#I tested an agent in the ElevenLabs Dashboard. Can you explain the difference between the Call cost
1 messages · Page 1 of 1 (latest)
- Call cost: What ElevenLabs charges for running the call itself (real‑time orchestration and audio pipeline). It’s billed by connection duration and discounted by 95% during silences longer than 10 seconds. Calls made in Setup & Prompt Testing in the dashboard are billed at half cost. LLM charges are not included in this figure. (elevenlabs.io)
- Credits (LLM): The number of ElevenLabs credits deducted to cover the LLM spend used by your agent. ElevenLabs converts the LLM’s dollar cost into credits at a plan‑specific rate (e.g., Business: 8,333 credits per $1 LLM cost). (help.elevenlabs.io)
- LLM cost: The underlying dollar amount attributable to the language model (based on tokens), shown in the dashboard. Model pricing is denominated in USD per 1M tokens and varies by provider/model; ElevenLabs passes these LLM costs through separately from the call cost. (elevenlabs.io)
Sources:
Introduction - Conversational voice AI agents | ElevenLabs Documentation
How much does ElevenLabs Agents (formerly Conversational AI) cost? – ElevenLabs
Models | ElevenLabs Documentation
Can you explain what the Allow list is (in security section) when setting up an agent?
The Allow list is a security control that restricts which web origins can connect to your agent. You add approved hostnames (e.g., example.com, app.example.com, localhost:3000); when a client tries to connect, ElevenLabs checks the request’s origin against this list and rejects connections from non‑listed domains. (elevenlabs.io)
Key details:
- It’s configured in an agent’s authentication settings (Dashboard or API). (elevenlabs.io)
- Exact hostname matching is used; subdomains must be added separately (no wildcard). (elevenlabs.io)
- You can specify up to 10 hostnames. (elevenlabs.io)
- For stronger security, you can combine an allowlist with signed URLs so requests must come from an approved domain and include a valid signed URL. (elevenlabs.io)
Sources:
Agent authentication | ElevenLabs Documentation
Agent authentication | ElevenLabs Documentation
Agent authentication | ElevenLabs Documentation
Agent authentication | ElevenLabs Documentation
Agent authentication | ElevenLabs Documentation
OK, so this is the website where the agent is used?
Yes. The Allow list is a list of web domains/hostnames (origins) that are permitted to embed or connect to your agent from the browser (e.g., via the widget/JS SDK). Only requests whose Origin matches an allowed hostname will be accepted; others are rejected. You can add entries like example.com, app.example.com, or localhost:3000 (up to 10). For stronger security, you can also require signed URLs in addition to the allow list. (elevenlabs.io)
Can an agent be integrated into a scrom package hosted on a LMS?
Short answer: there’s no ElevenLabs doc that mentions SCORM or LMSs specifically. However, Agents can be embedded on “any website” via the HTML widget. If your LMS/SCORM package lets you add custom HTML/JS, you can paste the embed snippet and it will work, provided you meet the documented requirements:
- Use the widget embed code in your page content. (elevenlabs.io)
- Widgets currently require a public agent with authentication disabled. (elevenlabs.io)
- Add the LMS host/domain to the agent’s Allowlist (Security tab) to restrict usage to that site. (elevenlabs.io)
What’s missing: ElevenLabs does not publish SCORM-specific instructions; the above is based on the official “embed on any website” guidance and the documented allowlist/auth rules. (elevenlabs.io)
Sources:
Widget customization | ElevenLabs Documentation
Widget customization | ElevenLabs Documentation
Widget customization | ElevenLabs Documentation
Widget customization | ElevenLabs Documentation
What would happen if no material was loaded into the knowledge base? How would the agent work?
Short answer: the agent still works, but it won’t be grounded in your own content.
- With no knowledge-base documents, the agent answers using the selected LLM’s base knowledge plus whatever you define in the System Prompt/behavior settings. Tools and workflows you’ve connected still function. (help.elevenlabs.io)
- Retrieval-Augmented Generation (RAG) cannot be used without at least one knowledge-base document, so the agent won’t retrieve or cite your materials. RAG exists to ground answers in your content and reduce hallucinations. (elevenlabs.io)
- If you need domain-specific, accurate responses from your own docs/FAQs, add them to the Knowledge Base (optional) and enable RAG. (help.elevenlabs.io)
Sources:
What can I create with ElevenLabs Agents (formerly Conversational AI) – ElevenLabs
Retrieval-Augmented Generation | ElevenLabs Documentation
What can I create with ElevenLabs Agents (formerly Conversational AI) – ElevenLabs
How is the agen embedded into a webpage
Here’s the supported way to embed an ElevenLabs Agent on a website:
- Make the agent public: In your agent’s Advanced settings, disable authentication. The current widget requires a public agent. (elevenlabs.io)
- Paste the embed in your site’s body:
<elevenlabs-convai agent-id="YOUR_AGENT_ID"></elevenlabs-convai>
<script src="https://unpkg.com/@elevenlabs/convai-widget-embed" async type="text/javascript"></script>
Place this in your main HTML (e.g., index.html) for site‑wide availability. (elevenlabs.io) - Secure by domain: Add your site’s hostnames to the agent Allowlist (Security tab) so only those domains can load the widget. (elevenlabs.io)
- Optional configuration in HTML attributes: You can set things like variant, language, UI text, colors, and markdown link behavior directly on the <elevenlabs-convai> tag. (elevenlabs.io)
Notes:
- For no‑code platforms, see the Webflow guide which uses the same snippet. (elevenlabs.io)
- If you need deeper control (framework components or direct sockets), use the SDKs or the WebSocket API instead of the widget. (elevenlabs.io)
Sources:
Widget customization | ElevenLabs Documentation
Widget customization | ElevenLabs Documentation
Widget customization | ElevenLabs Documentation
Widget customization | ElevenLabs Documentation
Agents Platform in Webflow | ElevenLabs Documentation
Agents Platform | ElevenLabs Documentation
Can an IP address be put in the Allow list for an Agent rather than a URL?
Short answer: The documentation only supports allowlisting by hostname (domain or subdomain, optional port), with exact matching. It does not document allowing raw IP addresses or full URLs. If you need to restrict access, use a hostname in the allowlist and/or signed URLs. (elevenlabs.io)
"We strongly recommend setting up an allowlist when using overrides." What does 'overrides' mean?
Overrides are per-conversation runtime settings that let you replace an agent’s default configuration. When enabled (per field, in the agent’s Security tab), you can supply values that completely override defaults such as the system prompt, first message, language, LLM, voice/tts parameters (voice_id, stability, speed, similarity), and text-only mode. Overrides are disabled by default; sending an override for a field that hasn’t been enabled will return an error.
Because overrides can change core behavior, ElevenLabs recommends restricting which sites can connect and supply them by configuring an Allowlist so only approved hostnames can use your agent (including the web widget).
Could a localhost be used in the Allow list?
Yes. Agents support allowlists of hostnames, and the docs explicitly show adding localhost with a port (e.g., localhost:3000) as an allowed entry. (elevenlabs.io)
Notes:
- Allowlists do exact hostname matching; add each subdomain separately. (elevenlabs.io)
- You can specify up to 10 hostnames in the allowlist. (elevenlabs.io)
- When embedding the web widget, define your allowed domains in the agent’s Allowlist (Security tab). (elevenlabs.io)
Sources:
Agent authentication | ElevenLabs Documentation
Agent authentication | ElevenLabs Documentation
Agent authentication | ElevenLabs Documentation
Widget customization | ElevenLabs Documentation
Where do I find my API key?
- Dashboard: Log in, then go to Developers in the left sidebar → API Keys. Create or copy your key there. Note: you can only view the full key once; after creation only the name and last four characters are visible. (help.elevenlabs.io)
- Workspaces (service accounts): If you’re using a multi-seat workspace, open the profile menu → Workspace settings → Service Accounts to create/manage API keys for a service account. (elevenlabs.io)
- Keep your key secret; don’t share it or expose it in client-side code. (help.elevenlabs.io)
Sources:
How do I authorize myself using an API key? – ElevenLabs
Service Accounts and API Keys | ElevenLabs Documentation
How do I authorize myself using an API key? – ElevenLabs
Does the agent stick strictly to the Knowledge Base? What if I asked the agent the time?
Short answer: not unless you configure it that way.
- By default, a Knowledge Base augments the agent’s underlying LLM; it doesn’t force the model to answer only from those documents. ElevenLabs describes the KB as additional, domain‑specific information “on top of the base understanding of the LLM,” and as a way to “go beyond its pre‑trained data.” (help.elevenlabs.io)
- You can make an agent stick to the KB (or a specific scope) via your system prompt and guardrails. ElevenLabs’ own docs agent is prompted to “only use [the provided documentation] to answer questions about ElevenLabs,” with guardrails to stay on ElevenLabs topics. (elevenlabs.io)
- RAG can be enabled to retrieve and ground responses in the KB, reducing off‑KB hallucinations, but it’s a retrieval/grounding feature—not a hard restriction by itself. (elevenlabs.io)
If you ask “What time is it?”:
- The platform doesn’t list a built‑in “current time” system tool. To answer with real‑time accuracy, add a client or server tool that returns the current time; tools are the mechanism for accessing external/real‑time data. Alternatively, prompt the agent to decline out‑of‑scope questions. (elevenlabs.io)
Sources:
What can I create with ElevenLabs Agents (formerly Conversational AI) – ElevenLabs
Building an effective Voice Agent for our own docs
Retrieval-Augmented Generation | ElevenLabs Documentation
Tools | ElevenLabs Documentation
I added lcalhost:3000 to the allow list but get this message: Hostname must consist of a domain and an optional port
It’s a typo: you added lcalhost:3000. The allowlist expects a hostname (exact match) and optionally a port. Use localhost:3000 instead. The Agent Authentication docs show localhost:3000 as a valid example and note that allowlists match hostnames exactly. (elevenlabs.io)
I have copied and pasted in localhost:3000 and it is still showing the error 'Hostname must consist of a domain and an optional port'.