7 days free on Starter, Pro, and Business for a website's first subscription, with limited usage and no charge until day 8. Upgrade mid-trial anytime, or cancel before the charge through Manage subscription in your dashboard. Pick the plan by the job you want done:
No add-on subscriptions for features — optional message packs are the only extra. No glue code. What your plan includes is in your account the moment it starts (usage is capped at the trial limit during a free trial).
Most AI chat tools bill by credits, content caps, or per resolved conversation — so a busy month quietly costs you more. Here, your plan price is the price: each plan carries a monthly message allowance, and optional message packs add more.
Try Starter, Pro, and Business free for 7 days on a website's first subscription. Each plan bundles its features — no separate feature shopping.
7-day trials on Starter, Pro, and Business require a card. We charge $29 on day 8 for Starter, $99 for Pro, and $399 for Business, plus applicable tax — cancel anytime in the billing portal, no charge if you cancel before day 7. Billed monthly after that.
Running in production today: role-based access with granular permissions · team accounts with per-section operator access and one-step transfer of ownership · append-only audit logs (CSV export for platform admins) · TOTP two-factor auth · HMAC-signed service-to-service calls · self-serve website data deletion plus a per-site subject-access export. On the roadmap: SSO / SAML / OIDC, roles you define yourself and an organisation boundary spanning several companies, bring-your-own LLM keys, and on-prem / VPC deployment. See the public roadmap, then talk to sales about your contract.
Below is what works in production today. Items that are still on the roadmap live in their own subsection at the bottom — and the headline ones on the public roadmap.
| Feature | Starter$29/mo | Pro$99/mo | Business$399/mo | CustomTalk to sales |
|---|---|---|---|---|
| Capacity | ||||
| Messages / month | 1,000 / month | 4,000 / month | 20,000 / month | Custom |
| Pages crawled | 500 | 2,500 | 10,000 | Custom |
| Uploaded files (PDF · DOCX · TXT) | 10 MB | 100 MB | 1 GB | Custom |
| Re-crawl frequency | Weekly | Daily | Every 12h | Scoped per contract |
| AI quality | Latest generation | Latest generation | Latest generation | Latest generation |
| Knowledge search | Standard | Smart | Premium | Premium |
| Voice TTS | — | 150,000 chars / mo | 500,000 chars / mo | Custom |
| Voice STT | — | Unlimited (browser) | Unlimited (browser) | Unlimited (browser) |
| Calendar booking | — | 100 events / mo | 500 events / mo | Custom |
| Webhooks | Enabled | Enabled | Enabled | Enabled |
| Answers & knowledge | ||||
| PDF, DOCX, TXT uploads (total) | 10 MB total | 100 MB total | 1 GB total | Custom |
| Visitor document analysis (price a file, understand a file, or your own template) | — | 200 uploads / mo, 10 MB, local parsers | 500 uploads / mo, 25 MB, Gemini native PDF | Unlimited + custom templates |
| Grounded / low-confidence / no-evidence labelling | ||||
| Keeps itself fresh — auto re-crawl | Weekly | Daily | Every 12h | Scoped per contract |
| Business type auto-detection | ||||
| Editable AI persona & tone | ||||
| Auto language detection | ||||
| Visual result cards | ||||
| Lead capture | ||||
| Lead inbox in your admin dashboard | ||||
| Lead capture (intent + scoring) | ||||
| Customize the lead form | ||||
| Conversation auto-fill | ||||
| Outbound lead webhook | ||||
| HMAC-signed webhooks + dead-letter queue | ||||
| Leads CSV export | ||||
| Calendar booking | 100 / mo | 500 / mo | Custom | |
| Operator invites with per-section access | ||||
| Hand the account over to someone else | ||||
| Voice calls from the widget | ||||
| Live chat translation for operators | ||||
| Proactive messages (the agent speaks first) | 1 rule, standard wording | 3 rules + your own wording | Unlimited rules | Unlimited rules |
| AI follow-up emails (cold-lead autopilot) | 40 / mo | 200 / mo | 1,000 / mo | Unlimited |
| Appointment reminders | ||||
| Live Chat | ||||
| Human handoff with context | ||||
| Live Agent — talk to a real human, with your team | 2 seats | 10 seats | Custom seat count | |
| Search past live-chat conversations | ||||
| Post-chat rating (CSAT) | ||||
| Voice | ||||
| Voice input (Speech-to-Text) | Unlimited | Unlimited | Unlimited | |
| Voice output (Text-to-Speech) | 150k chars / mo | 500k chars / mo | Custom | |
| Brand & widget | ||||
| Language + theme | ||||
| Bubble position + size | ||||
| Bubble label + visibility | ||||
| Brand colors (light + dark) | ||||
| Bubble offset fine-tune | ||||
| Proactive auto-show + delay | ||||
| Persona — bot avatar + logo | ||||
| Custom bubble icon | ||||
| Remove "Powered by" (white-label) | ||||
| Analytics | ||||
| Visitor Engagement — the number that grows from day one | ||||
| Overview dashboard | ||||
| Message Analysis | ||||
| Knowledge Gap Detection | ||||
| Conversations & outcomes | ||||
| Conversation topics | ||||
| Per-topic sentiment | ||||
| Geographic & Languages | ||||
| Export Data | ||||
| ROI & cost savings | ||||
| Chats with a reply (share of conversations) | ||||
| Document analysis insights | ||||
| Integrations & channels | ||||
| Slack native (OAuth + Block Kit) | 1 workspace | 1 workspace | 1 workspace | |
| Calendly | ||||
| Cal.com | ||||
| Custom AI actions (no-code tools) | 1 action | 5 actions | 15 actions | Unlimited |
| Shopify order status | 250 / mo | 1,000 / mo | Unlimited | |
| Public REST API + API keys | ||||
| AI agents on your site (customers' AI assistants) | ||||
| Your own AI assistant, connected to your account (MCP) | ||||
| WhatsApp Business channel | ||||
| Instagram DM channel | ||||
| Facebook Messenger channel | ||||
| Security & compliance | ||||
| Rate limiting + fingerprint protection | ||||
| Two-factor authentication (TOTP) | ||||
| High-uptime infrastructure | ||||
These items are committed and scoped — they'll land in the plans below as they ship. The headline items are also on the public roadmap.
“I built NovixAgent as a reliable, full-featured AI agent for websites — bringing grounded answers, lead capture, voice, live chat with automatic translation, booking, and multilingual conversations into one product. The goal is simple: give every business an agent that can do more than chat and still stay accurate.”

The visitor's connection and the call to the AI provider travel over TLS.
The chat and embedding AI providers we use exclude training on your content under their business-tier terms.
Privileged actions on your account are written to a journal no request reaching the product can rewrite or erase; an entry leaves it only when its published retention period ends.
Delete a website yourself; a paid one is held recoverable for a grace period before its data is destroyed. Download a subject-access export of the records of a site when you need one, whichever plan you are on.
We don't ask you to trust us — we ask you to check.
We build Custom contracts case-by-case. Tell us about your deployment shape, compliance roadmap, or integration needs and we'll discuss what's feasible together.
Talk to salesYes — all three. Connect a WhatsApp Business number, an Instagram professional account, or a Facebook Page in Dashboard → Channels, and incoming customer messages use the same agent knowledge, lead capture, and conversation history as your website. They are included on Pro and up, and channel replies draw on the same monthly message allowance rather than a separate charge. On Pro and up, Live Agent can hand the thread to an operator. Messages the agent starts are a separate capability: appointment reminders can already go out over WhatsApp, while WhatsApp follow-ups on the lead-status playbook are still coming soon — that playbook sends email only.
No. Answers that go through retrieval are tagged grounded, low-confidence, or no-evidence: a reply backed by your content shows the source pages or files beneath it once the reply has finished, except when the visitor asked for your team, and when nothing in your content was retrieved for the question the agent says it can't help with that one rather than guessing, and offers to put the visitor in touch with your team. Citations are assembled on our side from the passages actually retrieved, and the agent is instructed not to write a source list of its own, so the Sources shown beneath a reply come from what was retrieved. On Business, the Messages tab lists the questions visitors asked that your content could not answer, as they wrote them (the first 200 characters, where the site stores question text), so you know what content to publish next. How much of your tier-1 volume gets deflected depends on how thoroughly your own content covers what visitors ask — that's the lever you control.
Yes — that's the AI Assistant, and once you turn follow-up on it is handled for you. For each lead status you pick a channel and, if you like, a goal; when a lead reaches that status the assistant writes a short message and sends it automatically. Booking businesses can also send automatic appointment reminders a set number of hours before the meeting — by email, or on Pro and up over WhatsApp once you've connected WhatsApp Business, where the reminder is a Meta-approved template with Confirm and Reschedule buttons. Because nobody reads these automatic messages before they go out, each one is written only from the goal you set for that status, your reply template, your business type, the topic of the request and the visitor's first name (only when the name as given holds nothing but letters, spaces, apostrophes and hyphens, and its first word has at most one apostrophe or hyphen inside it), plus the booking details for a reminder — never from the visitor's own words. If a written reminder would carry any link other than the booking's own links (its reschedule or cancel link and, when the meeting's location is a web address, that address), a link to your website, or a bare address already written in your own settings or the booking details — counting any bare address that ends in a real top-level domain as a link — it is not sent: it is kept as a draft in the lead for you to send or skip. A WhatsApp reminder is the approved template itself, so this check does not apply to it; one is skipped rather than sent, and uses none of the allowance, when the name the customer gave has no first name the template can use — most often because it carries a title or initial with a dot, a comma or a digit, or a first name joined more than once. For a personal answer to exactly what a visitor wrote, open the lead and use Generate AI reply; you read that draft and send it yourself. Email follow-up is live on every plan (40 AI drafts a month on Starter, 200 on Pro, 1,000 on Business, unlimited on Custom). The monthly allowance is spent when a message is written, so sending it is free; a reminder held as a draft or skipped because the booking was cancelled has still used its unit. Every follow-up email goes out under a sender name taken from your website's domain (a workspace with no website sends under our product name), with replies coming back to you and a one-click unsubscribe. Still coming soon: WhatsApp on the lead-status playbook (that playbook is email-only today), and an AI voice confirmation call for a booked appointment — a customer who left a phone number would get a consent-based confirmation/reminder call, booking-only and never cold outreach.
Start a trial on Starter, Pro, and Business — your pick — for a website's first subscription (one trial per customer) and you get that plan's features for 7 days, no charge; usage during the trial is limited. We require a card so the renewal happens automatically — no surprise bills, no manual upgrade flow. You can cancel in the billing portal before day 8 at zero cost. Already getting value on day 5? Move up a plan mid-trial — the change applies right away, the trial keeps its end date, and the first charge for the new plan falls on that date. Prefer to start paying sooner? “Start paid plan now” ends the trial immediately.
Two reasons. One: it filters out bot signups — agents are expensive to host, so we'd rather not run free for accounts that never convert. Two: it removes the friction of a separate upgrade flow. On day 8 → automatic renewal if you keep it, $0 if you cancelled. No re-entering card data.
Two layers of protection. First, the 7-day trial is a real test drive of the plan you pick, with usage limited for the length of the trial — cancel before day 8 and you're charged $0. Second, on monthly billing the worst case is one paid month while you decide; a yearly plan is paid for the whole year up front, and cancelling it stops the renewal rather than refunding the rest of that year. We don't refund unused time unless the law requires it, but if something is genuinely broken on our side (the agent didn't behave as advertised), email us and we'll work it out case by case.
Yes. Switch to Pro or Business anytime during the trial — the change applies right away, the trial keeps its end date, and the first charge for the new plan falls on that date. Usage stays limited until the trial ends. If you would rather start paying sooner, “Start paid plan now” ends the trial immediately and charges for the plan then; if the payment does not go through, the trial simply continues.
Anytime. Upgrades take effect immediately — a prorated charge for the rest of the cycle, and the bigger message limit applies right away rather than next cycle (messages you've already used this month still count against it, and your reset date doesn't move). Downgrades take effect at period end — your widget keeps working, your settings stay saved.
Yes, and there is no contract beyond the period you've paid for. Cancel in the billing portal and the renewal stops; you keep access until the end of that period.
Cancelling stops the next renewal; it does not end the plan early. You keep full access until the end of the year you paid for, and we don't refund the unused time unless the law requires it. If something is genuinely broken on our side, contact us and we'll work it out.
One visitor question plus the agent's reply = one message. So a 6-turn conversation is 6 messages. The count is taken when the question is admitted, before the agent retrieves anything, so a reply that declines for want of evidence still counts against the allowance. What doesn't count is a failure on our side: if the model errors out or your knowledge base is briefly unavailable, the message is credited straight back.
Yes — when you hit your monthly cap the agent stops answering questions, including inside a conversation already in progress, until your allowance resets. The widget itself does not go away: it stays on the page, says the allowance is spent and asks for contact details, showing your lead form where you have one set up, and your own operators can still reply in live chat, which never draws on the message allowance. That reset is monthly on your website's own reset date, so a yearly plan still refreshes every month rather than once a year. You can also upgrade to the next tier at any point (instant, prorated).
Yes — every account supports unlimited websites. Each website has its own plan, its own agent, its own settings, templates and analytics. So you can run a Starter site for your blog and a Business site for your agency client side by side, billed separately.
Yes — Live Agent is a real-time human handoff inside the same widget (see /use-cases/live-chat): a visitor clicks "Talk to a human" and reaches an operator directly, no separate live-chat tool to install. The button only shows when an operator is online and within your configured working hours. Your team works from a shared inbox in the dashboard with the visitor's page, device, location and captured lead already on screen, plus saved replies, file attachments, and a typing indicator. It comes with operator seats included in your plan (2 seats on Pro, 10 seats on Business — invited teammates; you never use a seat yourself), separate from the AI message cap. When the AI cannot answer, it still offers to take the visitor's contact details, and that lead reaches your team by email, Slack, or webhook (see the CRM/Zapier/Slack question above). Automated team routing — auto-assignment, departments, SLA targets — is on the roadmap (see /roadmap).
Yes — on Pro and up, visitors can start a real voice call with your team straight from the chat window. Calling is a switch you turn on in your live-chat settings (it starts off), and each operator chooses whether they take calls. The call runs browser-to-browser: the visitor taps the call button, the browser asks for microphone permission, and they're connected — no phone numbers on either side, and no per-minute fees or per-call charges, because calls use the same operator seats as live chat. The call button follows the same presence rules as live chat, so it only appears while an operator is online. Every available operator rings for up to 45 seconds — online, not already on a call, and not opted out of calls — and whoever accepts first takes it; if nobody picks up, the visitor is offered your lead form where you have one set up, so a missed call still becomes a lead. In most live-chat tools, only the agent can start a call — visitor-initiated calling is usually an enterprise contact-center feature.
Yes — live chat is built for teams. Invite operators by email (a personal invite link, valid 72 hours, only that address can accept) and pick either the websites each one can staff or access to the whole account, which covers every website you add later too. The team shares one inbox with Mine / Unassigned / All views: assign a conversation to yourself in one click, or just reply to an unassigned chat and it's claimed automatically. A live indicator shows which teammate is already viewing a conversation, so two people can avoid answering the same visitor. Everyone shares the site's saved replies (any operator can add one), and each operator gets personal desktop notifications with sound plus their own visitor-facing identity — show a real name or a display alias. Plans include 2 seats for invited operators on Pro and 10 seats on Business; the account owner is always on top of that and never consumes a seat.
Yes — live chat translates in both directions, and it's included with Live Agent rather than sold as an add-on. Your operator reads each visitor message in their own language (in a browser with the built-in translator and an available language pair) and types replies in it; the visitor writes and reads in theirs. 38 languages are supported, the visitor's language is detected automatically and follows them if they switch mid-conversation, and there's nothing for the visitor to install, choose, or turn on — any browser, any device. Your operator sees what they typed while the chat stays open in their tab, with the exact version the visitor received one click away; after a reload, their earlier replies are shown translated back from the version the visitor received. The translation runs on the operator's own device, inside their browser, so the conversation text isn't sent to an outside translation service. Two things to know: it needs a desktop browser with the built-in translation engine on the operator's side, such as a recent Google Chrome or Microsoft Edge — Safari, Firefox, and mobile browsers don't have the engine, so an operator there works without translation and the dashboard says so, while visitors are unaffected whatever they use; and it is machine translation, not a human translator, which is why the original is always one click away. The first use of a language pair downloads a small model (one click, a few seconds), then it's instant. It runs on the operator seats your plan already includes — not per message, not per translated word.
Exactly what you chose when you invited them, and you choose one of two things. Name the websites, and they see those and no others: Live Chat is always included and everything else is an opt-in checkbox — Leads, Analytics, Knowledge Base, Integrations, AI Templates, Chat Configuration — with an invite starting on Live Chat and Leads and every other box unticked. Or give access to the whole account, which reaches every website you have now and every one you add later. That answer carries its own set of the same six boxes: the invite form hands out all six, and you can switch any of them off afterwards without pinning the person back to named websites. Shared sections open in an operator mode scoped to the job: Knowledge Base and Integrations are read-only, and integration cards do not show secrets or API keys; on Leads an operator can move a lead through the pipeline but has no Delete; AI Templates and Chat Configuration are editable when you tick them. Six things are never shared and cannot be ticked for anybody: billing, account settings, website management, team management, Channels, and the outbound Assistant — connecting a messaging channel and reaching back out to a lead both act as you, so they stay with the account owner. You can edit a teammate's permissions or revoke access at any time — it applies immediately.
Yes — every plan. The agent replies in the language of the visitor's message, and the widget's own interface follows the visitor's language where it is one of the widget's 31. A visitor from Germany sees the chat in German, a visitor from Mexico sees it in Spanish, with no extra setup on your end. Your knowledge base stays in whatever language you wrote it; the agent handles the translation when it matters. Source-side multi-language content translation (translating retrieved chunks before answering, for sites that are English-only) is on the roadmap for Business and Custom — see /roadmap.
Custom is sales-led — for teams whose procurement requirements aren't met by Starter / Pro / Business yet. Common asks here (SSO/SAML, an organisation boundary spanning several companies, roles you define yourself, on-prem or VPC deployment, BYO LLM keys, a SOC 2 report) are on our public roadmap (see /roadmap) rather than shipped today — we'd rather have an honest conversation about timelines than promise something we can't deliver. Pricing is per contract. Reach out via /contact to start a conversation.
It stays yours. Conversations travel over TLS, the chat and embedding AI providers we use exclude training on your content under their business-tier terms, and you can delete a website and its data yourself (a paid website is held recoverable for a grace period first) and export its leads, or a subject-access copy of the site, whenever you need to. Our own support staff see conversations with email addresses (accented and non-Latin letters included), phone numbers written with a plus, with spaces, dashes or slashes between the digit groups, or in dotted pairs, and card numbers (13 to 19 digits) automatically masked server-side (a name typed into the chat stays as written unless the visitor gave it in one of your lead forms in that same chat) — raw content requires a separate, audited, MFA-gated workflow. GDPR-aligned consent banner included. Full details at /security.
Status today: the underlying controls (TLS, audit logs, MFA, signed webhooks, encryption, role-based access, no-training terms with our chat and embedding AI providers) are running in production. The third-party SOC 2 Type II attestation is on the audit calendar — window opening Q4 2026, report expected Q2 2027 — and ISO 27001 is planned for 2027. SSO/SAML, an organisation boundary spanning several companies, roles you define yourself and on-prem or VPC deployment are on the roadmap for the Custom contract tier — see /roadmap for the live status. If a missing certification is a deal-blocker for your team, tell us at /contact?subject=security — we'd rather know before you sign than after. Full security primer at /security.
Visitor questions go unanswered without an agent, and some of those visitors don't come back. 7 days free on Starter, Pro, and Business. Live today. Cancel anytime in the billing portal.
The claims on this page are ones you can check.
Read what ships, what is coming and what we do not have yet: the changelog, the roadmap and the security page. Then start a trial and test the agent on your own content.