HermesAgentMail logoHermesAgentMail
Features

Everything your agents need to run email safely.

HermesAgentMail is more than an inbox API — it is the operations layer around agent email: routing, approvals, policy, MCP tools, and full auditability.

The problem

Agents need more than inboxes.

Giving an AI agent an email address is easy. Trusting it with customer communication is harder. HermesAgentMail adds the missing operations layer: routing, permissions, approvals, memory, task creation, and visibility.

Inbox APIs stop too early

Most email APIs deliver the message. HermesAgentMail helps decide what the agent should do next.

Autonomous replies create risk

Agents need policy boundaries: who they can reply to, when approval is required, what attachments are allowed, and how much they can send.

Teams need visibility

Operators need to see pending approvals, agent decisions, created tasks, failed sends, and audit logs in one place.

Workflow engine

The workflow layer your agents need.

HermesAgentMail turns raw email into structured work.

Intent classification

Detect support requests, sales leads, invoices, bug reports, scheduling requests, newsletters, and urgent issues.

Agent routing

Route messages to the right Hermes agent based on address, sender, workspace, keywords, or policy.

Memory lookup

Let agents retrieve approved context before drafting a reply.

Task creation

Turn important emails into follow-up tasks, finance tasks, support tasks, or engineering tasks.

Approval policies

Require human approval for first-time senders, risky topics, external recipients, attachments, or high-value actions.

Agent collaboration

Allow one agent to ask another agent for help before a reply is sent.

Live activity feed

Show email events, agent decisions, tasks, and approvals in one live, auditable activity feed.

Audit trail

Record what was received, who or what handled it, what was drafted, who approved it, and when it was sent.

Features

Everything agents need to handle email safely.

Programmable primitives for routing, approving, tasking, and sending — built for controlled agent workflows, not human inbox clutter.

Controlled agent addresses

Claim addresses like support@hermesagentmail.com, luke@hermesagentmail.com, or finance@hermesagentmail.com and connect each one to an agent workflow.

Human approval by default

Agents can draft and prepare replies while humans stay in control of what leaves the workspace.

Policy-based sending

Control who agents can email, when they need approval, how often they can send, and whether attachments are allowed.

Email-to-task workflows

Convert inbound messages into support tickets, finance reviews, sales follow-ups, research summaries, or engineering tasks.

Hermes-native routing

Wake the right Hermes agent from inbound email and route the message with structured context.

MCP-controlled tools

Expose safe tools like get_email, search_thread, draft_reply, request_approval, send_reply, mark_handled, and create_task_from_email.

Queue-first reliability

Accept inbound events quickly, queue work safely, and process asynchronously with rate limits.

Workspace isolation

Keep each customer, workspace, agent, address, API key, and message scoped correctly.

Audit logs

Track message reads, drafts, approvals, sends, rejected actions, webhook events, and API usage.

Delivery tracking

Monitor queued, sent, delivered, bounced, failed, complained, and suppressed messages.

Secure attachments

Control attachment download, storage, scanning, size limits, and agent permissions.

Usage and abuse limits

Protect customers and the platform with inbound, outbound, per-agent, and per-workspace limits.

Architecture

Built for bursty email and controlled agent action.

HermesAgentMail does not process AI replies inside the inbound webhook. The webhook is only the front door. It verifies the provider event, stores metadata, checks idempotency, enqueues work, and returns quickly. Workers handle retrieval, storage, routing, agent execution, approvals, and outbound delivery.

Email received→
Verified webhook→
Idempotency check→
Durable queue→
Worker processing→
Message storage→
Intent classification→
Agent router→
Memory / context lookup→
Draft or task→
Policy engine→
Human approval→
Outbound queue→
Send provider→
Delivery event→
Activity update→
Audit log
Policy engine

Agent autonomy with boundaries.

Give agents the ability to work without giving them unlimited permission to send.

Support Agent
support@hermesagentmail.com
  • Can read inbound support email
  • Can draft replies
  • Requires approval for first-time senders
  • Cannot send attachments
  • Max 50 outbound emails/day
  • Auto-reply allowed only for trusted domains
Finance Agent
finance@hermesagentmail.com
  • Can receive invoices
  • Can extract invoice details
  • Can create approval tasks
  • Cannot send payment instructions
  • All outbound replies require approval
Research Agent
research@hermesagentmail.com
  • Can receive newsletters and reports
  • Can summarize and store memory
  • Can create research notes
  • Cannot send external replies automatically
Sales Agent
sales@hermesagentmail.com
  • Can qualify inbound leads
  • Can draft follow-ups
  • Requires approval for pricing promises
  • Auto-reply disabled for new domains

MCP-ready for agent runtimes.

Agents do not get raw SMTP credentials or unrestricted inbox access. They call controlled tools. HermesAgentMail enforces workspace boundaries, policies, approvals, usage limits, and audit logs.

get_email(message_id)search_thread(thread_id)search_memory(query)classify_email(message_id)draft_reply(message_id)request_human_approval(draft_id)send_approved_reply(approval_id)create_task_from_email(message_id)assign_to_agent(message_id, agent_id)mark_handled(message_id)get_agent_policy(agent_id)check_send_permission(agent_id, recipient)

Built around Hermes workflows.

HermesAgentMail is designed for Hermes agents, not just generic automation. An email can wake the right agent, provide structured context, search memory, create a task, request approval, and log every step as the workflow changes.

  1. 1Email to support@hermesagentmail.com
  2. 2HermesAgentMail receives and queues
  3. 3Policy selects Support Agent
  4. 4Agent searches memory
  5. 5Agent drafts reply
  6. 6Task created
  7. 7Human approves
  8. 8Reply sent
  9. 9Activity feed updated
  10. 10Audit log written
Use cases

Agent email workflows for real teams.

Claim root-domain addresses and point them at the agent workflow that should handle the work.

Customer support workflow

support@hermesagentmail.com

Classify customer questions, draft replies, create support tasks, and require approval for sensitive answers.

Founder operator workflow

luke@hermesagentmail.com

Let an operator agent triage requests, summarize threads, create tasks, and prepare replies for approval.

Finance workflow

finance@hermesagentmail.com

Receive invoices and receipts, extract details, create approval tasks, and prevent unauthorized payment replies.

Research workflow

research@hermesagentmail.com

Collect newsletters, reports, documents, and requests, then summarize them into searchable memory.

Sales workflow

sales@hermesagentmail.com

Handle inbound leads, qualify requests, draft follow-ups, and require approval before pricing commitments.

Engineering workflow

dev@hermesagentmail.com

Turn monitoring alerts, GitHub notifications, bug reports, and customer issues into engineering tasks.

See it work end to end.

Create root-domain agent addresses, route inbound messages into workflows, require approvals, and make every action visible.

Read the API docs