Incoming email is classified, routed and turned into chains of actions automatically. No manual sorting.
FAQ
How is this different from mail rules?
Mail rules fire on rigid conditions ("if the subject contains X"). TaskFlow AI reads the meaning: it extracts company, contact, amount, tone and intent, then acts — creating a deal, task, ticket or escalation. It is not a filter but 9 AI agents that parse a letter the way an employee would.
Can it run on-premise or in a closed perimeter?
Yes. TaskFlow is deployed via Docker Compose on your server (16+ CPU / 64 GB RAM) in 2 weeks. With a local LLM (Qwen, LLaMA) it runs fully offline, with no internet access.
Where is data stored and who can access it?
All data and email stay on your infrastructure in single-tenant mode — one instance per customer. Mailbox credentials are encrypted (Fernet), access is split across 4 roles via OIDC/SSO, and every action is written to an audit log.
Can several mailboxes be connected at once?
Yes. Exchange (EWS), IMAP, POP3 and SMTP are supported — personal and shared mailboxes, processed in parallel.
What happens if the AI gets it wrong?
Every AI action is logged and can be reverted. SLA escalations fire when the AI is not confident. An employee can re-route a letter manually, and that becomes a training example for the system.
How long does deployment take?
The first pilot is up in 2 weeks. Extending connectors and tuning to your processes takes another 4–6 weeks.