Indexed Collection (for RAG applications)¶
An indexed collection lets your chatbot search through your documents to find relevant information before responding. Instead of relying on the AI's built-in knowledge, the chatbot retrieves answers from files you upload — such as PDFs, reports, or wiki pages.
When should I use this?¶
When you want your chatbot's responses to be grounded in your uploaded documents.
Common examples include:
- A customer support chatbot that answers questions from your product documentation or FAQs
- An HR assistant that looks up company policies from an internal wiki or handbook
- A research tool that searches across uploaded reports, studies, or reference materials
- An onboarding guide that walks new users through your own uploaded training content
How it works¶
To search documents by meaning, OCS uses an embedding model — this technique is called Retrieval-Augmented Generation (RAG). See Local Index Optimization for a full explanation.
Which should I use?¶
In OCS, there are two types of indexes:
| Remote Index | Local Index | |
|---|---|---|
| Managed by | Your LLM provider (e.g. OpenAI) | OCS |
| Setup | Simpler — the provider handles everything | More steps — you choose the embedding model |
| Embedding model | Selected by the provider | You choose |
| Chunking | Handled by provider, not configurable | Configurable per file set |
| Collections per LLM node | Max 2 (OpenAI limit) | Unlimited |
| Best for | Getting started quickly | More control, or more than 2 collections |
If you are new to indexed collections, start with a Remote Index. Switch to a Local Index if you need more than 2 collections or want to choose a specific embedding model for your content type.
Remote Index¶
Remote indexes are hosted and managed by your LLM provider. Files are uploaded to the provider, which handles all indexing. The embedding model is chosen by the provider.
Supported providers¶
- OpenAI (using the responses API)
OpenAI Remote Collection Limit
When using OpenAI as your LLM provider with remote (OpenAI-hosted) indexed collections, you can select a maximum of 2 collections per LLM node. This is a limitation imposed by OpenAI's API, not Open Chat Studio.
- Local indexes are NOT affected by this limit
- Non-OpenAI providers are NOT affected by this limit
- If you attempt to select more than 2 remote collections with OpenAI, you will receive a validation error
Supported file types¶
Supported files are determined by the selected provider:
- OpenAI - See the OpenAI docs
Local Index¶
Local indexes are hosted and managed by OCS. When you create a local index, you choose which embedding model to use. Different models suit different types of content, so choosing the right one can improve retrieval accuracy. See Local Index Optimization for guidance.
Indexing Options¶
- Supported LLM providers: OpenAI, Voyage AI, Google Gemini
- Supported file types: pdf, txt, csv, docx
- Supported embedding models: You can see the list of embedding models for the LLM provider you have selected.
Chunking and Optimization¶
When you upload a document to a local index, OCS breaks it into smaller parts called chunks and stores them in the index. The default chunking settings work well for most use cases.
For advanced configuration — including chunk size, chunk overlap, and embedding model selection — see Local Index Optimization.
Document Sources for Indexed Collections¶
Instead of uploading files manually, you can connect OCS to an external document source — such as a Confluence space or GitHub repository — and have it fetch and index content automatically on a schedule. This keeps your OCS indexed collection (for both remote and local indexes) current without manual uploads.
Document-source updates reach published chatbots automatically
When a document-source sync runs and updates the collection's content, those changes are applied to your published chatbot without requiring a republish. See Collections and published chatbots for more detail.
Currently supported sources: Confluence and GitHub.
For configuration steps, authentication setup, and how to monitor sync status, see Set Up Document Sources.