An AI that actually answers — grounded in your content
Train an AI agent on your URLs, docs, FAQs, and Q&A pairs. It answers in your voice, cites its sources, calls your APIs to take actions, and hands off to a human the moment it's not confident.
CustomerHero AI
Online · Powered by RAG
What's your refund policy for annual plans?
Annual plans can be refunded pro-rata within the first 30 days. After that, we credit unused months toward future renewals.[1]
Want me to start a refund for your annual plan?
Carla joined the conversation
Hi! I can see the full thread — let me help finalise that refund.
Point it at the content you already have
No prompt engineering. No fine-tuning. Drop in URLs, upload PDFs and Word docs, paste in Q&A pairs, or write text directly. The agent re-indexes automatically when content changes.
URL crawl
Crawl an entire site or a single page; respects robots.txt and JS-rendered content.
Files
PDF, DOCX, CSV, TXT — parsed and chunked automatically.
Q&A pairs
Author exact answers for high-intent questions; never fragmented.
Free-form text
Paste in policies, scripts, or product notes — usable as-is.
Retrieval-augmented, not hallucinated
Every answer is grounded in your content via hybrid retrieval — vector search plus keyword search — re-ranked by an LLM judge for relevance. Inline citations link back to the exact source.
Hybrid search
Vector + BM25 keyword retrieval catches both semantic and exact-match queries.
Reranking
A reranker reorders candidates by true relevance before the model answers.
Inline citations
Every claim is annotated with the source it came from — auditable, traceable.
Query rewriting
Follow-ups like "how much is it?" are rewritten into standalone search queries automatically.
Beyond answers — your agent can take action
Wire up actions to your existing APIs. The agent decides when to call them, confirms with the customer when it should, and the request is executed server-side with your credentials — never exposed in the browser.
Tool calling
Define actions with JSON schemas; the agent decides when each is appropriate.
Server-side execution
API keys live on the server. The browser never sees your credentials.
Confirm before mutating
Read-only actions run silently. Mutations ask the customer to confirm first.
Real-world examples
Look up an order, cancel a subscription, generate a quote, schedule a meeting.
Handoff to humans the moment it matters
The agent knows what it doesn't know. Low-confidence answers, sensitive topics, and explicit "talk to a human" requests trigger an escalation — with the full conversation history attached so your team picks up exactly where the AI left off.
Confidence thresholds
Per-chatbot tuning for when to deflect vs. escalate.
Smart routing
Route to the right team based on labels, business hours, and load.
Full context handoff
Agents see the entire conversation, including what the AI tried.
Re-engagement
AI can rejoin closed conversations to follow up automatically.
Built for production, not for demos
Guardrails & evaluation
Set a system prompt to control tone, refuse out-of-scope questions, and never make claims outside your content. Run a continuous eval harness to track regressions before they reach customers.
Retrieval you control
Tune retrieval per chatbot — top-K, similarity thresholds, and reranking — so each bot answers from exactly the right context. As new models ship, your bots get better without a migration.
vs Chatbase, CustomGPT, Dante AI
Other RAG chatbot tools stop at "answer questions." CustomerHero ships a full inbox, real-time human handoff, multi-channel routing, and proactive engagement on top of the same agent — at SMB pricing.