Vectara RAG
Vectara RAG asks your Vectara corpus a question and directly returns an answer written from the documents found. Vectara takes care of searching, ordering the results and summarising them, so you do not need to connect a model or an embeddings component.
When to use it
- Answering questions about your documents stored in Vectara with a single component.
- Building an answer search for employees or customers without designing an agent.
- Getting the answer in a specific language.
Configuration
The access details are typed in the node itself.
| Field | What it does |
|---|---|
| Vectara Customer ID (required) | Your Vectara account ID. |
| Vectara Corpus ID (required) | The ID of the corpus that holds your documents. |
| Vectara API Key (required) | The Vectara access key. |
| Search Query | The question you want answered. |
| Reranker Type | How results are ordered before summarising: mmr (default), which gives varied results; rerank_multilingual_v1; or none, no reordering. |
Advanced options
| Field | What it does |
|---|---|
| Hybrid Search Factor | How much exact words weigh against meaning, from 0 to 1. At 0 only meaning counts and at 1 only words. Default: 0.005. |
| Metadata Filters | Conditions to search only among documents with certain data, using Vectara’s filter syntax. |
| Number Of Results To Rerank | How many results are reordered. Default: 50. |
| Diversity Bias | From 0 to 1; higher gives more varied results. Only with mmr. Default: 0.2. |
| Max Results To Summarize | How many results are used to write the answer. Default: 7. |
| Response Language | The answer’s language, by its code, for example spa for Spanish, or auto to detect it. Default: eng. |
| Prompt Name | Vectara’s summarising instructions. Default: vectara-summary-ext-24-05-sml, the only one available on Growth plans; the others are for Scale plans. |
Output
The Answer output is the written answer, ready to show in a chat or send by email.
Example
An HR department wants to answer questions about the collective agreement.
- Upload the agreement to a Vectara corpus.
- Add Vectara RAG with the account and corpus details.
- Connect the employee’s question to Search Query and choose
spaorengin Response Language. - Show Answer in the chat.
Good practices
- Set Response Language to match your users: by default it answers in English.
Related components
- Vectara: stores documents in Vectara and searches them.