Instead of an abstract feature list: here's what SupportFlowAI actually looks like in typical situations.
A customer writes "My export to Excel has been failing since the update." SupportFlowAI recognizes the thread, searches the knowledge base, and finds three similar, already answered requests from the mailbox archive. The local language model drafts a reply with a confidence of 0.71 — below the full-automation threshold. The draft lands in the review queue, a support agent tweaks two sentences and approves it. Only now is the reply actually sent.
A very frequently asked question ("How do I reset my password?") comes in. The knowledge search finds a nearly identical, often-confirmed answer — confidence 0.93, above the configured threshold of 0.85. Since full automation is enabled for this account (previously confirmed via checkbox and re-entry of the password by the system administrator), the reply is sent without any manual intervening step — both the activation and this automatic send appear in the audit log.
A team is migrating from an old mail system and has three years of support history as
an .mbox export. Instead of a live IMAP account, a
file-import account is set up with the archive directory — SupportFlowAI searches it
recursively, builds question/answer pairs from it, and populates the knowledge base, all
without any ongoing mailbox access. Live operation remains disabled for this account.
After a few months of operation, it becomes apparent that several question/answer pairs
for the same outdated product version are still showing up actively in knowledge search.
Via /Admin/Curation, duplicates and outdated entries are surfaced
and cleaned up in a targeted way — the knowledge base stays current without needing to be
rebuilt from scratch.
The Getting Started guide walks you through the complete initial setup.
Open Getting Started →