AI chatbots, agents and workflows
Build a chatbot on your own content, test it against the exact configuration you are about to publish, then embed it on your site or call it from your application. One platform for the whole loop.
Support Assistant
Active · 2 collectiones
Platform
Each piece works on its own and they compose: a chatbot can answer from a collection, an agent can call a workflow, and every conversation lands in the same inbox and CRM.
Write the instructions, attach a collection, try it in the playground, then publish. Replies stream token by token and cite the sources they used.
Give an agent a tool set, a memory window and a structured output shape. Every tool request is shown with its decision, and nothing runs without approval.
Build a workflow from typed nodes, validate it before it runs, and read every run as a step-by-step timeline with its inputs, outputs and failures.
Add text, a URL or a file. Watch it move through ingestion, chunking, embedding and indexing, then test retrieval before a chatbot ever uses it.
A workspace API key and one endpoint. The same streaming contract the widget uses, so your app, your backend and your mobile client all behave alike.
Paste a script tag, allow your domains, and the widget appears in your brand colour. No credentials ever reach the page.
Widget, API, playground and agent runs land in the same place, with AI summaries and a reply box when a person should step in.
Contacts, stages, tags, notes and a merged activity timeline, linked to the conversations they came from.
Conversations, messages, resolution rate and token spend over time, compared against the period before.
Outgoing webhooks, Slack and SMTP today, with credentials encrypted at rest and never returned to the browser.
Deployment
Put the assistant on your website, or call it from whatever you already run. Both paths hit the same chatbot, the same knowledge and the same streaming response.
For a marketing site, a help centre or a product page. Paste the script, list the domains that may use it, and the widget matches your brand colour. Nothing secret ever reaches the page.
<script
src="https://app.example.com/embed/widget.js"
data-chatbot="cb_YOUR_EMBED_KEY"
async
></script>For your own frontend, a backend service, a mobile app or an internal tool. Authenticate with a workspace API key; the key identifies the tenant, so your code never names one.
const response = await fetch(
"https://app.example.com/api/v1/public/chat",
{
method: "POST",
headers: {
Authorization: `Bearer ${process.env.DOT_API_KEY}`,
"Content-Type": "application/json",
},
body: JSON.stringify({
chatbotId: "YOUR_CHATBOT_ID",
messages: [{ role: "user", content: "Do you ship to the EU?" }],
}),
},
);
// Server-sent events: the same stream the widget consumes.
for await (const event of readChatStream(response)) {
if (event.type === "text-delta") process.stdout.write(event.delta);
}How it works
The same five steps whether you are shipping a support bot on a marketing site or an agent inside your own product.
Create a chatbot or an agent, write its instructions and pick a model. No infrastructure to stand up.
Attach a collection so answers come from your content, with citations back to the source.
Run it in the playground against the exact configuration you are about to publish.
Embed it on your site or call it from your own application with an API key. Same assistant, same contract.
Read conversations, follow up in the CRM, and watch usage and cost in analytics.
Create a workspace, point it at your content, and see a grounded answer in the playground before you publish anything. No credit card, no sales call.