How SupportBot works

Overview

SupportBot is an AI support agent that lives inside your existing chats — Discord, Slack, Telegram, WhatsApp, Signal — plus a drop-in HTTP API + chat widget for any website. It imports your community's full message history, extracts every solved problem into a Knowledge Base, and answers new questions with citations to the original threads.

The same brain runs across every channel: a single Knowledge Base, a single Inbox for human-handoff cases, a single dashboard. You don't run a separate bot per platform.

Who it's for: communities (Discord/Telegram), B2B SaaS support (Slack/widget), high-throughput direct-message support (WhatsApp/Signal Business). Free during Beta with no message cap.

Getting started

Three steps to a live bot in any channel:

Sign in on /dashboard with your phone number (Telnyx SMS OTP). No email, no password.
Add a chat: click Connect, pick the platform, and either invite the bot directly (Discord/Slack/Telegram) or scan a QR code (WhatsApp/Signal). The bot starts listening immediately.
Import history (optional but recommended): the dashboard offers a one-shot import of the chat's prior messages so the very first answer already cites real solutions instead of guessing from training data alone.

That's it — the bot answers @mentions, replies to questions matching its judgement criteria, and surfaces ambiguous cases to your Inbox.

Messenger integrations

Same brain, different transport. Each platform has its quirks but the experience is consistent:

Discord

Add the bot via OAuth with the standard scopes (read messages, send messages, read message history). Works in text channels, threads, and forums.
Forums: the bot opens each post and replies on-thread — sources cite the post that solved a similar question previously.
Per-guild branding: if you set a brand in the dashboard, the bot member gets that name + avatar inside your server (Discord per-guild profile API).

Slack

OAuth install, channels:join scope. The bot watches public channels you invite it to. Per-message, the bot appears with the brand's name + icon (Slack username/icon override per-message).
Threaded replies: every answer goes in-thread on the question's thread, never in the main channel.

Two paths: shared bot (fastest) or your own BotFather bot (lets you set a custom @username + avatar). The dashboard explains both.
Group messages, threads, and private DMs all work.

Linked-device QR scan from a WhatsApp Business phone. Bot appears as that phone's profile across all chats it joins.
Group joins are by invite link; admins can also add the bot's number directly.

Signal

Linked-device QR scan against a Signal account. Same brain across DMs and groups.

History import

The single highest-leverage feature. Most communities sit on years of "we already answered that" — and SupportBot harvests it.

What it does

Pulls every message in your chat (text, images, voice, video transcripts).
Identifies question → answer pairs using a multi-step LLM gate (skips greetings, off-topic chatter, etc.).
Saves each pair as a Case with the original message link as a citation source.

Per-platform mechanics

Discord: uses the bot's read-message-history scope; runs in chunks of 5K messages with a progress bar in the dashboard.
Slack: conversations.history pagination, same chunked progress.
Telegram: backed by the Telethon user-session you authorize once via QR.
WhatsApp / Signal: requires a one-time scan of WhatsApp Business / Signal Desktop with full chat scrollback enabled.

You can re-run import at any time; the bot deduplicates and only ingests new messages.

Knowledge base

Every Case extracted from history (and every confirmed-good answer the bot writes since) is embedded into a per-group vector store (ChromaDB) plus a structured "what's the question / what's the resolution" record. New incoming questions:

Get embedded with the same model.
Top-K cases are pulled by semantic similarity + recency boost.
The synthesizer LLM writes a fresh answer cited to those cases — never copy-pastes the old reply verbatim.

What you can do with it

Browse: the dashboard shows every Case with its question + answer + source link. You can edit, delete, or merge.
Upload docs: drop in PDF/Markdown/HTML files and the bot indexes them alongside the chat-derived cases.
Tag & filter: tag cases by topic so the bot biases toward the right ones in retrieval.

Auto-brand

One URL, one click — the chat widget reskins to match your site. The pipeline:

Probe the URL through a headless browser (real DOM + computed styles, not just HTML).
Extract a brand brief: brand-identity color (not marketing-hero gradient), secondary accent, fonts, density, shadow recipe, header treatment, button shape, and the brand's voice.
Code CSS that reproduces the host site's design language — gradient headers for monobank-style brands, white-clean for Grammarly-style, solid for Nova Poshta, etc.
Critic reviews the rendered widget against the homepage and pushes corrections.
Iterates 4× then persists the final brand to your group. The widget renders that on every page-load thereafter.

API: POST /v1/auto-brand/stream emits SSE so a UI can show live progress. Soft-capped at 3/day per group.

Chat widget

Drop-in <script> tag — same brain as the messenger integrations, just exposed via a floating bubble on your site:

72px launcher in the corner, expands up to a 410×660 chat panel on click (responsive — shrinks to fit smaller viewports).
Threads: end-users get persistent conversation history (localStorage on their device, plus end_user_id binding for cross-device continuity).
Image upload, file attachments, and (where the brand allows) human-handoff into the same Inbox the messenger questions use.
Auto-brand applied: if you've generated a brand for this group, the widget reskins automatically.

The bubble you're seeing on this page right now is the widget — try it.

Inbox & human handoff

Not every question should be answered automatically. The Inbox catches:

Cases the bot is unsure about (low confidence + no good Knowledge Base match).
Explicit "talk to a human" / escalation requests.
Questions tagged for review (you can mark categories as always-Inbox).
SLA escalations: if the bot hasn't replied within X minutes (configurable per group), the case is auto-Inboxed.

Your team handles them in the dashboard or directly in the source channel — every reply you send becomes a future Knowledge Base case.

Cases & sources

Every answer the bot writes carries up to 5 source citations linked to the original message thread. Click a citation in any messenger and you jump to a per-case page on supportbot.info showing the question, the answer, the conversation around it, and similar cases.

Sources are verifiable — readers can see the exact context.
Wrong answer? Click "edit" on the case page; the bot retrains within minutes.
Anonymized sharing: the public case URL hashes member identifiers — no personal data leaks even when shared externally.

Pricing

Free during Beta — no message cap, no per-seat pricing. After Beta:

Free: 1 active group, basic Inbox, community support.
Pro — $39/mo: 3 active groups, full Inbox + analytics, priority support.
Business — $99/mo: unlimited active groups, branded widgets, SLA escalation, custom KB sources.

See the live pricing matrix on the homepage.

Privacy & data

The short version:

Phone numbers of regular group members are not stored — only irreversible SHA-256 hashes used for thread continuity.
Admins and designated support team members are deanonymized (their names appear instead of hashes).
Private DMs between members are not processed.
Messages from groups where the bot is not a member are not processed.

Full details: Privacy Policy & Terms of Service.

Need help?

Start here:

Developer API reference — every endpoint, every field, with curl examples.
How it works — the long-form story behind the architecture.
Privacy & ToS — exactly what we store and what we don't.
Dashboard — manage your groups, KB, Inbox.

Or just click the chat bubble in the bottom-right of this page — that's the same SupportBot, hosted on supportbot.info, and it's primed with all of these docs.