Tested, not just promised
Test results
We test every feature against real community messages before you rely on it. These are the numbers from our most recent test runs.
Sentiment accuracy
We ran 900 real community messages through the sentiment engine and compared every result against a human review. 894 were classified correctly — including Web3 slang, emoji and sarcasm, which a general-purpose model usually gets wrong.
Answer quality
Across 450 automated answers, only 2 were wrong. The bot answers word-for-word from your approved replies, and stays silent when it does not have an approved answer instead of guessing.
Question detection
Out of 4,038 messages tested, the bot correctly picked out 96.4% of genuine member questions while ignoring small talk, reactions and chatter.
How we test
Tests use real messages from live Telegram and Discord communities, reviewed by a person to establish the correct answer. We re-run these tests whenever the underlying models or prompts change, and publish the updated numbers here.