brain-cache 0.3.2 → 0.3.4

package/dist/{askCodebase-GXUB6477.js → askCodebase-WOFIJMSU.js}

@@ -1,16 +1,16 @@
 #!/usr/bin/env node
-import {
-  formatTokenSavings
-} from "./chunk-GGOUKACO.js";
 import {
   runBuildContext
-} from "./chunk-H4XSGUHL.js";
+} from "./chunk-QNPB2UD2.js";
 import "./chunk-5FXXZBZV.js";
+import {
+  formatTokenSavings
+} from "./chunk-GGOUKACO.js";
 import "./chunk-BF5UDEIF.js";
 import "./chunk-GR6QXZ4J.js";
 import "./chunk-V4ARVFRG.js";
-import "./chunk-6MACVOTO.js";
-import "./chunk-MSI4MDIM.js";
+import "./chunk-DTJNPRN6.js";
+import "./chunk-DDDLTKDT.js";
 import {
   childLogger
 } from "./chunk-3SFDFUEX.js";

package/dist/{buildContext-AHUWIN6M.js → buildContext-QIMH5LQY.js}

@@ -1,13 +1,13 @@
 #!/usr/bin/env node
 import {
   runBuildContext
-} from "./chunk-H4XSGUHL.js";
+} from "./chunk-QNPB2UD2.js";
 import "./chunk-5FXXZBZV.js";
 import "./chunk-BF5UDEIF.js";
 import "./chunk-GR6QXZ4J.js";
 import "./chunk-V4ARVFRG.js";
-import "./chunk-6MACVOTO.js";
-import "./chunk-MSI4MDIM.js";
+import "./chunk-DTJNPRN6.js";
+import "./chunk-DDDLTKDT.js";
 import "./chunk-3SFDFUEX.js";
 export {
   runBuildContext

package/dist/{chunk-MSI4MDIM.js → chunk-DDDLTKDT.js}

@@ -38,7 +38,8 @@ var IndexStateSchema = z.object({
   dimension: z.number().int(),
   indexedAt: z.string().datetime(),
   fileCount: z.number().int(),
-  chunkCount: z.number().int()
+  chunkCount: z.number().int(),
+  totalTokens: z.number().int().default(0)
 });
 
 // src/services/capability.ts

package/dist/{chunk-6MACVOTO.js → chunk-DTJNPRN6.js}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import {
   IndexStateSchema
-} from "./chunk-MSI4MDIM.js";
+} from "./chunk-DDDLTKDT.js";
 import {
   DEFAULT_EMBEDDING_DIMENSION,
   EMBEDDING_DIMENSIONS,

package/dist/{chunk-H4XSGUHL.js → chunk-QNPB2UD2.js}

@@ -1,7 +1,6 @@
 #!/usr/bin/env node
 import {
-  assembleContext,
-  countChunkTokens
+  assembleContext
 } from "./chunk-5FXXZBZV.js";
 import {
   RETRIEVAL_STRATEGIES,
@@ -18,16 +17,15 @@ import {
 import {
   openDatabase,
   readIndexState
-} from "./chunk-6MACVOTO.js";
+} from "./chunk-DTJNPRN6.js";
 import {
   readProfile
-} from "./chunk-MSI4MDIM.js";
+} from "./chunk-DDDLTKDT.js";
 import {
   DEFAULT_TOKEN_BUDGET
 } from "./chunk-3SFDFUEX.js";
 
 // src/workflows/buildContext.ts
-import { readFile } from "fs/promises";
 import { resolve } from "path";
 async function runBuildContext(query, opts) {
   const profile = await readProfile();
@@ -64,15 +62,7 @@ async function runBuildContext(query, opts) {
   const results = await searchChunks(table, queryVector, strategy);
   const deduped = deduplicateChunks(results);
   const assembled = assembleContext(deduped, { maxTokens });
-  const uniqueFiles = [...new Set(assembled.chunks.map((c) => c.filePath))];
-  let estimatedWithoutBraincache = 0;
-  for (const filePath of uniqueFiles) {
-    try {
-      const fileContent = await readFile(filePath, "utf-8");
-      estimatedWithoutBraincache += countChunkTokens(fileContent);
-    } catch {
-    }
-  }
+  const estimatedWithoutBraincache = indexState.totalTokens;
   const reductionPct = estimatedWithoutBraincache > 0 ? Math.round((1 - assembled.tokenCount / estimatedWithoutBraincache) * 100) : 0;
   const result = {
     content: assembled.content,

package/dist/claude-md-section-YPMDPAPW.js

@@ -0,0 +1,34 @@
+#!/usr/bin/env node
+
+// src/lib/claude-md-section.ts
+var CLAUDE_MD_SECTION = `
+## Brain-Cache MCP Tools
+
+**IMPORTANT: Use brain-cache tools as your FIRST action before reading files or using Grep/Glob.**
+
+### build_context (use for understanding)
+
+Call \`mcp__brain-cache__build_context\` with the user's question BEFORE reading individual files. This retrieves semantically relevant code from across the entire repo, deduplicates it, and returns a token-budgeted context block.
+
+**Always use for questions like:**
+- "How does X work?" / "Explain X end to end"
+- "Walk me through the flow of X"
+- "What does this page/feature/component do?"
+- "Explain the architecture" / "How is the project structured?"
+- "What happens when Y is called?"
+- Any question that requires understanding code across multiple files
+- Debugging unfamiliar code paths or understanding error flows
+
+**Do NOT skip this tool and jump to reading files.** build_context returns better results with fewer tokens than manually reading files.
+
+### search_codebase (use for locating)
+
+Call \`mcp__brain-cache__search_codebase\` instead of Grep or Glob when locating functions, symbols, types, definitions, or implementations. It uses semantic search \u2014 finds code by meaning, not just keyword match.
+
+### doctor (use for diagnosing)
+
+Call \`mcp__brain-cache__doctor\` first when any brain-cache tool fails or returns unexpected results.
+`;
+export {
+  CLAUDE_MD_SECTION
+};

package/dist/cli.js

@@ -5,23 +5,23 @@ import {
 
 // src/cli/index.ts
 import { Command } from "commander";
-var version = "0.3.2";
+var version = "0.3.4";
 var program = new Command();
 program.name("brain-cache").description("Local AI runtime \u2014 GPU cache layer for Claude").version(version);
 program.command("init").description("Detect hardware, pull embedding model, create config directory").action(async () => {
-  const { runInit } = await import("./init-SXC4MWOR.js");
+  const { runInit } = await import("./init-L4CMLYVY.js");
   await runInit();
 });
 program.command("doctor").description("Report system health: GPU, VRAM tier, Ollama status").action(async () => {
-  const { runDoctor } = await import("./doctor-3RIVSSNB.js");
+  const { runDoctor } = await import("./doctor-DRVKTWPW.js");
   await runDoctor();
 });
 program.command("index").description("Index a codebase: parse, chunk, embed, and store in LanceDB").argument("[path]", "Directory to index (defaults to current directory)").option("-f, --force", "Force full reindex, ignoring cached file hashes").action(async (path, opts) => {
-  const { runIndex } = await import("./workflows-TWA2GDHJ.js");
+  const { runIndex } = await import("./workflows-ZC24RRNX.js");
   await runIndex(path, { force: opts.force });
 });
 program.command("search").description("Search indexed codebase with a natural language query").argument("<query>", "Natural language query string").option("-n, --limit <n>", "Maximum number of results", "10").option("-p, --path <path>", "Project root directory").action(async (query, opts) => {
-  const { runSearch } = await import("./search-ZP4NKKAZ.js");
+  const { runSearch } = await import("./search-RGK2FEM5.js");
   await runSearch(query, {
     limit: parseInt(opts.limit, 10),
     path: opts.path
@@ -30,12 +30,12 @@ program.command("search").description("Search indexed codebase with a natural la
 program.command("status").description(
   "Show index stats: files indexed, chunks stored, last indexed time"
 ).argument("[path]", "Project root directory (defaults to current directory)").action(async (path) => {
-  const { runStatus } = await import("./status-JYNMLSXZ.js");
+  const { runStatus } = await import("./status-4IWQTGPA.js");
   await runStatus(path);
 });
 program.command("context").description("Build token-budgeted context from codebase for a query").argument("<query>", "Natural language query string").option("-n, --limit <n>", "Maximum number of search results", "10").option("-b, --budget <tokens>", "Token budget for assembled context", "4096").option("-p, --path <path>", "Project root directory").option("--raw", "Output raw JSON (MCP transport compatible)").action(
   async (query, opts) => {
-    const { runBuildContext } = await import("./buildContext-AHUWIN6M.js");
+    const { runBuildContext } = await import("./buildContext-QIMH5LQY.js");
     const result = await runBuildContext(query, {
       limit: parseInt(opts.limit, 10),
       maxTokens: parseInt(opts.budget, 10),
@@ -61,7 +61,7 @@ ${formatTokenSavings({ tokensSent: result.metadata.tokensSent, estimatedWithout,
 program.command("ask").description(
   "Ask a natural language question about the codebase \u2014 retrieves context locally, reasons via Claude"
 ).argument("<question>", "Natural language question about the codebase").option("-b, --budget <tokens>", "Token budget for context retrieval", "4096").option("-p, --path <path>", "Project root directory").action(async (question, opts) => {
-  const { runAskCodebase } = await import("./askCodebase-GXUB6477.js");
+  const { runAskCodebase } = await import("./askCodebase-WOFIJMSU.js");
   const result = await runAskCodebase(question, {
     path: opts.path,
     maxContextTokens: parseInt(opts.budget, 10)

package/dist/{doctor-3RIVSSNB.js → doctor-DRVKTWPW.js}

@@ -8,7 +8,7 @@ import {
 import {
   detectCapabilities,
   readProfile
-} from "./chunk-MSI4MDIM.js";
+} from "./chunk-DDDLTKDT.js";
 import {
   PROFILE_PATH
 } from "./chunk-3SFDFUEX.js";

package/dist/{init-SXC4MWOR.js → init-L4CMLYVY.js}

@@ -9,7 +9,7 @@ import {
 import {
   detectCapabilities,
   writeProfile
-} from "./chunk-MSI4MDIM.js";
+} from "./chunk-DDDLTKDT.js";
 import "./chunk-3SFDFUEX.js";
 
 // src/workflows/init.ts
@@ -86,13 +86,7 @@ async function runInit() {
     process.stderr.write("brain-cache: created .mcp.json with brain-cache MCP server.\n");
   }
   const claudeMdPath = "CLAUDE.md";
-  const brainCacheSection = `
-## Brain-Cache MCP Tools
-
-When answering "how does X work", "explain the architecture", or any question requiring multi-file reasoning, call \`mcp__brain-cache__build_context\` with the question BEFORE reading individual files. It returns pre-built, relevant context with token savings.
-
-When locating functions, symbols, types, or definitions, call \`mcp__brain-cache__search_codebase\` instead of using Grep or file-find tools. It searches the local vector index and returns ranked results.
-`;
+  const { CLAUDE_MD_SECTION: brainCacheSection } = await import("./claude-md-section-YPMDPAPW.js");
   if (existsSync(claudeMdPath)) {
     const content = readFileSync(claudeMdPath, "utf-8");
     if (content.includes("## Brain-Cache MCP Tools")) {

package/dist/mcp.js

@@ -91,7 +91,8 @@ var IndexStateSchema = z.object({
   dimension: z.number().int(),
   indexedAt: z.string().datetime(),
   fileCount: z.number().int(),
-  chunkCount: z.number().int()
+  chunkCount: z.number().int(),
+  totalTokens: z.number().int().default(0)
 });
 
 // src/lib/config.ts
@@ -796,6 +797,10 @@ async function runIndex(targetPath, opts) {
       delete updatedHashes[filePath];
     }
     const filesToProcess = [...newFiles, ...changedFiles];
+    let allFilesTotalTokens = 0;
+    for (const [, content] of contentMap) {
+      allFilesTotalTokens += countChunkTokens(content);
+    }
     if (filesToProcess.length === 0) {
       process.stderr.write(`brain-cache: nothing to re-index
 `);
@@ -811,7 +816,8 @@ async function runIndex(targetPath, opts) {
         dimension: dim,
         indexedAt: (/* @__PURE__ */ new Date()).toISOString(),
         fileCount: totalFiles2,
-        chunkCount: chunkCount2
+        chunkCount: chunkCount2,
+        totalTokens: allFilesTotalTokens
       });
       process.stderr.write(
         `brain-cache: indexing complete
@@ -902,7 +908,8 @@ async function runIndex(targetPath, opts) {
       dimension: dim,
       indexedAt: (/* @__PURE__ */ new Date()).toISOString(),
       fileCount: totalFiles,
-      chunkCount
+      chunkCount,
+      totalTokens: allFilesTotalTokens
     });
     const reductionPct = totalRawTokens > 0 ? Math.round((1 - totalChunkTokens / totalRawTokens) * 100) : 0;
     const savingsBlock = formatTokenSavings({
@@ -1080,7 +1087,6 @@ async function runSearch(query, opts) {
 }
 
 // src/workflows/buildContext.ts
-import { readFile as readFile5 } from "fs/promises";
 import { resolve as resolve3 } from "path";
 async function runBuildContext(query, opts) {
   const profile = await readProfile();
@@ -1117,15 +1123,7 @@ async function runBuildContext(query, opts) {
   const results = await searchChunks(table, queryVector, strategy);
   const deduped = deduplicateChunks(results);
   const assembled = assembleContext(deduped, { maxTokens });
-  const uniqueFiles = [...new Set(assembled.chunks.map((c) => c.filePath))];
-  let estimatedWithoutBraincache = 0;
-  for (const filePath of uniqueFiles) {
-    try {
-      const fileContent = await readFile5(filePath, "utf-8");
-      estimatedWithoutBraincache += countChunkTokens(fileContent);
-    } catch {
-    }
-  }
+  const estimatedWithoutBraincache = indexState.totalTokens;
   const reductionPct = estimatedWithoutBraincache > 0 ? Math.round((1 - assembled.tokenCount / estimatedWithoutBraincache) * 100) : 0;
   const result = {
     content: assembled.content,
@@ -1146,7 +1144,7 @@ async function runBuildContext(query, opts) {
 }
 
 // src/mcp/index.ts
-var version = "0.3.2";
+var version = "0.3.4";
 var log9 = childLogger("mcp");
 var server = new McpServer({ name: "brain-cache", version });
 server.registerTool(

package/dist/{search-ZP4NKKAZ.js → search-RGK2FEM5.js}

@@ -14,10 +14,10 @@ import {
 import {
   openDatabase,
   readIndexState
-} from "./chunk-6MACVOTO.js";
+} from "./chunk-DTJNPRN6.js";
 import {
   readProfile
-} from "./chunk-MSI4MDIM.js";
+} from "./chunk-DDDLTKDT.js";
 import "./chunk-3SFDFUEX.js";
 
 // src/workflows/search.ts

package/dist/{status-JYNMLSXZ.js → status-4IWQTGPA.js}

@@ -1,10 +1,10 @@
 #!/usr/bin/env node
 import {
   readIndexState
-} from "./chunk-6MACVOTO.js";
+} from "./chunk-DTJNPRN6.js";
 import {
   readProfile
-} from "./chunk-MSI4MDIM.js";
+} from "./chunk-DDDLTKDT.js";
 import "./chunk-3SFDFUEX.js";
 
 // src/workflows/status.ts

package/dist/{workflows-TWA2GDHJ.js → workflows-ZC24RRNX.js}

@@ -1,10 +1,10 @@
 #!/usr/bin/env node
-import {
-  formatTokenSavings
-} from "./chunk-GGOUKACO.js";
 import {
   countChunkTokens
 } from "./chunk-5FXXZBZV.js";
+import {
+  formatTokenSavings
+} from "./chunk-GGOUKACO.js";
 import {
   embedBatchWithRetry
 } from "./chunk-GR6QXZ4J.js";
@@ -20,10 +20,10 @@ import {
   readFileHashes,
   writeFileHashes,
   writeIndexState
-} from "./chunk-6MACVOTO.js";
+} from "./chunk-DTJNPRN6.js";
 import {
   readProfile
-} from "./chunk-MSI4MDIM.js";
+} from "./chunk-DDDLTKDT.js";
 import {
   DEFAULT_BATCH_SIZE,
   DEFAULT_EMBEDDING_DIMENSION,
@@ -345,6 +345,10 @@ async function runIndex(targetPath, opts) {
       delete updatedHashes[filePath];
     }
     const filesToProcess = [...newFiles, ...changedFiles];
+    let allFilesTotalTokens = 0;
+    for (const [, content] of contentMap) {
+      allFilesTotalTokens += countChunkTokens(content);
+    }
     if (filesToProcess.length === 0) {
       process.stderr.write(`brain-cache: nothing to re-index
 `);
@@ -360,7 +364,8 @@ async function runIndex(targetPath, opts) {
         dimension: dim,
         indexedAt: (/* @__PURE__ */ new Date()).toISOString(),
         fileCount: totalFiles2,
-        chunkCount: chunkCount2
+        chunkCount: chunkCount2,
+        totalTokens: allFilesTotalTokens
       });
       process.stderr.write(
         `brain-cache: indexing complete
@@ -451,7 +456,8 @@ async function runIndex(targetPath, opts) {
       dimension: dim,
       indexedAt: (/* @__PURE__ */ new Date()).toISOString(),
       fileCount: totalFiles,
-      chunkCount
+      chunkCount,
+      totalTokens: allFilesTotalTokens
     });
     const reductionPct = totalRawTokens > 0 ? Math.round((1 - totalChunkTokens / totalRawTokens) * 100) : 0;
     const savingsBlock = formatTokenSavings({

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "brain-cache",
-  "version": "0.3.2",
+  "version": "0.3.4",
   "description": "Local MCP-first context engine for Claude. Index your codebase, retrieve only what matters, and cut token usage.",
   "license": "MIT",
   "type": "module",