undefined,
+          collection?.name
+        );
 
         const jobId = await ctx.jobManager.startTypedJobWithLock(
           "embed",
@@ -74,7 +90,7 @@ export function handleEmbed(
             if (!embedResult.ok) {
               throw new Error(
                 `MODEL_NOT_FOUND: Embedding model not cached. ` +
-                  `Model: ${modelUri}, Preset: ${preset.name}. ` +
+                  `Model: ${modelUri}. ` +
                   `Run 'gno models pull embed' first.`
               );
             }
@@ -108,6 +124,7 @@ export function handleEmbed(
                 statsPort,
                 embedPort,
                 vectorIndex,
+                collection: collection?.name,
                 modelUri,
                 batchSize: 32,
               });

package/src/mcp/tools/index-cmd.ts

@@ -14,7 +14,7 @@ import { normalizeCollectionName } from "../../core/validation";
 import { embedBacklog } from "../../embed";
 import { defaultSyncService } from "../../ingestion";
 import { LlmAdapter } from "../../llm/nodeLlamaCpp/adapter";
-import { getActivePreset } from "../../llm/registry";
+import { resolveModelUri } from "../../llm/registry";
 import {
   createVectorIndexPort,
   createVectorStatsPort,
@@ -101,9 +101,12 @@ export function handleIndex(
           runUpdateCmd: false,
         };
 
-        // Get model from active preset
-        const preset = getActivePreset(ctx.config);
-        const modelUri = preset.embed;
+        const modelUri = resolveModelUri(
+          ctx.config,
+          "embed",
+          undefined,
+          collection?.name
+        );
 
         const jobId = await ctx.jobManager.startTypedJobWithLock(
           "index",
@@ -137,7 +140,7 @@ export function handleIndex(
             if (!embedResult.ok) {
               throw new Error(
                 `MODEL_NOT_FOUND: Embedding model not cached. ` +
-                  `Model: ${modelUri}, Preset: ${preset.name}. ` +
+                  `Model: ${modelUri}. ` +
                   `Run 'gno models pull embed' first.`
               );
             }
@@ -171,6 +174,7 @@ export function handleIndex(
                 statsPort,
                 embedPort,
                 vectorIndex,
+                collection: collection?.name,
                 modelUri,
                 batchSize: 32,
               });

package/src/mcp/tools/index.ts

@@ -13,6 +13,7 @@ import type { ToolContext } from "../server";
 import { normalizeTag } from "../../core/tags";
 import { handleAddCollection } from "./add-collection";
 import { handleCapture } from "./capture";
+import { handleClearCollectionEmbeddings } from "./clear-collection-embeddings";
 import { handleEmbed } from "./embed";
 import { handleGet } from "./get";
 import { handleIndex } from "./index-cmd";
@@ -213,7 +214,12 @@ const syncInputSchema = z.object({
     .describe("Run the collection's configured update command before syncing"),
 });
 
-const embedInputSchema = z.object({});
+const embedInputSchema = z.object({
+  collection: z
+    .string()
+    .optional()
+    .describe("Collection name to embed. Omit to embed all collections"),
+});
 
 const indexInputSchema = z.object({
   collection: z
@@ -230,6 +236,17 @@ const removeCollectionInputSchema = z.object({
     .describe("Collection name to remove"),
 });
 
+const clearCollectionEmbeddingsInputSchema = z.object({
+  collection: z
+    .string()
+    .min(1, "Collection cannot be empty")
+    .describe("Collection name to clean"),
+  mode: z
+    .enum(["stale", "all"])
+    .default("stale")
+    .describe("Cleanup mode: stale models only, or all embeddings"),
+});
+
 const createFolderInputSchema = z.object({
   collection: z.string().min(1, "Collection cannot be empty"),
   name: z.string().min(1, "Folder name cannot be empty"),
@@ -799,7 +816,7 @@ export function registerTools(server: McpServer, ctx: ToolContext): void {
 
     server.tool(
       "gno_embed",
-      "Generate vector embeddings for all unembedded chunks. Async: returns a job ID. Poll with gno_job_status.",
+      "Generate vector embeddings for all unembedded chunks, optionally scoped to one collection. Async: returns a job ID. Poll with gno_job_status.",
       embedInputSchema.shape,
       (args) => handleEmbed(args, ctx)
     );
@@ -818,6 +835,13 @@ export function registerTools(server: McpServer, ctx: ToolContext): void {
       (args) => handleRemoveCollection(args, ctx)
     );
 
+    server.tool(
+      "gno_clear_collection_embeddings",
+      "Remove stale or all embeddings for one collection.",
+      clearCollectionEmbeddingsInputSchema.shape,
+      (args) => handleClearCollectionEmbeddings(args, ctx)
+    );
+
     server.tool(
       "gno_create_folder",
       "Create a folder inside an existing collection.",

package/src/sdk/embed.ts

@@ -202,7 +202,7 @@ export async function runEmbed(
 
   const backlogResult = force
     ? await getActiveChunkCount(db)
-    : await stats.countBacklog(modelUri);
+    : await stats.countBacklog(modelUri, { collection: options.collection });
   if (!backlogResult.ok) {
     throw sdkError("STORE", backlogResult.error.message, {
       cause: backlogResult.error.cause,
@@ -264,6 +264,7 @@ export async function runEmbed(
         statsPort: stats,
         embedPort,
         vectorIndex,
+        collection: options.collection,
         modelUri,
         batchSize,
       });