Embage: Build AI chatbots and browser voice agents

Build customer agents that answer from your documents, capture requests and act in connected apps.

embage.com
Embage screenshot

What is Embage, in plain language?

Embage is an AI chatbot builder for websites. It lets a business create a customer-facing assistant, give it company information and approved tools, then put a chat window on its site. A visitor can ask about a return, leave a sales enquiry or request an appointment. Depending on the tools you configure, the agent can look up information, create a record or complete an action.

It suits small support and sales teams, service businesses and agencies building assistants for clients. The platform also offers browser voice and Telegram chat, although their capabilities differ. It is currently in early access, so test the actual workflow before depending on it for important customer requests.

Answer customer questions from your own information

Import a web page, write an article or upload text-based PDF, Word, PowerPoint, Excel, CSV, EPUB, text or Markdown files. A knowledge sub-agent searches that material for the main agent, helping it use your policies rather than rely only on general model knowledge.

There are practical boundaries: scanned image-only PDFs and password-protected files cannot be read. A URL import fetches one page without following its links, and imported pages do not refresh automatically. Re-import changed pages and keep policies current. Search can return a weak match, so instruct the agent to admit when the material does not answer a question and offer a human handoff. The knowledge-base guide explains these conditions.

Turn a conversation into a useful business outcome

A Datastore sub-agent can collect the fields for a ticket, lead or feedback record while the visitor talks. Missing required fields or invalid values reject the save, allowing the agent to ask for corrections. The agent adds records; editing existing records remains a dashboard task.

Connect applications through sub-agents to do more: check stock or orders with Shopify, find meeting slots with Google Calendar, or prepare follow-up emails with Gmail. Test each helper and pin only the actions it should use, up to 15 per app. For emails requiring approval, allow draft actions rather than sending actions. Embage advertises more than 360 available app connections; availability is different from granting permission to every action.

After a conversation, a review provides a 0–100 score, sentiment, a summary, action items and knowledge gaps. Use these as operational leads for review and improvements, rather than proof that an answer or action was correct.

Build and check a first website agent

  1. Create an account and organisation, choosing a database region.
  2. Add a small set of current policies or product documents.
  3. Create a knowledge-base sub-agent, assign its knowledge tool and test it in the playground. Add separate helpers for records or connected apps as needed.
  4. Create the main agent, choose its use case and configure personality, helpers, review, model and widget style.
  5. Copy the floating-chat script or inline iframe from Embed into your website.
  6. Test a supported question, an unsupported question and a complete request. Check the reply, saved fields, app result and handoff, then review the conversation record.

The official quickstart provides the configuration sequence. If you restrict allowed domains, standalone share links stop working too.

Understand browser voice and Telegram before choosing a channel

Browser voice needs microphone permission and supports interruption. During a handoff, a teammate sees the transcript and types replies that are spoken to the visitor. It does not supply phone numbers or SIP connections.

Telegram supports chat, but its conversations do not currently run sub-agents or search the knowledge base. Do not assume a workflow tested on the website will work unchanged in Telegram. Public REST API access and outgoing webhooks are also not yet available.

Free plan, paid plans and separate model costs

Prices exclude taxes unless stated otherwise, and subscriptions renew until cancelled. Cancel before the trial ends to avoid being charged.

Checked on 3 November 2026 using the public USD monthly display:

  • Free: $0 platform plan with your own OpenRouter or Vercel AI Gateway key; 200 knowledge searches, 500 Datastore writes and 250 integration calls per month.
  • Starter: $19/month, 1,000 AI credits and 100 voice sessions per month.
  • Pro: $59/month, 3,800 AI credits and 500 voice sessions per month.
  • Enterprise: custom pricing and limits.

Own-key chat replies consume no Embage reply credits, but model usage is billed by your provider. Free does not include voice; voice counts by session, not minute. Paid plans offer a seven-day trial. Different resources have separate allowances: an upgrade does not necessarily raise every quota.

When an allowance runs out, that feature stops until renewal or an upgrade; unused allowances do not roll over. Review the current plan limits and subscription terms for taxes, renewal and cancellation before subscribing.

What customer data passes through the service?

Embage says it does not use your content or visitor conversations to train AI models. It stores transcripts and operational logs, but not voice recordings. Relevant content still reaches the selected AI providers under their terms. A database region choice therefore does not mean all processing stays in one country.

Content and conversations are retained while the account is active. Account or organisation deletion is available in settings, with possible legal retention exceptions. You retain ownership of added content and conversations, and must have rights to the material you supply. Review the privacy policy before collecting customer information.

Frequently asked questions

Is the free chatbot completely free to operate?

The platform plan is free within its quotas when you bring your own key. Your AI provider may charge for model usage, and knowledge searches, records and app calls remain limited.

Can I use a scanned PDF as the knowledge base?

No. The documented importer needs readable text and cannot process image-only scanned PDFs or password-protected files.

Can the voice agent answer telephone calls?

The documented voice channel runs in a browser. It is not a phone-number or SIP service, and requires microphone access.

Are customer conversations used to train AI?

Embage states that it does not train models on them. Selected model providers still receive relevant content, so their terms and your own privacy obligations also matter.

Official resources

FluentPaid
An AI assistant inside your Mac apps, with your choice of cloud or local models.
AI Assistants & ChatbotsAI Paraphrasing Tools · AI Workflow Automation Tools · AI Document Chat Tools +2Visit ↗
VaaniPaid
Build multilingual calling agents with business knowledge, tools and memory.
AI Voice AgentsAI Transcription Tools · AI Workflow Automation Tools · AI Sales Assistants +1Visit ↗