grepmax 0.5.1 → 0.5.3

package/dist/commands/mcp.js

@@ -194,6 +194,23 @@ const TOOLS = [
             properties: {},
         },
     },
+    {
+        name: "summarize_directory",
+        description: "Generate LLM summaries for indexed code in a directory. Run after indexing. Summaries are stored and returned in search results. Requires the summarizer server on port 8101.",
+        inputSchema: {
+            type: "object",
+            properties: {
+                path: {
+                    type: "string",
+                    description: "Directory to summarize (absolute or relative). Defaults to current project root.",
+                },
+                limit: {
+                    type: "number",
+                    description: "Max chunks to summarize per call (default 200, max 5000). Run again to continue.",
+                },
+            },
+        },
+    },
 ];
 // ---------------------------------------------------------------------------
 // Helpers
@@ -505,9 +522,11 @@ exports.mcp = new commander_1.Command("mcp")
                 else {
                     lines.push("Callers: none");
                 }
-                // Callees
+                // Callees (cap at 15)
                 if (graph.callees.length > 0) {
-                    lines.push(`Calls: ${graph.callees.join(", ")}`);
+                    const capped = graph.callees.slice(0, 15);
+                    const suffix = graph.callees.length > 15 ? ` (+${graph.callees.length - 15} more)` : "";
+                    lines.push(`Calls: ${capped.join(", ")}${suffix}`);
                 }
                 else {
                     lines.push("Calls: none");
@@ -574,7 +593,12 @@ exports.mcp = new commander_1.Command("mcp")
                 if (entries.length === 0) {
                     return ok("No symbols found. Run 'gmax index' to build the index.");
                 }
-                const lines = entries.map((e) => `${e.symbol}\t${e.path}:${e.line}`);
+                const lines = entries.map((e) => {
+                    const rel = e.path.startsWith(projectRoot)
+                        ? e.path.slice(projectRoot.length + 1)
+                        : e.path;
+                    return `${e.symbol}\t${rel}:${e.line}`;
+                });
                 return ok(lines.join("\n"));
             }
             catch (e) {
@@ -585,9 +609,10 @@ exports.mcp = new commander_1.Command("mcp")
     }
     function handleIndexStatus() {
         return __awaiter(this, void 0, void 0, function* () {
-            var _a, _b, _c, _d;
+            var _a, _b, _c;
             try {
                 const config = (0, index_config_1.readIndexConfig)(config_1.PATHS.configPath);
+                const globalConfig = (0, index_config_1.readGlobalConfig)();
                 const projects = (0, project_registry_1.listProjects)();
                 const db = getVectorDb();
                 const stats = yield db.getStats();
@@ -608,7 +633,7 @@ exports.mcp = new commander_1.Command("mcp")
                 }
                 const lines = [
                     `Index: ~/.gmax/lancedb (${stats.chunks} chunks, ${fileCount} files)`,
-                    `Model: ${(_b = config === null || config === void 0 ? void 0 : config.embedModel) !== null && _b !== void 0 ? _b : "unknown"} (${(_c = config === null || config === void 0 ? void 0 : config.vectorDim) !== null && _c !== void 0 ? _c : "?"}d, ${(_d = config === null || config === void 0 ? void 0 : config.embedMode) !== null && _d !== void 0 ? _d : "unknown"})`,
+                    `Model: ${(_b = config === null || config === void 0 ? void 0 : config.embedModel) !== null && _b !== void 0 ? _b : "unknown"} (${(_c = config === null || config === void 0 ? void 0 : config.vectorDim) !== null && _c !== void 0 ? _c : "?"}d, ${globalConfig.embedMode})`,
                     (config === null || config === void 0 ? void 0 : config.indexedAt)
                         ? `Last indexed: ${config.indexedAt}`
                         : "",
@@ -625,6 +650,32 @@ exports.mcp = new commander_1.Command("mcp")
             }
         });
     }
+    function handleSummarizeDirectory(args) {
+        return __awaiter(this, void 0, void 0, function* () {
+            const dir = typeof args.path === "string"
+                ? path.resolve(args.path)
+                : projectRoot;
+            const prefix = dir.endsWith("/") ? dir : `${dir}/`;
+            const limit = Math.min(Math.max(Number(args.limit) || 200, 1), 5000);
+            try {
+                const db = getVectorDb();
+                const { summarized, remaining } = yield (0, syncer_1.generateSummaries)(db, prefix, (done, total) => {
+                    console.log(`[summarize] ${done}/${total} chunks`);
+                }, limit);
+                if (summarized === 0) {
+                    return ok("No chunks to summarize (all have summaries or summarizer unavailable)");
+                }
+                const remainMsg = remaining > 0
+                    ? ` (${remaining}+ remaining — run again to continue)`
+                    : "";
+                return ok(`Summarized ${summarized} chunks in ${path.basename(dir)}/${remainMsg}`);
+            }
+            catch (e) {
+                const msg = e instanceof Error ? e.message : String(e);
+                return err(`Summarization failed: ${msg}`);
+            }
+        });
+    }
     // --- MCP server setup ---
     const transport = new stdio_js_1.StdioServerTransport();
     const server = new index_js_1.Server({
@@ -656,6 +707,8 @@ exports.mcp = new commander_1.Command("mcp")
                 return handleListSymbols(toolArgs);
             case "index_status":
                 return handleIndexStatus();
+            case "summarize_directory":
+                return handleSummarizeDirectory(toolArgs);
             default:
                 return err(`Unknown tool: ${name}`);
         }

package/dist/commands/summarize.js

@@ -61,11 +61,12 @@ exports.summarize = new commander_1.Command("summarize")
         : "";
     const { spinner } = (0, sync_helpers_1.createIndexingSpinner)("", "Summarizing...");
     try {
-        const count = yield (0, syncer_1.generateSummaries)(vectorDb, rootPrefix, (done, total) => {
+        const { summarized, remaining } = yield (0, syncer_1.generateSummaries)(vectorDb, rootPrefix, (done, total) => {
             spinner.text = `Summarizing... (${done}/${total})`;
         });
-        if (count > 0) {
-            spinner.succeed(`Summarized ${count} chunks`);
+        if (summarized > 0) {
+            const remainMsg = remaining > 0 ? ` (${remaining}+ remaining — run again)` : "";
+            spinner.succeed(`Summarized ${summarized} chunks${remainMsg}`);
         }
         else {
             spinner.succeed("All chunks already have summaries (or summarizer unavailable)");

package/dist/lib/index/syncer.js

@@ -64,7 +64,7 @@ const project_root_1 = require("../utils/project-root");
 const pool_1 = require("../workers/pool");
 const index_config_1 = require("./index-config");
 const walker_1 = require("./walker");
-function generateSummaries(db, pathPrefix, onProgress) {
+function generateSummaries(db, pathPrefix, onProgress, maxChunks) {
     return __awaiter(this, void 0, void 0, function* () {
         let summarizeChunks;
         try {
@@ -72,23 +72,24 @@ function generateSummaries(db, pathPrefix, onProgress) {
             summarizeChunks = mod.summarizeChunks;
         }
         catch (_a) {
-            return 0;
+            return { summarized: 0, remaining: 0 };
         }
         // Quick availability check
         const test = yield summarizeChunks([
             { code: "test", language: "ts", file: "test" },
         ]);
         if (!test)
-            return 0;
+            return { summarized: 0, remaining: 0 };
+        const queryLimit = maxChunks !== null && maxChunks !== void 0 ? maxChunks : 50000;
         const table = yield db.ensureTable();
         const rows = yield table
             .query()
             .select(["id", "path", "content", "defined_symbols"])
             .where(`path LIKE '${pathPrefix}%' AND (summary IS NULL OR summary = '')`)
-            .limit(50000)
+            .limit(queryLimit)
             .toArray();
         if (rows.length === 0)
-            return 0;
+            return { summarized: 0, remaining: 0 };
         let summarized = 0;
         const BATCH_SIZE = 5;
         for (let i = 0; i < rows.length; i += BATCH_SIZE) {
@@ -124,7 +125,11 @@ function generateSummaries(db, pathPrefix, onProgress) {
             }
             onProgress === null || onProgress === void 0 ? void 0 : onProgress(summarized, rows.length);
         }
-        return summarized;
+        // Estimate remaining (rows.length was capped by queryLimit)
+        const remaining = rows.length === queryLimit
+            ? queryLimit - summarized // at least this many more
+            : 0;
+        return { summarized, remaining };
     });
 }
 function flushBatch(db, meta, vectors, pendingMeta, pendingDeletes, dryRun) {
@@ -472,26 +477,6 @@ function initialSync(options) {
                     metaCache.delete(p);
                 });
             }
-            // --- Summary post-processing (sequential, single process) ---
-            if (!dryRun && indexed > 0) {
-                const sumTimer = (0, logger_1.timer)("index", "Summarize");
-                onProgress === null || onProgress === void 0 ? void 0 : onProgress({
-                    processed,
-                    indexed,
-                    total,
-                    filePath: "Generating summaries...",
-                });
-                const summarized = yield generateSummaries(vectorDb, rootPrefix, (count, chunkTotal) => {
-                    onProgress === null || onProgress === void 0 ? void 0 : onProgress({
-                        processed: count,
-                        indexed,
-                        total: chunkTotal,
-                        filePath: `Summarizing... (${count}/${chunkTotal})`,
-                    });
-                });
-                sumTimer();
-                (0, logger_1.log)("index", `Summarize: ${summarized} chunks`);
-            }
             syncTimer();
             // Write model config so future runs can detect model changes
             if (!dryRun) {

package/mlx-embed-server/server.py

@@ -96,9 +96,18 @@ def load_model():
     print("[mlx-embed] Model ready on Metal GPU.")
 
 
+async def idle_watchdog():
+    while True:
+        await asyncio.sleep(60)
+        if time.time() - last_activity > IDLE_TIMEOUT_S:
+            print("[mlx-embed] Idle timeout, shutting down")
+            os._exit(0)
+
+
 @asynccontextmanager
 async def lifespan(app: FastAPI):
     load_model()
+    asyncio.create_task(idle_watchdog())
     yield
 
 

package/mlx-embed-server/summarizer.py

@@ -11,6 +11,7 @@ endpoints run on the event loop thread, avoiding Metal thread-safety crashes.
 import asyncio
 import logging
 import os
+import re
 import signal
 import socket
 import time
@@ -38,7 +39,7 @@ MODEL_ID = os.environ.get(
 )
 PORT = int(os.environ.get("MLX_SUMMARY_PORT", "8101"))
 IDLE_TIMEOUT_S = int(os.environ.get("MLX_SUMMARY_IDLE_TIMEOUT", "1800"))  # 30 min
-MAX_TOKENS = 100  # summaries should be one line
+MAX_TOKENS = 40  # summaries are ~20 tokens, one line
 
 model = None
 tokenizer = None
@@ -48,7 +49,7 @@ _mlx_lock = asyncio.Lock()
 
 SYSTEM_PROMPT = """You are a code summarizer. Given a code chunk, produce exactly one line describing what it does.
 Be specific about business logic, services, and side effects. Do not describe syntax.
-Do not use phrases like "This function" or "This code". Start with a verb."""
+Do not use phrases like "This function" or "This code". Start with a verb. /no_think"""
 
 def build_prompt(code: str, language: str, file: str, symbols: list[str] | None = None) -> str:
     parts = [f"Language: {language}", f"File: {file}"]
@@ -79,8 +80,12 @@ def summarize_chunk(code: str, language: str, file: str, symbols: list[str] | No
         max_tokens=MAX_TOKENS,
         verbose=False,
     )
+    # Strip thinking tokens if present
+    text = re.sub(r"<think>.*?</think>", "", response, flags=re.DOTALL).strip()
+    if not text:
+        text = response.strip()
     # Take first line only, strip whitespace
-    summary = response.strip().split("\n")[0].strip()
+    summary = text.split("\n")[0].strip()
     # Remove common prefixes the model might add
     for prefix in ["Summary: ", "summary: ", "- "]:
         if summary.startswith(prefix):
@@ -97,9 +102,18 @@ def load_model():
     print("[summarizer] Model ready on Metal GPU.")
 
 
+async def idle_watchdog():
+    while True:
+        await asyncio.sleep(60)
+        if time.time() - last_activity > IDLE_TIMEOUT_S:
+            print("[summarizer] Idle timeout, shutting down")
+            os._exit(0)
+
+
 @asynccontextmanager
 async def lifespan(app: FastAPI):
     load_model()
+    asyncio.create_task(idle_watchdog())
     yield
 
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "grepmax",
-  "version": "0.5.1",
+  "version": "0.5.3",
   "author": "Robert Owens <robowens@me.com>",
   "homepage": "https://github.com/reowens/grepmax",
   "bugs": {

package/plugins/grepmax/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "grepmax",
-  "version": "0.5.1",
+  "version": "0.5.3",
   "description": "Semantic code search for Claude Code. Automatically indexes your project and provides intelligent search capabilities.",
   "author": {
     "name": "Robert Owens",

package/plugins/grepmax/skills/gmax/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: gmax
 description: Semantic code search. Use alongside grep - grep for exact strings, gmax for concepts.
-allowed-tools: "mcp__grepmax__semantic_search, mcp__grepmax__search_all, mcp__grepmax__code_skeleton, mcp__grepmax__trace_calls, mcp__grepmax__list_symbols, mcp__grepmax__index_status, Bash(gmax:*), Read"
+allowed-tools: "mcp__grepmax__semantic_search, mcp__grepmax__search_all, mcp__grepmax__code_skeleton, mcp__grepmax__trace_calls, mcp__grepmax__list_symbols, mcp__grepmax__index_status, mcp__grepmax__summarize_directory, Bash(gmax:*), Read"
 ---
 
 ## What gmax does
@@ -67,6 +67,10 @@ List indexed symbols with definition locations.
 ### index_status
 Check centralized index health — chunks, files, indexed directories, model info.
 
+### summarize_directory
+Generate LLM summaries for indexed code in a directory. Summaries are stored and returned in search results. Run after indexing a new directory.
+- `path` (optional): Directory to summarize. Defaults to project root.
+
 ## Workflow
 
 1. **Search** — `semantic_search` to find relevant code (pointers by default)