mail-triage-agent

Mail Triage Agent is a small, self-hosted Python agent that watches an inbox for emails from chosen senders or containing chosen keywords, analyzes each match with an LLM, and writes a report plus a draft reply for the user to review, edit, and send. Filtering is configurable, and the analysis output includes a summary, category, priority, and suggested action for each email.

It works with standard IMAP/SMTP mail providers and lets the user bring their own AI provider and API key, including Anthropic Claude, OpenAI, Google Gemini, or any OpenAI-compatible API. It is a plain Python script run on the user's own machine, and credentials stay in a local .env file.

Before drafting anything, the agent runs a deterministic sender trust check covering SPF/DKIM/DMARC authentication, Reply-To versus From mismatches, and display-name brand-impersonation heuristics. The project describes this check as a heuristic rather than a guarantee, reading whatever the mail server already reports.

Draft replies are plain text files that the user reads and edits; nothing is sent until the user runs the send command. A local state file records already-processed messages so repeated runs do not create duplicate drafts.