Summary Agentic File Search tackles traditional RAG limitations by offering an AI agent for dynamic file system navigation. It targets users needing to extract information from complex, cross-referenced documents, providing more context-aware and relevant search than static embedding methods.

How It Works It uses a three-phase strategy: Parallel Scan (preview all), Deep Dive (extract relevant), and Backtrack (follow cross-references). This agentic approach, powered by Google Gemini 3 Flash and LlamaIndex Workflows, avoids chunking context loss and identifies logical connections missed by traditional RAG. Six file tools (scan_folder, preview_file, parse_file, read, grep, glob) enable dynamic exploration.

Quick Start & Requirements Clone the repo, cd into it, and install via uv pip install . or pip install .. Configuration requires a .env file with a GOOGLE_API_KEY from Google AI Studio. Usage is via CLI (uv run explore --task "...") or Web UI (uv run uvicorn fs_explorer.server:app --host 127.0.0.1 --port 8000).

Highlighted Details

• Tools: scan_folder, preview_file, parse_file, read, grep, glob.
• Document Support: PDF, DOCX, PPTX, XLSX, HTML, Markdown (via local Docling).
• LLM: Google Gemini 3 Flash with structured JSON output.
• Cost: ~$0.001 per query with token tracking.
• Web UI: Real-time WebSocket streaming, folder browser, logs, citations, cost stats.
• Test Data: Includes sample datasets (data/test_acquisition/, data/large_acquisition/).

Maintenance & Community The README provides no details on contributors, community channels, sponsorships, or roadmap.

Licensing & Compatibility Released under the MIT License, generally permitting commercial use and integration into closed-source projects.

Limitations & Caveats Effectiveness depends on Google Gemini 3 Flash quality and Docling parser accuracy. Requires a Google API key; primarily focused on file system exploration.