Support that never sleeps and answers only from what you gave it.

Answers from your own pages. Free to start, no card.

Built from the knowledge you give it, answering visitors around the clock with no one watching it.

Chat Studio draws only on the knowledge you hand it — your pages and documents, with the data Minimal already serves joining as a second channel in December 2026 — never the model’s general knowledge. Ten personas at launch set the voice and the boundary: support, sales, docs, onboarding, HR and more.

Start free Schedule a Personal Session

persona → knowledge → publish → one line on your site

One assistant, start to finish

A store’s help desk, built from its shipping page.

Three steps, in the order the studio walks you through them. The example the product ships with: a Support Agent answering shipping and returns questions.

Support Agent: patient, concrete, admits gaps. It answers product questions, walks through fixes, and knows when to send someone to a human. Tone layers on top — friendly & concise, precise & technical, playful — without changing what it refuses.

The persona step: persona cards for Start blank, Support Agent, Sales Qualifier, Docs Assistant, Onboarding Guide and HR & Policy Desk, tone chips underneath, and a live preview of what the visitor sees beside them — Start blank opens Advanced with your own instructions. Every persona shares one hard rule that editing cannot remove: answer only from what you gave it, never the model’s general knowledge.

Paste a site’s address and tick “Read the whole site”: Minimal follows the sitemap, llms.txt and links on the same domain, up to 200 pages and the first 50,000 words of each. One page on its own gives its first 2,000 words; uploads take .txt, .md, .pdf and .docx up to 20MB each; pasted text works too.

The knowledge step after reading a whole site from one address: 22 sources attached, a note that pages are re-read every 7 days, and each source listed with its chunk count and age — Every page becomes its own source, and pages are re-read automatically every 7 days. Uploaded files and pasted text don’t change on their own, so they are never re-read. A source that indexed to zero chunks shows no badge at all — absence of a count is the signal.

Publishing gives you two things: a public link on your workspace domain, and an embed snippet — one line, on any page you control, whatever it’s built on: WordPress, Shopify, Webflow, or hand-rolled HTML. Nothing is visitor-reachable until you publish. Test a real question in the preview first, not in the abstract.

The publish step: an accent colour for the send button, the provider and the model the assistant runs on, and the visitor preview beside them — Live, with the link and embed code ready to copy — an inline panel is the alternative to the floating widget. The accent colour applies to the send button, not full-page branding. The banner is the studio saying a provider key is missing *before* a visitor finds out.

Two channels of knowledge, and nothing else.

Every answer traces to a source you added. There are two kinds.

Exclusive knowledge — what you hand it.

Website pages, uploaded documents, pasted text. Read when added, listed as sources, refreshed one by one on your say-so.

Minimal-served data — what already runs. Public December 2026

The tables and APIs the layer serves for your agents, apps and dashboards, answered over live — no copy made, and the same permissioning applies. In private use today; opens to every Chat Studio account in December 2026.

What it never says. Anything outside those sources. When the answer is not in the knowledge, it says so — and with contact capture on, asks where to send the follow-up instead of guessing.

What stays yours. Authentication for your users, third-party integrations, and any action beyond answering — refunds, bookings, writes to your systems. An assistant that must act on your data is an Agent Studio job.

Where the knowledge and conversations live: see Security and Transparency.

Bring your own key

Your model bill, at provider prices.

Assistants answer on keys you bring — Claude, OpenAI, Gemini, or the open models through OpenCode GO. Model usage lands on your own provider account at the provider’s price, so the AI line is one you already control. Every assistant runs on a key you add.

Chat Studio settings: Bring Your Own Key slots for Claude, OpenAI, Gemini, OpenCode GO and AI credits, and an embeddings provider choice underneath — Keys are sealed server-side and never shown again after saving — a visitor’s message can arrive at any hour with no browser open, so the key lives where the assistant runs. Add a second key and it takes over mid-answer if the first is rejected. The bottom panel moves every assistant onto one provider and model in one action; knowledge, persona and public links stay exactly as they were.

Channels

Alerts go where you already look.

Channels are for you, not visitors — a broken key, a bad rating, a knowledge source that stopped answering. Slack, Microsoft Teams, Google Chat, Mattermost / Rocket.Chat and Discord get a readable message. A webhook gets versioned event JSON — Zapier, n8n, Make or your own endpoint, HMAC-signed. Email confirms before anything is sent.

The Channels view: seven destinations — Slack, Microsoft Teams, Google Chat, Mattermost / Rocket.Chat, Discord, Webhook and Email — with no channels registered yet — Registered channels are shared by every assistant; coverage decides who posts where. Test sends one real message through the channel immediately, so a misconfiguration surfaces now — and a channel with no event types selected is saved but inert.

After it is live

See what visitors asked, and what it couldn’t answer.

Every conversation is searchable, tool calls included. Questions the assistant could not answer from your knowledge are listed as misses, and a lead captured in chat lands in a queue, so the gaps in your documents show up as a to-do list.

The Conversations view: answer feedback, retrieval misses and queued leads at the top, a searchable list of what visitors asked, and the open conversation beside it — Answer feedback, retrieval misses and the queue of leads and unread conversations, above the conversation itself.

The Analytics view: conversations, messages, visitors and the share rated helpful over 30 days, with daily conversations and token usage charted below — Is anyone using your assistants, how much, and is it working: 30 days of conversations, messages, visitors and the share rated helpful.

Compared · Chatbase & Intercom Fin

Same job, three different bills.

Chatbase and Intercom Fin are the two products people evaluating Chat Studio ask about most. Both are real, capable products — the difference is mechanism: what you're billed for, and whether the model itself is something we resell to you. Figures below are each vendor's own publicly listed pricing, checked September 2026; both change pricing during the year, so verify directly before you decide.

	Chatbase	Chat Studio	Intercom Fin
Pricing mechanism	Message credits by plan tier, $40–$500/mo — credit cost changes with the model you pick	Knowledge storage by plan, plus requests answered from a prepaid balance at 500 per $1	$0.99 per resolved conversation, on top of a seat-priced Intercom plan
Bring your own model key	No — usage always runs on Chatbase's own credit system	Yes, and it is required on every tier — Claude, OpenAI, Gemini, or open models via OpenCode GO	No — runs on Intercom/Fin's own models only
Free tier	$0 — 50 message credits/mo, 1 agent, 400 KB training data	$0, no card — 1 assistant, all ten personas. Free covers our side of the infrastructure, within limits; you bring your own model key.	No free tier — a 14-day trial, then billed per resolution
Answers from	Trained data, plus optional connected actions (bookings, order lookups) through integrations	Only what you give it — your pages and documents; Minimal-served data from December 2026. Says so when the answer isn't there.	Your help center and connected help-desk data, with human handoff below a confidence threshold
Persona breadth	Support-first; additional agents configured separately	Ten at launch — support, sales, docs, onboarding, HR	Support-first, built into Intercom's own inbox
Where it runs	Embed widget, plus native integrations (Zendesk, Salesforce, Shopify, Slack, WhatsApp, and more)	One line of code, on any page you control	Native to Intercom's messenger; bolts onto Zendesk, Salesforce, HubSpot, Freshworks or Zoho at the same per-resolution rate
Beyond the sticker price	Branding removal (~$99/mo), extra agents, auto-recharge credits	Nothing from us beyond the plan price and the per-request meter. Your AI provider bills model usage at the provider’s own price.	An underlying Intercom seat plan (Essential, Advanced or Expert) is required

What they still do that we don't, today.

Chatbase ships native integrations to Zendesk, Salesforce, Shopify, Slack and WhatsApp; Chat Studio embeds on any page you control but doesn't connect to a help desk yet. Answering from Minimal-served data is in private use and opens publicly in December 2026. Intercom Fin publishes real resolution-rate case studies from its customers; our own usage numbers will appear here once we have them. Chat Studio is in beta.

The bill has two lines, and you control one of them.

The plan price and the per-request meter are the only things we charge. Model usage runs on your own key, at provider prices, billed by your provider — the two stay separate, so every line on the invoice is readable.

The Wallet: a prepaid credit balance and requests included this month, top-up buttons for $10, $50 and $100, and the rate card beside them — Prepaid and capped by design. When the balance reaches zero the assistant shows the contact form and nothing is billed past it. Every plan answers 500 requests a month before the balance is touched, and each top-up carries a 2% bonus.

See pricing

Put one on your site with one line of code.

Build the assistant on our hosted setup, test it in the preview against a real question, and paste one line into your page. Or book a session and we build it with you.

Start free Schedule a Personal Session

Next on the shelf

Where to go from here.

Ask StudioJ.27

Ask your database. Keep the answer.

Ask ClickHouse or Postgres in plain English; the answer is a dashboard that keeps answering as the data moves.

Open →

API Bayß

A dataset in, a priced API out.

Drop a file; we code, host, back up and serve the endpoints. You set the price and keep all of it.

Open →

Start free

$0, no card.

One assistant, all ten personas, and your own model key.

Start free →