Choose a new password
· plan:
· id:
Install
Paste this into your website's HTML, just before </body>.
Branding
PNG, JPG, SVG, WebP, GIF, ICO — max 500 KB.
Tight-cropped square image (no transparent padding around the mark) fills the circle edge-to-edge. If your file has built-in padding around the logo, the circle will show that same padding — please re-crop before uploading.
Tight-cropped square image (no transparent padding around the mark) fills the circle edge-to-edge. If your file has built-in padding around the logo, the circle will show that same padding — please re-crop before uploading.
Or paste an external URL:
The small strip at the bottom of the chat panel.
Must start with http:// or https://. Leave empty for plain text. Opens in a new tab.
Widget appearance
Between 280 and 800. Auto-shrinks on narrow screens.
Between 380 and 1200. Auto-shrinks on short screens.
Diameter of the floating chat icon. 36–120.
Also applied to every avatar circle in the widget. Empty = use accent color.
Distance from the left/right edge of the page. 0–200.
Distance from the top/bottom edge of the page. 0–200.
Allowed origins
Exact origin strings (no trailing slash). Only these sites can load the widget.
No origins yet. Type a URL above and click Add (or press Enter).
Cloudflare Turnstile
Free at dash.cloudflare.com → Turnstile. Add each allowed origin's hostname to your widget's hostname list.
Never displayed after saving. Only sent again to overwrite.
AI Platform Integration
Choose how this bot talks to an AI platform. General uses a single provider's native conversation API for everything.
Uses each platform's native chat API. OpenAI uses the Responses API; Anthropic uses Messages; Gemini uses GenerateContent.
Your existing independent provider settings remain unchanged and are available below.
Advanced mode — separate providers for chat, embeddings and voice, with your own crawled content as context — is coming soon.
General provider
Encrypted at rest. Never displayed.
General mode calls the platform's chat API directly with no local retrieval / reranking. Knowledge comes from the OpenAI vector store IDs below.
Comma-separated
vs_... IDs from your OpenAI dashboard (Storage > Vector Stores). When set, OpenAI's Responses API does hosted retrieval via the file_search tool — no local indexing needed. Leave blank for plain LLM chat without any knowledge base.Voice in General mode is only supported for OpenAI.
Uses the visitor's browser to transcribe speech into the input box. No API cost.
Streams audio via WebRTC directly to OpenAI Realtime. Charged per audio token.
Chat model
Powers the text conversation. Provider & key set here only affects chat.
Encrypted at rest. Never displayed.
Voice
Browser opens WebRTC directly to the provider — audio never touches our server.
Uses the browser's SpeechRecognition API. Free, but only supported in Chromium/Safari.
Streams audio via WebRTC directly to the provider. Charged per token.
Embeddings (knowledge search)
Used when indexing sources and when answering with RAG context.
Changing after indexing requires re-indexing all sources.
Runs on our own server. No key, no model choice, no per-query cost.
System prompt
Instructions the bot always sees before answering. Keep it focused.
Limits
Warning threshold. No auto-shutoff.
Select a conversation to view messages.
visitor:
Last 30 days
Loading…
Total spend
$
Total messages
Tokens in
Tokens out
Daily breakdown
No usage yet.
Crawl a website
BFS crawl from a start URL. Same origin only. Respects robots.txt.
Ingest a sitemap
Fetches sitemap.xml (index-safe) and indexes only listed pages.
Upload a document
Supported: .txt, .md, .pdf. Max 10 MB.
Sources
Embedding dim: · provider: · model:
No sources yet. Add one above.
| Type | Source | Status | Pages | Chunks | Indexed | |
|---|---|---|---|---|---|---|
Auto-refreshing every 3s while jobs are running…
Join an organization
Loading…
You've been invited to join as .
This invitation was sent to ,
but you're signed in as .
Sign out and register/sign in with the invited email.