FileChat RAG

Search, retrieve, and chat with documents securely stored in ANY Google Drive folder using semantic vector search (RAG). Use when: (1) a user asks to save a file for later, (2) a user asks a question about their saved files/documents, (3) a user asks you to retrieve or send them a previously stored file, (4) a user asks to sync or flush their document library. This skill automatically chunks, embeds, and indexes Google Drive documents into isolated persistent vector databases per folder.

FileChat RAG Skill

Your personal RAG (Retrieval-Augmented Generation) document library backed by Google Drive. Supports multiple Google Drive folders dynamically and allows choosing between Gemini or OpenAI for embeddings.

Setup & Bootstrap

If the user asks to use FileChat or asks a question about their files, FIRST verify that the required environment variables are set in /workspace/skills/filechat/.env:

1. EMBEDDING_PROVIDER (either gemini or openai)
2. GEMINI_API_KEY or OPENAI_API_KEY (Depending on the provider)

If they are missing, STOP and ask the user to provide them. Create the .env file like this:

bash

echo "EMBEDDING_PROVIDER=gemini" > ./skills/filechat/.env
echo "GEMINI_API_KEY=your_key_here" >> ./skills/filechat/.env

How to Sync the Library

When the user asks to "sync", "flush", or "update" a specific FileChat folder, you must run the ingestion script. This connects to Google Drive, downloads all new/changed files (including PDFs, resolving shortcuts, and traversing sub-folders), chunks the text, gets embeddings, and saves them to a local JSON vector database keyed by the Folder ID.

You must supply the FOLDER_ID to the script. If you don't know the folder ID the user wants, ask them.

bash

cd ./skills/filechat && node sync.js <FOLDER_ID>

How to Answer User Questions (RAG)

When a user asks a question about the contents of their documents, you MUST query the local vector store to fetch the relevant text chunks. You need the Folder ID.

bash

cd ./skills/filechat && node query.js <FOLDER_ID> "What does my medical discharge say?"

The output will give you the most relevant text snippets, the original file names, and the Google Drive File IDs. Use the text snippets to formulate a comprehensive answer for the user. Always cite the file name you are referencing.

How to Retrieve and Send a Physical File

If the user asks for the actual file (e.g., "Send me the discharge PDF"), first find the File ID using the query script.

Then, use the gws CLI to download the file into your workspace:

bash

gws drive files get --params '{"fileId": "<FILE_ID>", "alt": "media"}' --output /workspace/discharge.pdf

Then, reply to the user using the OpenClaw media attachment syntax: MEDIA:/workspace/discharge.pdf to send the physical file directly to their chat window.

How to Store a New File for the User

If the user uploads a file and asks you to "store" or "save" it:

1. Upload it to their specific FileChat Google Drive folder using gws:

   bash

   gws drive files create \
     --json '{"name": "filename.pdf", "parents": ["<FOLDER_ID>"]}' \
     --upload /path/to/uploaded/file.pdf
   

2. Trigger the sync process so the new file is immediately readable by the vector database:

   bash

   cd ./skills/filechat && node sync.js <FOLDER_ID>