Your Website. Your Game.
Your AI-Powered Live Chat

Not at all. Run one Docker command on a $5 server, then the browser-based Setup Wizard walks you through everything — AI provider + API key, widget design, embed snippet, Cloudflare Tunnel for HTTPS. Live in under 25 minutes.

TeleChat Enterprise connects to OpenAI, Claude, or Gemini. When a visitor messages, the AI responds instantly using your Knowledge Base (RAG). You monitor every conversation in the Admin Dashboard's Chat Explorer and take over with one click — the AI pauses automatically when an agent replies, and resumes the moment the conversation is resolved.

The Admin Dashboard (Enterprise) is a full web-based control center. It includes Chat Explorer for live conversations, AI Settings with multi-provider routing, Knowledge Base & Brain Graph for training your bot, Chat Design editor, App & Site Profiles, User & Role Management (RBAC), Budget controls, and Reports & Analytics.

Yes — any platform that supports HTML. Paste a single <script> tag before </body>. No plugins, no dependencies, no framework requirements. One line of code, any website.

Yes! Pro and Enterprise include a native Unity package (.unitypackage) with a WhatsApp-style chat UI. Players tap a floating button, type a message, the AI replies in real time, and your team takes over from the Admin Dashboard whenever they want. Works on WebGL, iOS, Android, Windows, macOS. Real-time SSE connection, typing indicators, date separators — all built in.

(1) You pay once — they charge $45-$295/month, forever. (2) Your data stays on your server — they store every conversation, dossier, and uploaded document on theirs. (3) Built-in AI with YOUR OpenAI/Claude/Gemini key — they charge extra per AI reply or lock it behind their highest tier. (4) Multilingual mood detection + escalation alerts in 15 languages — they don't. (5) Multi-brain knowledge with per-site allow-lists — they don't. (6) Works inside Unity games — they don't.

Zero limits. TeleChat runs on your own server — no artificial caps. A $5 VPS handles 500+ concurrent visitors. Need more? Just upgrade your server, no license changes needed.

Built-in Cloudflare Tunnel provides free HTTPS without opening ports. The admin API uses JWT auth with rate limiting. Enterprise adds RBAC user management, audit logs, and budget controls for full team security.

Basic: Self-hosted core — Docker deploy, real-time SSE messaging, file attachments, idempotent / race-safe operations, nightly backups. Pro: adds the Setup Wizard, Unity game client, Cloudflare Tunnel (free HTTPS), widget design wizard, and 30-day email support. Enterprise: adds the full Admin Dashboard with Chat Explorer, AI auto-replies (OpenAI/Claude/Gemini), RAG knowledge base, multi-brain support, Brain Graph, multilingual mood detection + escalation alerts, internal notes & watchers, multi-site profiles, RBAC team management with audit log, budget controls, canned responses, and reports.

Pro includes 30-day email support. All tiers include documentation, the interactive setup guide, and future updates.

AI replies work in every language the connected provider supports (OpenAI / Claude / Gemini cover 95+ languages). The mood detection layer ships first-class lexicons for 15 languages — English, Turkish, Spanish, French, German, Portuguese, Italian, Arabic, Russian, Chinese, Japanese, Korean, Dutch, Polish, Hindi — so the operator sees colour-coded sentiment + escalation alerts on visitors in those languages. The Chat Explorer UI itself is English; visitor + agent messages render in any UTF-8 script (CJK, RTL, Cyrillic, Devanagari) out of the box.

Yes — Enterprise's multi-site profiles + multi-brain knowledge are designed for it. One install hosts many sites, each with its own widget design, AI system prompt, allow-listed knowledge brains, and operator team. A site bound to brain-a only reads brain-a's documents — no cross-tenant leakage. Strict scope filters at the chunk level + an audit-tested isolation contract (the Postgres + file-vector layers BOTH enforce the allow-list, no SQL-WHERE-only trust).

Every visitor message is scored on polarity (−1…+1), intensity (0…1), urgency tier (low / medium / high / critical), and emotion (happy / frustrated / angry / sad / anxious / confused / neutral). Two-tier pipeline: a 50-µs on-prem lexicon classifier runs first; an optional LLM Tier-3 re-verifies low-confidence or critical hits using your chosen provider. The Chat Explorer row paints its colour from the live aggregate, the badge shows the emotion, a red banner pops if a critical signal lands. Sticky audit flag preserved so a supervisor can reconstruct what happened even after the visitor calmed down.

Both directions. Visitors send images and PDFs from the widget; agents reply with images, video, manuals — anything up to 100 MB per file. URLs are signed and routed through an allow-listed proxy (path-traversal blocked, only /api/uploads/* persisted), and the thread renders inline image previews. Internal notes can carry attachments too — useful for sharing a one-page playbook with a colleague mid-conversation.

Toggle the composer from Reply to Internal note — the message lands in the thread with a dashed amber border, every agent on the team sees it, the visitor never does (visitor SSE filter is default-deny by allow-list). Great for "this customer escalated last week, be careful" tags, or for a senior agent leaving a one-line answer for a junior to send.

Zero. There's nothing to cancel — you bought a one-time license, the code runs on your server, the database is your Postgres. Daily backups land in your filesystem (or any S3-compatible bucket you configure). Export the database, walk away, run the same install on a new VPS — the export reads back identically. No "vendor cloud", no "we keep your conversations" — your machine is the canonical store.

Chat Explorer surfaces live presence (who's online), typing indicators (per agent + per visitor), watchers (everyone currently viewing this chat), and a race-safe assignment system — two agents click "Take" on the same conversation at the same instant, only one wins, the other gets a clean message instead of a silent overwrite. Keyboard shortcuts (J/K next/prev, R reply, E resolve, I toggle AI) keep an experienced operator at ~3 chats a minute.

Knowledge Base page → drag-drop PDFs / DOCX / HTML / ZIP archives, or paste URLs and the importer fetches them (with SSRF + content-type guards so a bad URL can't poison the index). Documents auto-chunk, embed against your chosen provider, and land in a brain you pick from the chip strip. Need a separate brain per product line? Click New brain, give it a name, drop documents in. Bind the brain to a site under Apps & Sites → Knowledge.