Inbox Manager — triage, draft, and route email in your own voice
A long-lived email agent for one mailbox. It reads incoming threads, recalls what it knows about the sender, drafts replies that sound like you, and routes everything via labels / archive / Slack — and it never sends without your explicit approval.
Build this with agnt_
Skip the copy-paste. We'll spin up a builder session prepopulated with this blueprint's spec — providers, schedule, database schema, and the questions the agent should ask you to personalize it for your product.
Sign up free · no credit card
The motion
The Inbox Manager is a persistent agent attached to a connected Gmail or Outlook mailbox. For every incoming thread it pulls the full conversation, recalls memory about the sender and related threads, and decides what to do: label and archive low-value mail, draft a contextual reply in your own writing voice, or ping you on Slack. Replies are drafted but never sent without your explicit sign-off — approvals run through an in-app review or a Slack approve/decline button, and the actual send is a deterministic dispatch only after you say yes. Your writing voice is learned per mailbox at onboarding and injected at runtime, so the agent matches how *you* write (work vs personal can differ) — nothing about your tone is baked into the agent itself. It remembers contacts, communication patterns, and thread outcomes in a workspace memory shared with your other agents.
The agent learns how you write per mailbox at onboarding and drafts replies that match — greeting, tone, length, sign-off. No tone is baked in.
Every outbound reply is shown to you in full and gated behind an in-app review or a Slack approve/decline before it ever leaves your account.
Labels, archives, and Slack pings happen per your triage rules so only the threads that need you reach you.
Contact profiles, communication patterns, and thread outcomes land in a workspace memory shared with your other agents.
Connect a mailbox; the agent triages every incoming thread — labeling, archiving, drafting voice-matched replies, and escalating decisions to you. Sends are always gated behind your explicit approval.
Build this with agnt_
Skip the copy-paste. We'll spin up a builder session prepopulated with this blueprint's spec — providers, schedule, database schema, and the questions the agent should ask you to personalize it for your product.
Sign up free · no credit card
Or copy a prompt into another platform
Prefer to build with OpenClaw, Hermes, or Claude Code? Drop this prompt into your agent of choice — it seeds the goal, the agntdata endpoints to use, and a step-by-step plan.
Help me build an Inbox Manager — a long-lived email agent that triages one mailbox, drafts replies in the user's own voice, and never sends without explicit human approval.
REFERENCE DOCS
- agntdata API documentation: https://agntdata.dev/docs
WHAT TO BUILD
- A persistent (event-driven) agent attached to a connected Gmail or Outlook mailbox.
- It reads incoming threads, recalls memory about the sender, drafts a reply in the mailbox's writing voice, and either saves a draft or requests approval before any send.
- It labels / archives / pings Slack per the user's triage rules.
NON-NEGOTIABLE
- Never send an email until a human has seen the exact outgoing message and approved it. Approvals run through escalate_to_human or slack_request_approval (asynchronous — request, then end the turn).
- Voice is NOT baked into the agent. The mailbox's writing voice is injected at runtime; the agent matches it.
TOOLS
- Email: connection_gmail_* / connection_outlook_* (get_thread, get_message, search_messages, save_draft, send_email, reply_all, forward_message, label_message, list_labels, archive_message).
- Slack: slack_post_message, slack_request_approval.
- Builtin: memory_remember/recall/search/forget, webhook_send, escalate_to_human.
When you're ready, connect a mailbox and let the onboarding screening derive the voice profile + triage rules.Paste into OpenClaw to scaffold this agent. Tweak the inputs and goal at the top of the prompt.
How to build it
5 steps. Each one links to the underlying agntdata endpoints — open them in a new tab to inspect parameters and pricing as you build.
Authorize Gmail or Outlook. The first connect deploys one Inbox Manager for the workspace; later mailboxes attach to the same agent.
A pass over recent sent mail synthesizes your per-mailbox writing voice and a set of starter triage rules — editable before they apply.
Slack for approvals and FYIs.
Let the agent triage a handful of live threads and review its drafts + routing before turning it loose.
The agent runs continuously, triaging every incoming thread. Replies always wait for your approval.
Ship this blueprint today
One click spins up a builder session prefilled with this blueprint's spec. We'll ask you a handful of personalization questions, then generate the agent.
Related blueprints
Browse all →Hand any X username to this agent and get back a qualified, ICP-scored lead with a verified email and a resolved LinkedIn profile.
Hand any LinkedIn profile URL to this agent and get back a qualified, ICP-scored lead with a verified email and a website summary attached.
Give it a name + a domain (or a LinkedIn URL, or any combo of identifiers) — it returns a verified professional email or a structured failure. Three-step waterfall: agntdata orchestrator → Exa deep search → verifier. Stateless primitive, called by other agents.
Unlock the full build guide for Inbox Manager — triage, draft, and route email in your own voice
Step-by-step instructions, copy-paste prompts, and the endpoints this agent calls. Drop your email to read the rest.