Gemini File Search (RAG)

The gemini extension provides a complete UI for managing Google Gemini File Search Stores and using them as managed Retrieval Augmented Generation knowledge bases - without asking your team to design a chunking pipeline, vector database, background indexer, retrieval tool and administration UI from scratch.

Enabling it​

The extension self-enables when both of these are available:

1. A Gemini API key - GEMINI_API_KEY, or whatever the google provider resolved.
2. A registered IDbConnectionFactory, since the document catalog is an OrmLite table.

export GEMINI_API_KEY=...

Or programmatically:

services.AddPlugin(new ChatFeature {
    Variables = {
        ["GEMINI_API_KEY"] = context.Configuration["Gemini:ApiKey"]!,
    },
});

Without either prerequisite the extension logs why and disables itself - no routes, no tools, no UI.

To turn it off explicitly:

services.AddPlugin(new ChatFeature {
    DisableExtensions = ["gemini"],
});

Stores, categories and documents​

Teams create separate stores for departments, customers, products or projects, then drag and drop PDFs, Markdown, text and supported business documents into categories within them.

Uploads are:

• Content-addressed - written into App_Data/chat/cache under their SHA-256
• Deduplicated - the same document uploaded twice is stored once
• Asynchronous - a background worker performs the Gemini upload, so indexing never blocks the UI

Queued uploads are resumed on startup if the App shut down mid-upload.

Grounded conversations at three scopes​

Once indexed, a user can start a grounded conversation at whichever scope fits the question:

Grounding is implemented by including an OpenAI-shaped file_search tool in the thread, which the Google provider forwards to Gemini. Grounding sources are retained with responses, so users can verify where an answer came from instead of treating fluent output as evidence.

Synchronization​

The local catalog and Gemini's view of a store can drift - an upload failed, a document was removed remotely, metadata changed. Sync reconciles them:

• Matches by the SHA-256 hash Gemini keeps in each document's custom metadata, falling back to the document name
• Refreshes stale local fields
• Records what didn't line up as the document's state

Storage economics​

Google currently charges for embeddings when documents are indexed, whilst file storage and query-time embeddings are free. Retrieved document tokens are billed as normal model context, and stores persist until deleted with project capacity scaling by usage tier.

That removes the recurring vector-storage charge common in managed RAG architectures - an organization can maintain durable knowledge bases and pay primarily when content is indexed and when users actually query it.

What AI Chat adds around Google's service​

• Local document catalog and per-document status
• Store and category organization
• Background upload queue with resume-on-startup
• SHA-256 deduplication
• Remote/local synchronization reports
• Re-upload and delete controls
• User-scoped management
• One-click grounded chat creation

Per-user isolation​

File Stores and their document catalogs are scoped to the authenticated user, like every other piece of AI Chat state. A store one user creates is not visible to another.

For a genuinely shared organizational knowledge base, create it under a shared account or expose it through an API Tool that queries it on the user's behalf with your own authorization rules.

API​

Operational notes​

• The document catalog tables are created when AutoInitSchema is true.
• The upload worker is stopped cleanly on AppHost disposal via a registered shutdown handler.
• Cached document bytes live under App_Data/chat/cache and should be included in your backups - see Data & Storage.