scai 0.1.63 → 0.1.65

package/README.md

@@ -269,6 +269,34 @@ These features are experimental and subject to change:
 
 <br>
 
+## Commands
+
+### `index`
+The `index` command is used to manage and perform operations on the indexed files and repositories.
+
+#### `scai index start`
+
+Index supported files in the configured index directory.
+
+```bash
+scai index set <dir>
+```
+
+Set and activate the index directory.
+```bash
+scai index list
+```
+
+List all indexed repositories.
+```bash
+scai index switch
+```
+
+Switch active repository (by key or indexDir). Run without input for an interactive list of repositories.
+
+
+--- 
+
 > **Note:** Indexing very large repositories (millions of lines) may take **hours or days**. Please be patient, and only index huge codebases if you are ok with some extra processing taking place on your computer.
 
 The `scai index` command **automatically** starts a background daemon that continuously:
@@ -322,7 +350,7 @@ You can run it in two ways:
 1. **Inline question**
 
    ```bash
-   scai ask "How does mapcontrol work?"
+   scai ask "How does the controller work?"
    ```
 
 2. **Interactive prompt**
@@ -335,7 +363,7 @@ You can run it in two ways:
    , then type your question when prompted:
 
    ```
-   > How does mapcontrol work?
+   > How does the controller work?
    ```
 
    </br>

package/dist/db/fileIndex.js

@@ -3,77 +3,46 @@ 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';
 import { CANDIDATE_LIMIT } from '../constants.js';
 import { getDbForRepo } from './client.js';
-/**
- * 📄 Index a single file into the database.
- *
- * - Normalizes the file path to prevent OS-specific path bugs.
- * - Stores metadata like summary, type, lastModified, etc.
- * - Sets `embedding` to null initially — computed later.
- */
+import { scoreFiles } from '../fileRules/scoreFiles.js'; // 👈 NEW
 export function indexFile(filePath, summary, type) {
     const stats = fs.statSync(filePath);
     const lastModified = stats.mtime.toISOString();
     const indexedAt = new Date().toISOString();
     const normalizedPath = path.normalize(filePath).replace(/\\/g, '/');
-    const fileName = path.basename(normalizedPath); // Extracting the filename
-    // Insert into files table
+    const fileName = path.basename(normalizedPath);
     const db = getDbForRepo();
     db.prepare(sqlTemplates.upsertFileTemplate).run({
         path: normalizedPath,
-        filename: fileName, // Pass filename
+        filename: fileName,
         summary,
         type,
         lastModified,
         indexedAt,
         embedding: null
     });
-    // Insert into files_fts table for full-text search, including filename
     db.prepare(`
-  INSERT OR REPLACE INTO files_fts (rowid, filename, summary, path)
-  VALUES ((SELECT id FROM files WHERE path = :path), :filename, :summary, :path)
-`).run({
+    INSERT OR REPLACE INTO files_fts (rowid, filename, summary, path)
+    VALUES ((SELECT id FROM files WHERE path = :path), :filename, :summary, :path)
+  `).run({
         path: normalizedPath,
         filename: fileName,
-        summary: summary,
+        summary,
     });
     console.log(`📄 Indexed: ${normalizedPath}`);
 }
-/**
- * 🔍 FTS5 keyword-based search using SQLite's full-text index.
- *
- * Use this when you want:
- * - A simple keyword search.
- * - Fast fallback search that doesn't rely on embeddings.
- *
- * Returns a full `FileRow[]` (all known metadata, but no similarity score).
- */
 export function queryFiles(safeQuery, limit = 10) {
     console.log(`Executing search query: ${safeQuery}`);
     const db = getDbForRepo();
-    const results = db.prepare(`
+    return db.prepare(`
     SELECT f.id, f.path, f.filename, f.summary, f.type, f.last_modified, f.indexed_at
     FROM files f
     JOIN files_fts fts ON f.id = fts.rowid
     WHERE fts.files_fts MATCH ?
     LIMIT ?
   `).all(safeQuery, limit);
-    return results;
 }
-/**
- * 🧠 Combined semantic + FTS search (Hybrid).
- *
- * 1. Convert user query to embedding vector using OpenAI / Ollama etc.
- * 2. Perform a keyword-based FTS match to limit candidates.
- * 3. For each candidate, compute cosine similarity between query + file embedding.
- * 4. Blend the BM25 score and cosine similarity to produce a final score.
- * 5. Return top K most relevant results.
- *
- * ⚠️ This returns a **lighter-weight type** than `FileRow` — doesn't include id, timestamps, etc.
- * Use a wrapper type like `SearchResultWithScore` in calling code if you need both.
- */
 export async function searchFiles(query, topK = 5) {
     console.log(`🧠 Searching for query: "${query}"`);
     const embedding = await generateEmbedding(query);
@@ -93,70 +62,11 @@ export async function searchFiles(query, topK = 5) {
     LIMIT ?
   `).all(safeQuery, CANDIDATE_LIMIT);
     console.log(`FTS search returned ${ftsResults.length} results`);
-    if (ftsResults.length === 0) {
+    if (ftsResults.length === 0)
         return [];
-    }
-    const bm25Min = Math.min(...ftsResults.map(r => r.bm25Score));
-    const bm25Max = Math.max(...ftsResults.map(r => r.bm25Score));
-    const scored = ftsResults.map(result => {
-        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);
-            }
-            catch (err) {
-                console.error(`❌ Failed to parse embedding for ${result.path}:`, err);
-            }
-        }
-        const terms = query.toLowerCase().split(/\s+/);
-        const path = result.path.toLowerCase();
-        const summary = (result.summary || '').toLowerCase();
-        let termMatches = 0;
-        for (const term of terms) {
-            if (path.includes(term) || summary.includes(term)) {
-                termMatches += 1;
-            }
-        }
-        const matchRatio = termMatches / terms.length;
-        const termBoost = matchRatio >= 1 ? 1.0 : matchRatio >= 0.5 ? 0.5 : 0;
-        // 🧠 Final score with hybrid weighting (BM25 + Embedding + Term Boost)
-        finalScore = 0.4 * normalizedBm25 + 0.4 * sim + 0.2 * termBoost;
-        // ✅ Fuzzy score using string-similarity-js
-        const fuzzyScore = stringSimilarity(query.toLowerCase(), `${path} ${summary}`);
-        finalScore += fuzzyScore * 10;
-        return {
-            id: result.id,
-            path: result.path,
-            summary: result.summary,
-            score: finalScore,
-            sim,
-            bm25: result.bm25Score,
-        };
-    });
-    const sorted = scored
-        .sort((a, b) => b.score - a.score)
-        .slice(0, topK);
-    console.log(`Returning top ${topK} results based on combined score`);
-    return sorted;
-}
-/**
- * 🔢 Cosine similarity between two vectors.
- */
-function cosineSimilarity(a, b) {
-    const dot = a.reduce((sum, ai, i) => sum + ai * b[i], 0);
-    const magA = Math.sqrt(a.reduce((sum, ai) => sum + ai * ai, 0));
-    const magB = Math.sqrt(b.reduce((sum, bi) => sum + bi * bi, 0));
-    return dot / (magA * magB);
+    const scored = scoreFiles(query, embedding, ftsResults);
+    return scored.slice(0, topK);
 }
-/**
- * 🧠 Get all extracted functions for a given list of file IDs.
- *
- * - Used in `askCmd.ts` to fetch code snippets from relevant files.
- * - Grouped by file_id.
- */
 export function getFunctionsForFiles(fileIds) {
     if (!fileIds.length)
         return {};

package/dist/fileRules/scoreFiles.js

@@ -0,0 +1,71 @@
+import { stringSimilarity } from 'string-similarity-js';
+export function scoreFiles(query, embedding, candidates) {
+    const terms = query.toLowerCase().split(/\s+/);
+    const bm25Min = Math.min(...candidates.map(r => r.bm25Score));
+    const bm25Max = Math.max(...candidates.map(r => r.bm25Score));
+    return candidates.map(result => {
+        let finalScore = 0;
+        let sim = 0;
+        const path = result.path.toLowerCase();
+        const filename = result.filename.toLowerCase();
+        const summary = (result.summary || '').toLowerCase();
+        // 🎯 Normalize BM25
+        const normalizedBm25 = 1 - ((result.bm25Score - bm25Min) / (bm25Max - bm25Min + 1e-5));
+        // 🧠 Embedding similarity
+        if (result.embedding) {
+            try {
+                const vector = JSON.parse(result.embedding);
+                sim = cosineSimilarity(embedding, vector);
+            }
+            catch {
+                sim = 0;
+            }
+        }
+        // 🧩 Match ratio
+        let matchRatio = 0;
+        let matchedTerms = 0;
+        for (const term of terms) {
+            if (path.includes(term) || summary.includes(term))
+                matchedTerms++;
+        }
+        matchRatio = matchedTerms / terms.length;
+        const termBoost = matchRatio >= 1 ? 1.0 : matchRatio >= 0.5 ? 0.5 : 0;
+        // 🪜 Path heuristics
+        const isHtml = path.endsWith('.html');
+        const isSrc = path.includes('/src/') || path.includes('/controls/');
+        const isDoc = path.includes('/docs/') || path.includes('/mvndist/');
+        const isExactMatch = filename === `${terms[0]}.js`;
+        let weight = 1;
+        if (isHtml)
+            weight *= 0.85;
+        if (isDoc)
+            weight *= 0.8;
+        if (isSrc)
+            weight *= 1.2;
+        if (isExactMatch)
+            weight *= 1.5;
+        // 🧪 Fuzzy score
+        const fuzzyScore = stringSimilarity(query, `${path} ${summary}`);
+        // 🧮 Final composite score
+        finalScore =
+            (0.4 * normalizedBm25) +
+                (0.4 * sim) +
+                (0.2 * termBoost) +
+                (fuzzyScore * 0.5); // scale fuzzy match moderately
+        finalScore *= weight;
+        return {
+            id: result.id,
+            path: result.path,
+            summary: result.summary,
+            score: finalScore,
+            sim,
+            bm25: result.bm25Score,
+        };
+    }).sort((a, b) => b.score - a.score);
+}
+function cosineSimilarity(a, b) {
+    const dot = a.reduce((sum, ai, i) => sum + ai * b[i], 0);
+    const magA = Math.sqrt(a.reduce((sum, ai) => sum + ai * ai, 0));
+    const magB = Math.sqrt(b.reduce((sum, bi) => sum + bi * bi, 0));
+    return dot / (magA * magB);
+}

package/dist/index.js

@@ -24,7 +24,7 @@ import { reviewPullRequestCmd } from "./commands/ReviewCmd.js";
 import { promptForToken } from "./github/token.js";
 import { validateGitHubTokenAgainstRepo } from "./github/githubAuthCheck.js";
 import { checkGit } from "./commands/GitCmd.js";
-import { runSwitchCommand, runInteractiveSwitch } from "./commands/SwitchCmd.js";
+import { runInteractiveSwitch } from "./commands/SwitchCmd.js";
 import { execSync } from "child_process";
 import { fileURLToPath } from "url";
 import { dirname, resolve } from "path";
@@ -166,15 +166,10 @@ index
     Config.printAllRepos(); // 👈 simple and clean
 });
 index
-    .command('switch [input]')
-    .description('Switch active repository (by key or indexDir). Run without input for a list')
-    .action((input) => {
-    if (input) {
-        runSwitchCommand(input);
-    }
-    else {
-        runInteractiveSwitch();
-    }
+    .command('switch')
+    .description('Switch active repository (by interactive list only)')
+    .action(() => {
+    runInteractiveSwitch();
 });
 // This will help resolve the current directory in an ES Module
 cmd

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "scai",
-  "version": "0.1.63",
+  "version": "0.1.65",
   "type": "module",
   "bin": {
     "scai": "./dist/index.js"