Answers from the team that built it.

Under 2 minutes for a basic live chat. Sign up, create your first widget, copy a single <script> tag and paste it into your website's HTML before </body>. The chat bubble appears immediately. Configuring an AI agent takes another 2–3 minutes: choose a model, paste your API key, write a system prompt.

No. The 14-day free trial unlocks every Pro feature without a credit card. We only ask for payment when you choose a paid plan after the trial. You can also stay on the Free plan indefinitely with 1 agent, 1 widget and 100 chats per month.

Any website that lets you add a <script> tag: WordPress, Webflow, Wix, Shopify, custom React or Vue apps, static HTML, Next.js, Astro. The widget mounts inside a Shadow DOM with z-index 2147483647, so it never conflicts with your existing CSS or other tools.

Yes. The widget replacement is one line of code. Conversation history import is on the Pro and Enterprise roadmaps. For now, we suggest running both in parallel for a week and switching when you're comfortable. See our /intercom-alternative, /drift-alternative, /tidio-alternative and /crisp-alternative pages for side-by-side comparisons.

Anthropic Claude, OpenAI GPT, Google Gemini, Meta Llama (via OpenRouter) and Alibaba Qwen. Other models available on OpenRouter work too. You pick which model handles which widget and can switch in seconds, with no contract.

You paste your Anthropic, OpenAI, Google or OpenRouter API key into the AI Agent settings. NexTalk encrypts it with AES-256 at rest (per-tenant encryption key) and decrypts it only inside the AI worker process for each LLM request. The key never leaves the server, is never returned to the dashboard, and is never logged. You pay the LLM vendor directly at their cost. NexTalk does not mark up tokens or charge per resolution.

Intercom Fin charges $0.99 per AI outcome on top of seat pricing, so a spike in traffic means a spike in the bill. With BYOK, you pay the model vendor for tokens, which is predictable and transparent, and you keep the rate negotiated by your own enterprise account if you have one.

Yes, and we encourage it. Run Claude on your B2B sales page (high reasoning), a small GPT model on your docs (cheap and fast), Llama on an internal helpdesk (private) and Gemini on high-volume FAQ widgets. Each widget has its own model, prompt, schedule and persona.

Yes. Every AI agent has its own system prompt, with full control over tone, scope, knowledge boundaries, escalation triggers and brand voice. Test prompts before going live.

Configure escalation rules: keyword triggers (like “speak to human” or “refund”) and a fallback for when the AI can't answer. The AI hands the conversation to your human queue with the full chat history attached, so your agent picks up in context and never asks the visitor to repeat themselves.

Free ($0, 1 agent, 1 widget, 100 chats/mo), Starter ($9.99/mo, 3 agents, 2 widgets, 1,000 chats/mo), Pro ($19.99/mo flat, 10 agents, 5 widgets, 10,000 chats/mo, AI agents), Enterprise (custom: unlimited agents and chats, SSO, API access, priority support). See /pricing for the full breakdown.

$19.99/month covers your whole Pro workspace: up to 10 agents, 5 widgets, 10,000 chats. Not per seat. For comparison, Intercom Advanced is $85 per seat, so a 10-agent team pays $850/month (public price, June 1, 2026).

Zero. NexTalk does not charge per AI conversation, per resolution, per token, or any usage-based surcharge. You pay your LLM provider (Anthropic, OpenAI, Google, OpenRouter) directly at their published rates for tokens consumed.

We use Stripe. Monthly or annual billing. Upgrades and downgrades are prorated. Cancel anytime: your data remains accessible read-only for 30 days after cancellation, then is fully deleted per GDPR.

Yes, contact sales (/contact). We offer 50% off for verified non-profits and free Pro for solo open-source maintainers.

Yes, it's a flagship feature. Free includes 1 widget, Starter 2, Pro 5, and Enterprise as many as you need. Each widget has its own branding, AI agent, schedule, language, domain restriction and analytics. One workspace runs your marketing site, your docs, your customer portal and your e-commerce store, each one separately.

Brand colors, greeting text, agent avatar and name, welcome card content, chat bubble position (bottom-left or bottom-right), corner radius, font preference, and CSS overrides for advanced users (Enterprise). Match your brand without writing code.

Yes. Each widget supports an allowlist of domains. Embeds from any other origin are rejected, which protects against widget theft and unauthorized use on other sites.

Yes, fully responsive. It adapts to iPhone Safari, Android Chrome and in-app browsers. The chat opens as a bottom sheet on small screens instead of a floating window.

No. The widget loader is ~6 KB gzipped and loads asynchronously after page paint. The full widget bundle (~80 KB gzipped) only loads when the visitor opens the chat. No render-blocking, no Core Web Vitals impact.

Yes, via outgoing webhooks. Subscribe any URL to events: new_chat, new_message, conversation_closed, csat_submitted. HMAC-SHA256 signatures verify authenticity. Ready-made templates for Slack and Discord; works with Zapier, Make, n8n, Pipedream and custom internal systems.

Yes, on the Enterprise plan. OpenAPI 3.1 spec, JWT-authenticated, rate-limited. Programmatic access to conversations, messages, agents, widgets, AI agents and analytics.

Yes, on the Enterprise plan. SAML 2.0 and OIDC. SCIM provisioning available on request. We do not currently support consumer social sign-in (Google or Microsoft personal accounts) for the dashboard, only enterprise identity providers.

Yes. CSV export for conversations, messages, agents and CSAT in the dashboard. Full JSON export via API on Enterprise. GDPR-compliant data deletion on request.

Yes. Independent pentest in March 2026 by a third-party security firm. Critical and high findings were remediated within the disclosure window. Notable fixes: JWT moved from localStorage to HttpOnly, Secure, SameSite cookies (out of reach of XSS), AES-256 encryption for AI provider API keys at rest, strict CSP and security headers, and tenant isolation verified across every endpoint.

In PostgreSQL 16 hosted on Microsoft Azure (Europe region by default), with Redis 7 for cache and pub/sub. Conversations, messages, agents, settings and analytics are encrypted in transit (TLS 1.3) and at rest (Azure-managed keys, plus per-tenant application-layer encryption for sensitive secrets).

Yes. Tenant data is fully isolated. Visitor data minimization is built in (we don't track visitors across sites). Right to access and right to deletion are exposed via the dashboard and API. A Data Processing Agreement is available on the Enterprise plan.

Yes. Keys are encrypted with AES-256-GCM using a per-tenant data encryption key (DEK), which is itself encrypted with a key encryption key (KEK) rooted in Azure Key Vault. The plaintext key only exists in memory inside the AI worker process for the duration of each LLM call. Never logged, never returned to the dashboard.

Conversation history is retained per your plan's retention policy (default: 1 year on Free and Starter, 2 years on Pro, configurable on Enterprise). Visitors can request deletion at any time; you can also bulk-purge in the admin dashboard.

The backend is a horizontally scaled FastAPI cluster with a Redis pub/sub layer for cross-instance WebSocket fan-out, so capacity grows by adding instances. Talk to us (/contact) if you expect very high volume.

99.9% on Pro (best-effort), 99.95% on Enterprise (contractual with credits). We publish post-incident write-ups for any major event.

On request, for Enterprise customers. The full stack ships as Helm charts for Kubernetes (we use the same charts internally, on the same platform that runs Privum Cloud's SRExpert product). Contact sales (/contact) for self-host pricing and support terms.

The primary region is Europe (Azure West Europe / North Europe). A US region is available on Enterprise. Multi-region failover is on the roadmap.