ProvnAI private preview

Email-to-agent workflows need a control layer.

ProvnMail turns inbound email into governed agent-visible tasks only after custody, parsing, policy admission, runtime boundaries, and evidence are in place.

Open private dashboard See the architecture

allowsafe mail can become a task

quarantinerisky mail stops before task creation

evidenceevery decision leaves a durable trail

Capability and authority are separated.

The model can reason about email. ProvnMail decides what becomes visible, what requires review, what is refused, and what evidence is preserved before an agent gets authority.

Email custody

Raw MIME is received, hashed, stored, and normalized before agent exposure.

Policy admission

A2AVanguard decides whether inbound mail may become an agent-visible task.

Evidence envelopes

Every allow, quarantine, block, and unknown-recipient path is tied to durable audit material.

Current private-preview state.

Verified live path

Spacemail SMTP, Cloudflare Email Routing, Worker forwarding, Railway, PgBoss, Postgres audit/evidence, and dashboard views are wired and tested.

Still intentionally bounded

Outbound sending, attachment release workflows, production A2AVanguard gateway calls, and long-term retention controls are explicit next-phase items.