Needle indexes PDFs, DOCX, XLSX, plain text files, and more. When documents are uploaded to a collection, Needle automatically chunks, embeds, and indexes the content for semantic retrieval. The agent can search across all indexed file types in a single query.

Keyword search matches exact terms. Needle's semantic search understands meaning, so a query about 'return policy' finds relevant content even if the document says 'refund guidelines' or 'product exchange process.' This means the agent finds answers that keyword search would miss.

Yes. Needle returns source references with every search result, including the file name and the specific passage that matched. The agent presents these citations in its response, and can generate a download link for the original document.

No. Documents live in your Needle account. Tars queries Needle's search API in real time and receives text passages and metadata. The original files are never copied to or stored by Tars. You maintain full control over your document repository.

Needle automatically processes and indexes documents after upload. For most file types, content becomes searchable within minutes. Large documents or bulk uploads may take slightly longer. The agent can check collection stats to confirm indexing is complete.

Yes. When configuring the agent in Tars, you specify which Needle collection IDs the agent can query. You can restrict customer-facing agents to public documentation while internal agents search HR policies or engineering docs.

The max_distance parameter controls how strict the relevance threshold is, on a scale from 0 to 1. Lower values return only highly relevant results. Higher values cast a wider net. The agent uses this to balance precision and recall based on the query context.

FAQ pages require users to browse and find the right article. With Needle and Tars, customers ask questions in their own words and get precise answers extracted from your full document library. It handles questions that were never anticipated in a pre-written FAQ.