AI-powered analytics chat, built for production
Your customers ask questions of their data in plain language and get back real answers, as text or live charts, without leaving your product. You keep full control of the UI and UX.
Make analytics a reason to renew.
Your customers already ask ChatGPT anything they like, and expect every tool to answer as readily. Meeting that bar on your own data turns a reporting tab into a daily or weekly habit. Habits are hard to churn away from.
Fewer tickets, faster answers
Every question a customer answers for themselves is one that never becomes a support ticket or a data request for your team.
A daily or weekly habit
When the answer is one question away, people ask more often. Analytics stops being the tab nobody opens.
Without building it yourself
Matching that experience in-house means an orchestration layer, permissions and a chart pipeline. This is all of that, already built.
Conversational insights you can trust
Any product can put a language model behind a chat window. Keeping its answers grounded and its data access scoped is the harder problem. The AI Chatbot answers from the governed data models your dashboards already run on, authorised by the same security tokens.
Grounded in your models
Answers are retrieved from your own data models, the same governed semantic layer your dashboards run on.
Permissioned per user
Row-level security is enforced on every query, from the claims in the token you already issue.
Your UI, your control
Style the chat with your own design tokens, or build your own front end entirely.
Ask a question. Get more than an answer.
Everything your customers can do, grounded in the data and chart types you have made available.
Ask, and get a real answer
- Questions in plain language, against any dataset you have enabled.
- A text explanation, a live chart, or both.
- Thumbs up/down and an optional comment on every response.
Edit dashboards by asking
- Add a chart to the dashboard by describing it.
- Change an existing chart's data, measure or filters.
- Resize, reorder or remove charts without a drag.
Aware of what is on screen
- When it shares a token with a dashboard on the page, it reads that dashboard's charts, queries and layout, so anything it adds lands sized and positioned sensibly.
- With no dashboard open it still answers and previews the chart, asking "Shall I add this to your dashboard?" before anything changes.
One chat component. Your own model underneath.
Three pieces, cleanly separated: a chat interface in your product, an orchestration layer we host and secure, and the model you choose, called with your own key.
<em-ai-chat>, in your product
A web component for plain HTML, React, Vue, Angular or anything else. It authenticates with the same security tokens your dashboards already use.
AI Orchestrator
Hosted by Embeddable, US or EU. It resolves every request against your data models and enforces row-level security. Add your own system prompt so the assistant knows your domain and speaks in your product's voice.
Your LLM, your key
Anthropic, OpenAI, Google, OpenRouter, Mistral AI, Amazon Bedrock or Ollama. Saved once, stored encrypted, never sent to the client.
It runs on the same governed data your self-serve dashboard users already explore.
<script type="module" src="https://unpkg.com/@embeddable.com/ai-chat/dist/ai-chat/ai-chat.esm.js"></script>
<em-ai-chat
mode="inline"
orchestrator-url="wss://api.us.embeddable.com/ws"
embeddable-tokens="..."
></em-ai-chat>
Live in a few lines, no build step required
Or bring your own agent.
Would rather build the chat experience yourself? AI Endpoints exposes the same governed data and dashboard tools over a REST API. The reasoning stays with your agent, and the interface stays entirely yours.
A REST API for your agent
The /ai/v1 endpoints cover metadata, models, queries and full dashboard management. Every response is scoped automatically to that user's security context.
An MCP server, ready to run
A reference Model Context Protocol server, in Python with FastMCP, exposes the whole surface as tools any MCP-compatible agent can call.
The same governance, extended to conversation.
Letting your customers talk to their data should not mean loosening control over it.
One set of tokens
- It uses the security tokens you already mint. There is no separate chatbot credential to issue, rotate or leak, and row-level security applies to every question asked.
Read and write, controlled separately
- A read-only flag on the token decides whether the AI can change the dashboard, or only answer and preview. Set per token, so per user.
- The orchestrator rejects connections whose tokens belong to different workspaces.
Your key stays yours
- Your LLM provider key is stored encrypted, masked in API responses and never sent to the client.
Trusted by top SaaS teams building analytics into their products
From growth-stage SaaS teams to enterprise software platforms, Embeddable helps product teams launch customer-facing analytics faster without compromising on UX, flexibility, or control.
GDPR
SOC2-TYPE2
High Performer 2026
Highest User Adoption 2026
"It's a missing part of the ecosystem"
Director of Engineering, Asana
Start with the dashboards you already have.
Give your customers a reason to keep coming back, without taking on a second product to host, secure and maintain. Customer-facing analytics, built in code. Trusted at scale.
Same tokens as your dashboardsFull control of the UI
US and EU regions
Ask us