my agent doesnt actively use information in the knowledge base

[elfin pier]

AI-generated Answer for zqilk

Here are the most common, doc-backed reasons an Agent won’t use your knowledge base, and how to fix them:

• The documents aren’t actually attached to the Agent. In the dashboard, go to your Agent > Agent tab > Knowledge base > Add document, and confirm the document shows as a dependency of that Agent. (elevenlabs.io)
• RAG isn’t enabled (or indexing hasn’t finished). In the Agent’s Knowledge Base settings, toggle Use RAG. Each document must be indexed before retrieval works; large files can take a few minutes and you can check indexing status in the knowledge base list. (elevenlabs.io)
• The document is too small for RAG. Files under 500 bytes cannot be indexed and will only be injected into the prompt; they won’t be retrieved via RAG. (elevenlabs.io)
• You need the document always considered. Set its usage mode to Prompt (always in the system prompt) instead of Auto (retrieved only when relevant). Use Prompt sparingly to avoid context limits. (elevenlabs.io)
• You hit plan limits for RAG indexing. Check your tier’s total indexed document-size limits (e.g., 1MB Free, 20MB Creator, 100MB Pro, etc.). Exceeding these prevents effective retrieval. (elevenlabs.io)
• API configs don’t include the KB. If you configure via API, ensure documents are created and added under agent.prompt.knowledge_base, and that RAG is enabled in the agent’s prompt settings. (elevenlabs.io)
• FYI: The RAG option becomes available once the KB is large enough that retrieval is more efficient than direct prompt injection. (elevenlabs.io)

If you’ve verified all of the above and the Agent still ignores the KB, share your Agent ID, whether RAG is enabled, and the document indexing status with support so we can investigate. [e-mail]. (help.elevenlabs.io)

Sources:
Knowledge base dashboard | ElevenLabs Documentation
Retrieval-Augmented Generation | ElevenLabs Documentation
Retrieval-Augmented Generation | ElevenLabs Documentation
Retrieval-Augmented Generation | ElevenLabs Documentation
Retrieval-Augmented Generation | ElevenLabs Documentation
Knowledge base | ElevenLabs Documentation
Deploying enterprise knowledge to voice agents
How can I contact support? – ElevenLabs

[thorny finch]

my file is bigger than 500bytes, but it recommends me to not uising it becasue it might take extra and should disable it for faster response?

[elfin pier]

AI-generated Answer for zqilk

Short answer: If your document is larger than 500 bytes, it’s eligible for RAG indexing. The warning you’re seeing is about latency: enabling RAG adds roughly 500 ms per response. If you don’t need retrieval for this agent/flow, you can disable RAG for faster responses.

What the docs say

• RAG adds slight latency to agent responses (~500 ms). (elevenlabs.io)
• Documents smaller than 500 bytes can’t be indexed for RAG and are automatically injected into the prompt instead. Larger files (like yours) can be indexed. (elevenlabs.io)
• You can toggle “Use RAG” in the agent’s Knowledge Base settings. Disabling it removes the RAG overhead. (elevenlabs.io)
• For minimal overhead while keeping RAG on, keep documents in “Auto” usage mode (retrieved only when relevant). Avoid setting many docs to “Prompt,” which always injects them into context and can increase latency/cost. (elevenlabs.io)
• If you keep large docs, consider splitting them into smaller, focused pieces for better processing. (elevenlabs.io)

Practical guidance

• If latency is your top priority and the agent rarely needs KB grounding: disable RAG. (elevenlabs.io)
• If you need KB grounding: leave RAG on, keep usage mode = Auto, and verify the document finished indexing (large docs can take a few minutes). (elevenlabs.io)

If you’d like, tell me your agent’s KB settings (RAG on/off, usage mode) and I can point to the exact control to change. (elevenlabs.io)

Sources:
Retrieval-Augmented Generation | ElevenLabs Documentation
Retrieval-Augmented Generation | ElevenLabs Documentation
Retrieval-Augmented Generation | 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

[thorny finch]

if im using api to create a KB using file, and add it to the agent, how can i set is such that its always include in the prompt so the agent is aware of that? and should i do this?

[elfin pier]

AI-generated Answer for zqilk

Short answer

• Yes—you can force a KB document to always be included in the system prompt by setting its usage_mode to "prompt" in the agent’s prompt.knowledge_base configuration. Use this sparingly; ElevenLabs recommends leaving most documents in the default “Auto” (retrieve-when-relevant) mode to avoid context bloat. Documents under 500 bytes are automatically used in the prompt anyway and cannot be indexed for RAG. (elevenlabs.io)

How to do it via API

• Create the document from a file, then attach it to the agent with usage_mode set to "prompt":
  • Create file doc: conversational_ai.knowledge_base.documents.create_from_file(...)
  • Update agent: conversational_ai.agents.update(... conversation_config.agent.prompt.knowledge_base = [{ id: <doc_id>, type: "file", name: <name>, usage_mode: "prompt" }, ...])
    The base pattern for creating a KB doc and adding it to an agent is shown in the Knowledge base docs; the RAG page documents the usage_mode flag and shows how to modify it programmatically. (elevenlabs.io)

Should you do this?

• Only for short, critical content that must always condition the model (e.g., safety or policy snippets). Setting many documents to “Prompt” can exceed context limits; prefer “Auto” with RAG for large/occasional references. Note that enabling RAG adds slight latency (about 500 ms) to responses. (elevenlabs.io)

References

• Knowledge base: creation and attaching to agents. (elevenlabs.io)
• RAG: enabling, usage modes (“Auto” vs “Prompt”), <500-byte behavior, and latency note. (elevenlabs.io)

Sources:
Retrieval-Augmented Generation | ElevenLabs Documentation
Knowledge base | ElevenLabs Documentation
Retrieval-Augmented Generation | ElevenLabs Documentation
Knowledge base | ElevenLabs Documentation
Retrieval-Augmented Generation | ElevenLabs Documentation

[thorny finch]

when im debugging in editor locally, and for this data {{ $('Webhook').item.json.body.candidate.cv_file.content }}

its definitely defined which is a string, in the node where i try to do a manual mapping, it was green even before i execute the steps, but when i execute all the steps, it becomes red and the expression become undefined

[elfin pier]

AI-generated Answer for zqilk

Short answer: I can’t verify that expression behavior in your local editor from ElevenLabs’ documentation. It looks specific to your workflow tool (e.g., n8n) rather than the ElevenLabs API.

What I can confirm from our docs and how to debug from the ElevenLabs side:

• ElevenLabs webhook payload shape: top-level fields are type, event_timestamp, and data. Your mappings should read values from data (for example, data.transcript, data.agent_id). If your expression points to other paths (like body.candidate.cv_file.content), it won’t resolve for ElevenLabs webhooks. Please inspect the raw payload your endpoint receives and map from data accordingly. (elevenlabs.io)
• If you’re wiring this into n8n, verify the webhook payload format matches what your n8n node expects and check your agent logs when debugging. The n8n integration page’s troubleshooting section calls this out explicitly. (elevenlabs.io)
• Implement (or temporarily enable) logging in your webhook handler to print the exact JSON received before and after signature verification using the SDK’s construct_event/constructEvent helper, then map fields from event.data. This ensures you’re mapping from the parsed ElevenLabs event structure rather than a tool-specific wrapper. (elevenlabs.io)

If you still can’t reconcile the payload your node sees with the ElevenLabs event structure, please contact ElevenLabs Support and include the raw webhook payload, timestamps, and any request IDs so we can check delivery on our side. (help.elevenlabs.io)

Sources:
Webhooks | ElevenLabs Documentation
Connect n8n to ElevenLabs Conversational AI Voice Agents
Webhooks | ElevenLabs Documentation
How can I contact support? – ElevenLabs

[thorny finch]

i think part of the previous are just undefined?