FileChat RAG Skill

Your personal RAG (Retrieval-Augmented Generation) document library backed by Google Drive.
Supports multiple Google Drive folders dynamically, interactive folder routing, incremental sync, choosing between Gemini or OpenAI for embeddings, and connecting to Qdrant.

Setup & Bootstrap

FIRST verify that the required environment variables are set in /workspace/skills/filechat/.env:

1. 1. EMBEDDING_PROVIDER (either gemini or openai)
2. INLINECODE4 or INLINECODE5
3. Optional: QDRANT_URL and QDRANT_API_KEY (If absent, it uses local disk-based JSON).

Create the .env file like this:
CODEBLOCK0

Google Workspace Authentication:
Before running any commands, check if the system is authenticated by running:

npx @googleworkspace/cli auth status

npx @googleworkspace/cli auth login --services drive

Folder Management

The user can have infinite folders synced. You manage them using folders.js.

• - List Folders: INLINECODE10
• Add a Folder: node folders.js add "Taxes 2026" <FOLDER_ID> (Auto-discovers the ID via gws drive files list if you don't know it!)
• Set Default Folder: INLINECODE13

If the user asks to do something with a file/folder but doesn't specify which one, run node folders.js get-default to find the default ID. If no folders exist, ask them to set one up!

How to Sync the Library

When the user asks to "sync", "flush", or "update", you must run the ingestion script.

To sync a specific folder:
CODEBLOCK3

To sync EVERYTHING (all folders in the registry):
CODEBLOCK4

Note: Syncs are highly incremental and use a local cache! If a file hasn't been modified in Drive, the script will skip it instantly and output "0 chunks" embedded. This is NORMAL behavior. If you are debugging, testing, or the user specifically requests a hard flush, you MUST delete the cache files first:
CODEBLOCK5

How to Answer User Questions (RAG)

Query the local vector store or Qdrant for the target Folder ID to fetch relevant text chunks:

CODEBLOCK6
Use the snippets returned to answer the user.

How to Retrieve and Send a Physical File

Find the File ID using the query script, then download it:

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

How to Store a New File for the User

If the user uploads a file and asks you to save it (or implicitly sends a file per your automatic processing rules):

1. 1. Check their folders (node folders.js list).
2. If they didn't specify which folder, use the default folder. If no default is set, ask them!
3. Notify the user exactly which folder the file is being saved to.
4. Tell the user that you are now extracting the information and saving it in a vectordb.
5. If the file is an image or scanned document, make sure to extract the text using a vision model or OCR before it is embedded. (The sync script handles this natively).
6. Upload it to the correct folder using gws:

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

1. 7. Trigger node sync.js <FOLDER_ID> so the vector database chunks and embeds the file into the corresponding vectordb.

How to Test & Validate the Skill

If the user asks you to verify the skill is working, or if you just set it up and want to ensure end-to-end functionality, follow these exact steps:

1. 1. Verify Auth: Run npx @googleworkspace/cli auth status. Ensure it shows a valid token.
2. Verify Drive Access: Do a dry-run fetch of the target folder to ensure GWS can see the files.

   npx @googleworkspace/cli drive files list --params '{"q": "'\''<FOLDER_ID>'\'' in parents and trashed = false"}'
   

1. 3. Force a Clean Sync: Clear the cache for the test folder to guarantee a fresh run, then sync.

   rm -f ./skills/filechat/meta_<FOLDER_ID>.json ./skills/filechat/vector_db_<FOLDER_ID>.json
   node ./skills/filechat/sync.js <FOLDER_ID>
   

1. 4. Test the Vector Query: Run a generic query to verify the embeddings were saved and cosine similarity works.

   node ./skills/filechat/query.js <FOLDER_ID> "hello"