Function buildIndex

src/context/retrieval.ts:248–276

(
  projectRoot: string,
  embeddingModel: string,
  baseUrl: string,
  maxFiles: number,
  signal: AbortSignal | undefined,
  onProgress: (msg: string) => void,
)

}

export async function buildIndex(
  projectRoot: string,
  embeddingModel: string,
  baseUrl: string,
  maxFiles: number,
  signal: AbortSignal | undefined,
  onProgress: (msg: string) => void,
): Promise<Index> {
  const files = await walkSource(projectRoot, maxFiles);
  onProgress(`Indexing ${files.length} file(s)…`);
  const chunks: Chunk[] = [];
  // AST-aware chunking (semantic boundaries) with line-based fallback per file.
  const { astChunkFile } = await import('./ast-chunk.js');
  for (const f of files) {
    try {
      chunks.push(...await astChunkFile(f.rel, f.content));
    } catch {
      chunks.push(...chunkFile(f.rel, f.content, 40, 8));
    }
  }
  onProgress(`Embedding ${chunks.length} chunk(s) with ${embeddingModel}…`);
  const { skipped } = await embedChunksResilient(
    chunks,
    (texts) => ollamaEmbed(baseUrl, embeddingModel, texts, signal),
    { batchSize: 32, onProgress },
  );
  if (skipped > 0) onProgress(`  (skipped ${skipped} oversized/unembeddable chunk(s))`);
  return { projectRoot, embeddingModel, chunks, builtAt: Date.now() };
}

// ── Pure ranking / aggregation (unit-testable, no model) ────────────────────────