AI agents for the conversations ahead.
Gmail, Outlook, any IMAP mailbox, Slack, Teams, Telegram, WhatsApp, Messenger, Discord, LINE and more — 29 platforms, 25 through official APIs. Nine agents triage, draft, follow up and report. One executor sends, and it checks consent, quiet hours, budgets and unsubscribe before every message.
The AI workflow
From your goal to work you can review.
You direct
Your channels and instructions
Connect the supported inboxes and set the audience, purpose and response rules.
AI works
Triage → drafts → follow-ups
AI agents sort incoming messages, prepare replies and sequences, and hand supported sends to the shared executor.
You inspect
An outbox you can review
Inspect proposals and reports. Sending follows the connection, consent, budget and automation controls.
AI agents draft. One executor sends.
The AI inbox triage agent sorts what arrived; the AI support operator and sales follow-up agent draft replies; the email marketer writes sequences and newsletters. None of them can send. Every outgoing message becomes a row that the one executor sends — and broadcasts and marketing mail can never run on autopilot: a person approves them.
Hygiene is code, not a prompt
Automated mail is never answered. Threads follow Message-ID, never sender-plus-subject. Every follow-up and broadcast carries one-click unsubscribe; bounces and complaints suppress the contact; SPF, DKIM and DMARC are checked before any bulk send; deliverability pauses at 0.3 % complaints or 5 % bounces. Per-platform daily budgets sit below the published limits.
Your accounts, on the platform's terms
Connect with OAuth or a token; secrets are encrypted and decrypted in exactly one place. Slack's inbox is event-driven, not polled. Personal accounts are never driven through unofficial clients — iMessage, Signal, Zalo and KakaoTalk stay in the human lane: agents draft, you send.
How it runs
Connect
One card per platform in Settings; several accounts per platform is normal.
Triage
Inbound is classified, automated mail is filtered out, threads are rebuilt.
Draft and approve
Replies and follow-ups can run on autopilot; broadcasts always wait for you.
Report
Daily analytics per platform: sent, replied, bounced, unsubscribed.
Questions, answered
Will it send marketing email on its own?
Never. Broadcasts and marketing sends are forced to proposal status; only replies, follow-ups and group actions can be set to autopilot, and even those pass the consent and quiet-hours gates first.
Which platforms are automated?
25 of 29 through official APIs — Gmail, Microsoft 365, IMAP/SMTP, Resend, Postmark, SES, SendGrid, Slack, Teams, Google Chat, Matrix, WeCom, Feishu, DingTalk, Telegram, WhatsApp, Messenger, Instagram, Discord, LINE, Viber, Twitch, Intercom, Crisp, Zendesk. Signal, iMessage, Zalo and KakaoTalk are draft-only.
Where do my tokens live?
Encrypted per tenant, decrypted by one function at send time, never logged and never returned over the API. Webhooks verify the vendor signature before anything is stored.
Why might an approved message still be waiting?
Approval is one step. The outbound executor also checks consent, quiet hours, unsubscribe status and sending budgets. A provider rate limit can defer a message rather than send it immediately.
What happens when someone replies to a sequence?
A reply can stop the contact’s active sequence enrollment so follow-ups do not continue as though nobody answered. Review the conversation and the sequence status before deciding on the next message.
How do support drafts get the right context?
The inbox workflow rebuilds conversations into threads and uses the available messages and your instructions. Give the team accurate product and support context, then review replies that need a decision or additional information.
Answer everyone. Spam no one.
Twenty-nine platforms, one executor, every gate in code.