scai 0.1.45 → 0.1.46

package/dist/commands/AskCmd.js

@@ -8,7 +8,8 @@ import { buildContextualPrompt } from '../utils/buildContextualPrompt.js';
 import { generateFileTree } from '../utils/fileTree.js';
 import { log } from '../utils/log.js';
 import { PROMPT_LOG_PATH, SCAI_HOME, INDEX_DIR } from '../constants.js';
-const MAX_RELATED_FILES = 5;
+const MAX_RELATED_FILES = 3;
+const MAX_SUMMARY_LINES = 12;
 export async function runAskCommand(query) {
     if (!query) {
         query = await promptOnce('🧠 Ask your question:\n> ');
@@ -37,7 +38,6 @@ export async function runAskCommand(query) {
     // 🟩 STEP 2: Merge results (de-duplicate by full resolved path)
     const seen = new Set();
     const combinedResults = [];
-    // Merging results now ensures all have `id`
     for (const file of semanticResults) {
         const resolved = path.resolve(file.path);
         seen.add(resolved);
@@ -48,7 +48,7 @@ export async function runAskCommand(query) {
         if (!seen.has(resolved)) {
             seen.add(resolved);
             combinedResults.push({
-                id: file.id, // Ensure the id is included here
+                id: file.id,
                 path: file.path,
                 summary: file.summary || '',
                 score: 0.0,
@@ -73,9 +73,12 @@ export async function runAskCommand(query) {
     let code = '';
     let topSummary = topFile.summary || '(No summary available)';
     let topFunctions = [];
-    // Gather all file IDs from the combined results
+    // Truncate summary if needed
+    if (topSummary) {
+        topSummary = topSummary.split('\n').slice(0, MAX_SUMMARY_LINES).join('\n');
+    }
     const allFileIds = combinedResults
-        .map(file => file.id) // Now file.id exists
+        .map(file => file.id)
         .filter((id) => typeof id === 'number');
     const allFunctionsMap = getFunctionsForFiles(allFileIds);
     try {
@@ -89,9 +92,13 @@ export async function runAskCommand(query) {
     // 🟩 STEP 5: Build relatedFiles with functions
     const relatedFiles = combinedResults.slice(0, MAX_RELATED_FILES).map(file => {
         const fileId = file.id;
+        let summary = file.summary || '(No summary available)';
+        if (summary) {
+            summary = summary.split('\n').slice(0, MAX_SUMMARY_LINES).join('\n');
+        }
         return {
             path: file.path,
-            summary: file.summary || '(No summary available)',
+            summary,
             functions: allFunctionsMap[fileId]?.map(fn => fn.name) || [],
         };
     });

package/dist/db/fileIndex.js

@@ -4,6 +4,7 @@ import path from 'path';
 import { generateEmbedding } from '../lib/generateEmbedding.js';
 import { sanitizeQueryForFts } from '../utils/sanitizeQuery.js';
 import * as sqlTemplates from './sqlTemplates.js';
+import { stringSimilarity } from 'string-similarity-js';
 /**
  * 📄 Index a single file into the database.
  *
@@ -67,7 +68,6 @@ export async function searchFiles(query, topK = 5) {
     }
     const safeQuery = sanitizeQueryForFts(query);
     console.log(`Executing search query in FTS5: ${safeQuery}`);
-    // Step 1: Narrow candidate set using fast keyword match
     const ftsResults = db.prepare(`
     SELECT fts.rowid AS id, f.path, f.summary, f.type, bm25(files_fts) AS bm25Score, f.embedding
     FROM files f
@@ -80,30 +80,39 @@ export async function searchFiles(query, topK = 5) {
     if (ftsResults.length === 0) {
         return [];
     }
-    // Step 2: Compute score based on embedding similarity + BM25 score
     const bm25Min = Math.min(...ftsResults.map(r => r.bm25Score));
     const bm25Max = Math.max(...ftsResults.map(r => r.bm25Score));
     const scored = ftsResults.map(result => {
-        let finalScore = 0;
         let sim = 0;
+        let finalScore = 0;
+        const normalizedBm25 = 1 - ((result.bm25Score - bm25Min) / (bm25Max - bm25Min + 1e-5));
         if (result.embedding) {
             try {
                 const vector = JSON.parse(result.embedding);
                 sim = cosineSimilarity(embedding, vector);
-                const normalizedBm25 = 1 - ((result.bm25Score - bm25Min) / (bm25Max - bm25Min + 1e-5));
-                finalScore = 0.7 * sim + 0.3 * normalizedBm25;
             }
             catch (err) {
                 console.error(`❌ Failed to parse embedding for ${result.path}:`, err);
-                finalScore = 0;
             }
         }
-        else {
-            // Fallback to BM25-only score
-            finalScore = 1 - ((result.bm25Score - bm25Min) / (bm25Max - bm25Min + 1e-5));
+        const terms = query.toLowerCase().split(/\s+/);
+        const path = result.path.toLowerCase();
+        const summary = (result.summary || '').toLowerCase();
+        let termMatches = 0;
+        for (const term of terms) {
+            if (path.includes(term) || summary.includes(term)) {
+                termMatches += 1;
+            }
         }
+        const matchRatio = termMatches / terms.length;
+        const termBoost = matchRatio >= 1 ? 1.0 : matchRatio >= 0.5 ? 0.5 : 0;
+        // 🧠 Final score with hybrid weighting (BM25 + Embedding + Term Boost)
+        finalScore = 0.4 * normalizedBm25 + 0.4 * sim + 0.2 * termBoost;
+        // ✅ Fuzzy score using string-similarity-js
+        const fuzzyScore = stringSimilarity(query.toLowerCase(), `${path} ${summary}`);
+        finalScore += fuzzyScore * 10;
         return {
-            id: result.id, // Ensure the id is included here
+            id: result.id,
             path: result.path,
             summary: result.summary,
             score: finalScore,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "scai",
-  "version": "0.1.45",
+  "version": "0.1.46",
   "type": "module",
   "bin": {
     "scai": "./dist/index.js"
@@ -30,7 +30,8 @@
     "better-sqlite3": "^12.1.1",
     "commander": "^11.0.0",
     "fast-glob": "^3.3.3",
-    "proper-lockfile": "^4.1.2"
+    "proper-lockfile": "^4.1.2",
+    "string-similarity-js": "^2.1.4"
   },
   "devDependencies": {
     "@types/better-sqlite3": "^7.6.13",