The problem is not “an auto-replier.” The problem is volume: what matters, what needs a
reply, what can wait. Mellissa uses the same Composio Gmail OAuth as InboxPurge,
pulls new messages on a systemd timer (~5 min, per-user polling_minutes default 10),
scores 0–100 importance and actionability with a one-sentence reason, applies
your rules (floor/ceiling/mute), and writes reply drafts for high-priority mail.
Every draft is human-reviewed. Mellissa never auto-sends — you approve, then you send. That is the opposite of
a silent agent. InboxPurge unsubscribes senders; Mellissa triages threads. Do not merge the
landings. Public /mellissa/; app login. platform MiniMax path; dry keys → no drafts,
not a fake send.
If a control looks like Send, it is a bug. ReplyCraft is a paste-draft notepad; this is inbox-connected.
Not Gmail. Not InboxPurge’s sender table.
Floor/ceiling/mute. The LLM is not the policy.
Two apps, one OAuth vendor. Different jobs.
If you wanted auto-send, this README refuses you.
Will not unsubscribe-as-a-side-effect. Will not use a second Google Cloud app.
Not ReplyCraft paste. Not Pad.
Cards match Mellissa README implemented features.
Login for /mellissa/app. This page is not your inbox.
Open MellissaRequires a MelTuc account. Create one free.
AI Gmail triage — score, classify, draft replies with human approval. Shared Composio Gmail with InboxPurge. Health stays cheap. If this page disagrees with the signed-in app, trust the app — the product contract and tests enforce that rule.