Skip to main content

Email Ant

Turn client requests into careful, reviewable email drafts

Always human
Review policy
None
External actions
Up to 8
Knowledge notes

Current ANTS capability boundary, not an automation or performance claim.

The Problem

Client email is repetitive, but it is rarely low-stakes. A short reply can accidentally promise a date, imply that work is complete, or miss context from an earlier conversation. Drafting from scratch also forces people to switch repeatedly between the incoming message, account notes, policies, and their own writing standards.

The useful opportunity is not unattended sending. It is a faster, more consistent first draft that keeps its evidence visible and leaves the final judgment with a person.

How the Email Ant Works

The current product is a local-first, manual workflow. It has no inbox connection and performs no autonomous action.

1. Configure a narrow job

In Ant Studio, name the Email Ant, describe its role and purpose, choose a tone, add drafting instructions, and optionally add a signature and small trusted knowledge notes. A fixed policy requires human review and cannot be changed.

2. Add the request and context

Paste a client request into the request desk and add only the context you are permitted to use. The request, edits, decisions, Ant settings, and knowledge stay in browser storage unless you explicitly confirm optional AI drafting.

3. Generate or write a proposal

You can write locally without an AI provider. If secure drafting is configured, ANTS shows the exact request and Ant configuration that would be submitted to OpenAI before generating a structured proposal. The endpoint cannot send email or create a task.

4. Review every detail

Edit the reply and any follow-up proposal, verify the visible evidence, and record an approval or rejection locally. Approval is a review state only; it does not perform the proposed action.

What Makes It Different

  • Supervised by design: Every draft returns to a person and the approval policy is locked.
  • Local-first control: Browser storage, JSON export, and clear/reset controls work without an AI provider.
  • Evidence boundaries: Knowledge and request content are treated as untrusted reference material, never as permission to take action.
  • Honest capability: The current version drafts proposals only. It does not read an inbox, send messages, book meetings, or create tasks.

Expected Results

A useful pilot
Start with a small set of low-risk client requests. Review every draft, record what you changed, and compare drafting time, factual corrections, and tone consistency against your normal process. Keep mandatory review in place while you learn where the Ant helps and where it needs better context.

A good outcome is not an impressive automation percentage. It is a workflow that makes routine drafting faster without hiding uncertainty, inventing commitments, or weakening human accountability.

Try Email Ant locally

Add a request manually, prepare an editable reply, and keep sending outside ANTS.

Open Request Desk