Skip to main content
The Knowledge Base is a collection of documents that Sniko indexes with Retrieval-Augmented Generation (RAG). When an agent receives a user query, ElevenLabs retrieves the most relevant chunks from the assigned knowledge bases and injects them into the LLM context — allowing the agent to answer accurately from your content without hallucinating.
Documents are stored and indexed by ElevenLabs. Sniko applies customer isolation so each account only sees and manages its own documents.

​
Adding documents

Navigate to Knowledge Base in the sidebar. Three ingestion methods are available:
Provide a publicly accessible URL. ElevenLabs fetches and parses the page content automatically.
  • The document name defaults to the URL hostname (e.g. docs.example.com) if you leave the name field blank.
  • Maximum URL length: 2,048 characters.
1

Click 'Add Document' → 'From URL'

Enter the full URL including the protocol (https://).
2

Optionally set a display name

If omitted, Sniko derives the name from the URL hostname.
3

Submit

Sniko posts to the ElevenLabs /convai/knowledge-base/url endpoint. On success, the new document appears in your list.

​
Document management

From the Knowledge Base index page you can:

Rename

Update a document’s display name. The change is patched to ElevenLabs via PATCH /convai/knowledge-base/{id}.

View content

Preview the raw extracted text that ElevenLabs stored for the document.

Delete

Permanently removes the document and all its RAG indexes. If the document is assigned to agents, deletion still proceeds (those agents lose the knowledge source).
You can also view which agents depend on a given document to understand the blast radius before deleting.

​
RAG indexing

RAG (Retrieval-Augmented Generation) indexing splits a document into semantically meaningful chunks and encodes them as vector embeddings. At conversation time, the agent’s query is encoded and the nearest chunks are retrieved and inserted into the prompt.

​
Computing an index

After a document is uploaded you can trigger indexing on demand. Two embedding models are available:
ModelBest for
e5_mistral_7b_instructEnglish-dominant content, high accuracy
multilingual_e5_large_instructMulti-language content
1

Open the document

Click the document name in the Knowledge Base list.
2

Select an embedding model

Choose e5_mistral_7b_instruct for English content or multilingual_e5_large_instruct for multilingual content.
3

Click Compute Index

Sniko posts to /convai/knowledge-base/{id}/rag-index. Indexing runs asynchronously on ElevenLabs infrastructure.

​
RAG storage quota

The Knowledge Base index page displays a storage usage bar showing how much of your RAG quota is consumed. The default limit is 21 MB of indexed vector storage. Your subscription plan may grant a higher limit. 
RAG Storage: [████████░░░░░░░░░░░░] 8.4 MB / 21 MB used
You can query the live quota at any time from the RAG index overview endpoint.
Deleting a RAG index frees storage but the document itself remains. You can re-index the document later if needed.

​
Assigning knowledge bases to agents

A single agent can have multiple knowledge bases. A single knowledge base can be shared across multiple agents. During agent creation: Select one or more knowledge bases in the Knowledge Base step of the creation wizard. Sniko validates that each selected document exists in ElevenLabs before completing agent creation. On an existing agent: Open the agent’s configuration page and go to the Knowledge Base tab. Add or remove documents and save. Sniko sends the updated knowledge_base array inside conversation_config.agent.prompt to the ElevenLabs PATCH endpoint.
If you share a knowledge base across multiple agents, updating the document content (or re-indexing it) automatically improves responses for all agents that use it.

​
How assignments are stored

Internally Sniko tracks agent–knowledge base relationships in the agent_knowledge_bases database table (columns: agent_id, knowledge_base_id, user_id). This allows the platform to answer queries such as “which agents use document X?” without a round-trip to the ElevenLabs API.

Build docs developers (and LLMs) love

Get started for freeTalk to us