Documents

What this page is for

The Documents page is where you create collections, upload files, and attach them to chatbots. It’s the heart of Exolvra’s Retrieval-Augmented Generation (RAG) story — you provide the knowledge, Exolvra handles the indexing, the chatbot handles the conversation.

Use it when you want a chatbot that answers questions from your own content: product documentation, help articles, internal wikis, policy handbooks, SOPs. Anything written down that a bot should reference.

Layout

The Documents page has two main views:

1. Collection list — one card per collection showing name, description, document count, total size, and last updated
2. Collection detail — click into any collection to see the full list of uploaded documents with individual upload status, indexing state, and per-document actions

Each document card in the detail view shows:

• Filename and size
• Status — Uploading, Indexing, Ready, or Failed
• Chunk count — how many searchable chunks the document was split into
• Uploaded by and when

Common tasks

Create a new collection

1. Click New collection on the Documents page
2. Name it — the name becomes the collection’s id, used when linking it to a bot (product-docs, company-handbook, api-reference)
3. Add a description
4. Click Create

You land on the empty collection detail page, ready for uploads.

Upload documents

1. Open the collection detail page
2. Drag files onto the upload area, or click Upload to pick them
3. Supported formats: Markdown (.md), plain text (.txt), PDF (.pdf), and most plain-text formats
4. Files upload one at a time, then enter Indexing status — Exolvra chunks them, generates embeddings, and stores the vectors
5. When status flips to Ready, the chunks are searchable

Large documents (hundreds of pages) can take a few minutes to index. You don’t have to wait — leave the tab open or come back later.

Attach a collection to a chatbot

1. Go to /bots/{id}/edit
2. Section 04 — Knowledge, pick the collection from the dropdown
3. Save

The chatbot now does RAG on every user message — top-5 chunks from the collection are injected into its context before responding.

Test retrieval quality

Each collection detail page has a Search tab. Type a query to see which chunks would be returned for that query. Use this to test whether your content is indexing well — if queries that should find obvious chunks don’t, you may need to preprocess the source material (split huge files, add headers, strip HTML boilerplate).

Delete a document

Click the trash icon on the document card. Confirm. The document is removed from the collection and its chunks are deleted from the index. Any chatbots using this collection will no longer retrieve from it.

Delete a collection

From the collection detail page, click Delete collection. All documents inside are also deleted. Chatbots linked to this collection lose their knowledge — the link breaks, and the bot reverts to ungrounded responses on every turn. Fix the bot configuration before or after.

How retrieval works

When a user sends a message to a chatbot with knowledge attached:

1. Exolvra runs a semantic search across the attached collection using the user’s message as the query
2. The top N matches (default 5) are pulled back as chunks of text
3. These chunks are injected into the bot’s context under a ## Relevant Knowledge section in the system prompt
4. The bot sees the user’s message, its own personality, and the retrieved chunks — it decides how to respond

A well-grounded response cites the source document. Prompt the bot explicitly to cite sources if you want this behavior.

Common pitfalls

Uploading one giant PDF. Large monolithic files are hard to chunk well. If you can, split your source material into smaller, topic-focused documents before uploading. One-page-per-topic is often better than one-200-page-book.

Mixing unrelated content. A collection should represent one coherent knowledge base. If you want a bot that knows both product docs and HR policies, you probably want two bots with two collections, not one bot with one collection mixing both.

Not testing retrieval. Just because your documents uploaded doesn’t mean the bot will find them. Use the Search tab to sanity-check that obvious queries return the chunks you expect. If they don’t, the bot won’t either.

Using documents where memory fits. Documents are for curated external knowledge. Memory is for agent self-notes. If you’re thinking “I want the agent to remember this”, that’s memory. If you’re thinking “I want the bot to answer from this handbook”, that’s documents.

Where to go next

• Bots — link a collection to a chatbot
• Memory & knowledge — the difference between memory and documents
• Memory browser — the memory counterpart