Point SiteHelp at your website and help docs. It builds a knowledge base, answers visitor questions in your voice, and brings in a human exactly when it should.
Give SiteHelp a URL and it crawls your site to draft help articles automatically. Review, publish, and the assistant starts answering from them — grounded in your real content, so replies are accurate and on-brand.
Rules, hand-off and modes keep the AI helpful — and keep your team in control.
Match what a visitor types to your own rules and steer the answer — pricing, refunds, onboarding, anything.
The AI brings in a human the moment a conversation needs one, with full context attached.
Auto-reply, suggest-to-agent, or off — switchable per site so you choose how hands-on to be.
Let the AI run your tools mid-conversation — look up an order, restart a service — as signed webhooks or MCP tools you connect, with visitor confirmation or on autopilot, capped and audited.
The same knowledge base powers a branded, searchable self-service portal, so answers work even before a chat starts.
The two-way email inbox lands in the same queue, so the assistant's knowledge and your team cover every channel.
The whole widget — greeting, pre-chat, business hours, ratings — speaks the visitor's language, and the AI replies in whatever language they write in.
Batch-test up to 100 real questions before going live: see which answers are grounded, which hand off, and how fast — then fix the gaps and flip the switch.
Triggers can let the AI write the opening line from the page the visitor is on — "Comparing plans? Happy to help you pick" — with your copy as the fallback.
Callers talk to your grounded AI in real time, in their language — it answers from your knowledge base and hands off to a human or voicemail exactly when it should. Every call is transcribed and counts in your resolution analytics. See voice →
Connect an MCP server — your admin API, billing system, ops runbook — and the assistant calls those tools mid-conversation to actually resolve the request. Capped, audited, and injection-guarded. See MCP →
An AI assistant is only as good as its worst answer — so every part of the loop is instrumented.
Strict grounding mode makes the AI abstain and hand off to a human when your knowledge base has no real answer — no improvising.
Closed conversations are auto-scored on tone, resolution, accuracy and compliance, so you see quality drift before customers do.
Unanswered questions cluster into topics — paraphrases count as one — and become drafted articles overnight. Review, publish, and the gap never opens twice.
A strict, defensible AI resolution rate — bot answered, no human needed, chat finished — plus what those resolutions would cost at $0.99-per-resolution vendors.
Bring your own Anthropic or OpenAI key — one key can cover every site and brand — and pay raw provider cost, typically 10–40× less than $1-per-resolution vendors. Run the numbers →
Token usage and spend per site, per day, per purpose — so AI costs are a line item you control, not a surprise.
Answer instantly, around the clock, from your own knowledge — and escalate only when it counts.