#Knowledge base for agent states file up to 20mb, I uploaded 7mb json. Now it forces me to do some R

1 messages · Page 1 of 1 (latest)

tacit doveBOT
#
AI-generated Answer for osudcreator

Short answer

  • No, you don’t need to enable RAG to use a Knowledge Base. Agents can use Knowledge Base documents without RAG; non‑enterprise accounts can attach up to 20 MB (or 300k characters) of Knowledge Base content. (elevenlabs.io)
  • Your “RAG storage limit exceeded” error is because RAG has its own, smaller per‑workspace indexing quota by plan (separate from the 20 MB Knowledge Base cap). On Free the RAG quota is 1 MB, so a 3.5 MB index attempt will fail. (elevenlabs.io)

What the limits are

  • Knowledge Base (non‑enterprise): 20 MB total (or 300k characters). (elevenlabs.io)
  • RAG indexing (per workspace):
    • Free: 1 MB
    • Starter: 2 MB
    • Creator: 20 MB
    • Pro: 100 MB
    • Scale: 500 MB
    • Business: 1 GB
    • Enterprise: 1 GB+ (by agreement)
      Notes: Limits apply to the total original file size indexed for RAG; sub‑500‑byte docs can’t be RAG‑indexed and are used in the prompt instead. (elevenlabs.io)

How to proceed

  • If you don’t want RAG: turn off “Use RAG” in the agent’s Knowledge Base settings; the agent will still use the document via prompt context (subject to model/context limits). (elevenlabs.io)
  • If you do want RAG with a 7 MB JSON:
    • Upgrade to a tier whose RAG quota covers 7 MB (Creator or higher), or
    • Reduce/split the file into smaller, focused documents and re‑attach. (elevenlabs.io)

About the “See changes” altering true/false

  • This behavior isn’t documented in the Help Center or Docs. Please share screenshots and the exact steps with Support via the contact form so the team can investigate. (help.elevenlabs.io)

References

  • Knowledge Base limits and usage: Knowledge base; What can I create with ElevenLabs Agents. (elevenlabs.io)
  • RAG overview, toggle location, and per‑plan RAG storage limits: Retrieval‑Augmented Generation. (elevenlabs.io)
  • Contact Support: How can I contact support? (help.elevenlabs.io)

Sources:
Knowledge base | ElevenLabs Documentation
Retrieval-Augmented Generation | ElevenLabs Documentation
Knowledge base | ElevenLabs Documentation
Retrieval-Augmented Generation | ElevenLabs Documentation
Retrieval-Augmented Generation | ElevenLabs Documentation
Retrieval-Augmented Generation | ElevenLabs Documentation
How can I contact support? – ElevenLabs
Knowledge base | ElevenLabs Documentation
Retrieval-Augmented Generation | ElevenLabs Documentation
How can I contact support? – ElevenLabs

obsidian grotto
#

how do i test if agent see information in knowlegde base, he seems to browse also internet so its hard to tell

tacit doveBOT
#
AI-generated Answer for osudcreator

Short answer: use the built‑in testing/simulation APIs and/or force a document into the prompt while disabling external tools.

Practical ways to verify your agent is using your knowledge base:

  • Run a simulation or a test and inspect rag_retrieval_info

    • Simulate a conversation against your agent and check each turn in the response for rag_retrieval_info. If retrieval happened, you’ll see chunks with document_id you can match to your KB. (elevenlabs.io)
    • If you use the Tests feature, fetch the test invocation; the response explicitly returns rag_retrieval_info.chunks[ ].document_id and chunk_id for the turn being evaluated. (elevenlabs.io)

  • Temporarily prevent any “internet” tool usage

    • Agents only call out to the world via Tools (client, server, or MCP). If you remove tools or set MCP tools to “Always Ask” or Disabled, the agent cannot make external calls while you test KB grounding. (elevenlabs.io)

  • Force the doc into context (to prove it’s seen)

    • In the Knowledge Base/RAG settings, set the relevant document’s usage mode to Prompt so it is always included in the system prompt. Ask a question that quotes that document; the agent should answer correctly without retrieval. Use this sparingly because it consumes context. (elevenlabs.io)

  • Confirm RAG index readiness (if you rely on retrieval)

    • Check your document’s RAG index status and used_bytes via the Get RAG index endpoint before testing, so you’re sure the doc is retrievable. (elevenlabs.io)

Notes

  • RAG usage limits exist per plan; if you hit “RAG storage limit exceeded,” see the plan limits and consider switching specific docs to Prompt mode for testing. (elevenlabs.io)
obsidian grotto
#

I cannot save knowledge bas, cannot use it in worflow it all states "Fix knowledge base errors to proceed" it is impossible to use without that ? I tried chat with agent and he do not know abot data in that knowledge base

tacit doveBOT
#
AI-generated Answer for osudcreator

Short answer: the Help Center doesn’t document the “Fix knowledge base errors to proceed” message or any requirement to resolve it before using Agents/Workflows. I can’t verify why your save is blocked from official docs.

What the docs do confirm:

  • Knowledge Base is optional “additional information” for an Agent; you can create and use an Agent without it. Non‑enterprise users have up to 20 MB total Knowledge Base storage. (help.elevenlabs.io)
  • RAG is optional. It can’t be enabled for documents smaller than 500 bytes. When enabled, it increases how much Knowledge Base material can be assigned to an Agent, with the total limit depending on your subscription plan (exact per‑plan RAG limits are not listed in the Help Center). (help.elevenlabs.io)

Given there’s no published guidance on that specific error and you’re seeing “RAG storage limit exceeded,” the safest next step is to contact Support with screenshots and the file you’re trying to add so they can check your account’s limits and the ingestion pipeline. (help.elevenlabs.io)

Related references:

  • What can I create with ElevenLabs Agents (formerly Conversational AI) — Knowledge Base and RAG limits overview. (help.elevenlabs.io)
  • How can I contact support? — official contact form. (help.elevenlabs.io)

If you want, share the exact Help Center article(s) you’d like me to check against or I can keep looking for any newly published guidance and update you.

Sources:
What can I create with ElevenLabs Agents (formerly Conversational AI) – ElevenLabs
What can I create with ElevenLabs Agents (formerly Conversational AI) – ElevenLabs
How can I contact support? – ElevenLabs
What can I create with ElevenLabs Agents (formerly Conversational AI) – ElevenLabs
How can I contact support? – ElevenLabs

obsidian grotto
#

what is even rag

tacit doveBOT
#
AI-generated Answer for osudcreator

RAG (Retrieval‑Augmented Generation) is the feature that indexes your agent’s knowledge base and, at runtime, retrieves only the most relevant chunks to answer a user’s question—so the LLM doesn’t have to load entire documents into its prompt. This enables larger, more accurate, source‑grounded answers and typically adds about 500 ms of latency. (elevenlabs.io)

How it works in ElevenLabs:

  • You can toggle Use RAG in an agent’s Knowledge Base settings; documents are then indexed automatically, and you can configure the embedding model, max retrieved chunks, and vector distance. Very small files (<500 bytes) aren’t indexed and are used in the prompt instead. (elevenlabs.io)
  • RAG has plan‑based indexing limits (total original file size indexed for RAG): Starter 2 MB, Creator 20 MB, Pro 100 MB, Scale 500 MB, Business 1 GB, Enterprise 1 GB+ (higher by agreement). (elevenlabs.io)
  • RAG is specifically meant to let agents use larger knowledge bases than would fit in a prompt; enabling it increases how much info your agent can draw from. (help.elevenlabs.io)

In short: RAG is the retrieval layer that lets your agent ground answers in your indexed documents instead of relying only on prompt context. (elevenlabs.io)

Sources:
Retrieval-Augmented Generation | ElevenLabs Documentation
Retrieval-Augmented Generation | ElevenLabs Documentation
Retrieval-Augmented Generation | ElevenLabs Documentation
What can I create with ElevenLabs Agents (formerly Conversational AI) – ElevenLabs
Retrieval-Augmented Generation | ElevenLabs Documentation

obsidian grotto
#

even index size is 1/7mb, why allowingt upload 20mb files when it support only files under 1mb