Cost
Cache hit rate, per bot
A cold cache costs more than no cache at all, so we only pay the write premium when a read is plausible. You can see whether it is working.
Product · Insights
Every conversation is logged, every gap is ranked, and every message records what it cost. The most useful thing a support bot does is tell you what your documentation is missing.
100 free messages · No card required
Unanswered questions are ranked by how often they are asked. That list is the most honest documentation backlog you will ever be handed.
Tokens in, cached tokens, tokens out, latency and exact cost, on every single message. The economics of your own chatbot are never a mystery.
tokens_in 8,412
cached_tokens 8,398 (99.8%)
tokens_out 146
cost_usd $0.000061
latency 1,204 ms
Nothing is deleted until you say so. Set a retention window and older conversations remove themselves every night; export everything whenever you like. It is your data, held on your instructions.
Cost
A cold cache costs more than no cache at all, so we only pay the write premium when a read is plausible. You can see whether it is working.
Quality
A rising refusal rate means your knowledge base has drifted behind your product. That is a signal, not a failure.
Privacy
Visitors are identified by a salted hash of IP and user agent, never the raw address. It is what lets rate limiting survive a deletion request.