Chat Data

Chat Data is the customer support AI agent platform from Chat Data LLC: it feeds documents, web pages and audio/video into a RAG knowledge base, then deploys it across a website and messaging channels. GPT-4o, Claude and Gemini are all third-party models; what is built in-house is the retrieval, correction-capture and compliance layer. It is not the same company as chat4data in our directory.

Core capabilities

• Knowledge base training: upload PDFs, crawl web pages, connect Notion, with limits at both the per-Agent and account level
• 8 messaging channels: WhatsApp official and WhatsApp Web unofficial as two routes, plus Messenger, Instagram, Telegram, LINE, Slack, Discord and a website widget
• In-house RAG layer: each answer pulls only the 5 most relevant text chunks, with optional rerank; fix an answer once in Chat Logs and it is captured as a Q&A pair
• Human handoff: native escalation with context from Standard up, available as an add-on on lower tiers
• Workflows and voice: conditional branches, scheduling, real-time voice, and AI phone calls using your own Twilio
• Compliance and white-label: Standard includes a free HIPAA BAA; white-label and SSO are sold as add-ons

Where it fits Suited to small and mid-size agencies, cross-border independent-store support teams and small clinics that must sign a BAA: a budget in the low hundreds while needing channel breadth and compliance. Not suited to large enterprises that want minimum liability exposure, and teams in China still get no Chinese UI.

What sets it apart Chatbase is cheaper and Intercom is more complete; it sits in between: channel breadth, white-label resale, and a BAA included on the entry tier. Its DPA unusually names the Swiss OCI host, Azure OpenAI in Sweden and PostHog EU, and it openly admits that DeepSeek via OpenRouter follows a retention policy outside its own control.

Editorial take The biggest honest point: the free tier maps to the weakest basic models, and the official cookbook itself admits GPT-4o mini hallucinates and tells you to switch to a stronger model when answers are wrong. Third-party samples are thin: G2 has just 1 review, Trustpilot has 14 reviews at 3.3, and Reddit has zero discussion; the team is 2–5 people with MRR of roughly 17,600 USD. Worth buying if you have human fallback; avoid it if you do not.