Guide

Draft Personalized Emails with Manus AI

Use Manus AI to draft personalized outbound emails at scale. Provide context and a contact list, review AI-generated drafts, and send with the Nylas CLI across all major email providers. This workflow covers sales outreach, follow-ups, and newsletters.

Written by Aaron de Mello Senior Engineering Manager

Reviewed by Hazik

Verified — CLI 3.1.1 · Gmail, Outlook · last tested April 11, 2026

Why use Manus for email drafting

Manus AI drafts personalized emails by reading context you provide — previous threads, company details, or a template — and generating a tailored message for each recipient. According to McKinsey's 2023 report on generative AI, professionals spend roughly 28% of their workweek on email. Manus reduces per-email drafting time from several minutes to seconds while keeping a human review step before anything gets sent.

The Nylas CLI provides the execution layer. Manus calls nylas email smart-compose to generate each draft and nylas email send to deliver approved messages. The agent handles iteration across contacts; the user handles judgment on tone and accuracy.

Prerequisites

Two things are required before Manus can draft and send emails through the Nylas CLI: a Manus account with the Nylas Skill installed, and at least one authenticated email grant. Setup takes under 5 minutes if both accounts already exist.

  1. A Manus account with the Nylas CLI Skill installed. The Manus AI Skills setup guide walks through Skill installation.
  2. Nylas CLI authenticated with at least one connected grant. Run nylas auth whoami to confirm the grant is active.

Draft a single email

A single-email draft is the simplest Manus workflow: describe who the email is for and what it should say, and Manus generates one draft using nylas email smart-compose. The entire round-trip — prompt to rendered draft — typically completes in under 10 seconds. No email is sent until you explicitly approve.

In the Manus chat, type a natural-language instruction like:

"Draft a follow-up email to Sarah about the budget meeting."

Manus converts the natural-language request into a CLI command. The --prompt flag accepts freeform text that the Nylas AI layer uses to generate the draft body. Manus fills in the prompt based on your instruction:

nylas email smart-compose --prompt "Polite follow-up about the Q4 budget discussion. Ask for the updated spreadsheet."

The CLI returns a draft body as plain text. Manus displays the draft for review — nothing is sent at this stage, so you can request edits or discard the draft entirely.

Draft from context

Context-aware drafting means Manus reads an existing email thread before composing a reply, so the generated text references specific details from the conversation. According to Lavender.ai's 2024 email benchmark, personalized emails that reference prior context see 2.5x higher reply rates than generic outreach. Manus achieves this by chaining two CLI commands: a read, then a compose.

In the Manus chat, describe what thread to read and what the reply should say:

"Read my last email from Bob and draft a reply accepting his proposal."

Manus first retrieves the original message using nylas email read. The --json flag returns structured output that Manus parses to extract the sender, subject, body, and timestamps from the original thread:

nylas email read MESSAGE_ID --json

With the original message content in memory, Manus then calls smart-compose with a prompt that references specific details from Bob's email. The Nylas AI layer uses both the prompt text and any referenced context to generate the reply:

nylas email smart-compose --prompt "Reply to Bob accepting his proposal. Reference the timeline he mentioned and confirm the March 20 start date."

This two-step pattern — read then compose — produces drafts that reference dates, names, and details from the original thread rather than relying on generic phrasing.

Batch drafting

Batch drafting lets Manus generate one personalized email per contact from a list you provide in the chat. Manus processes contacts sequentially, calling nylas email smart-compose once per recipient with a tailored prompt. Each draft is shown individually for approval before the agent moves to the next contact. Batches of 10-20 contacts work best — small enough to review carefully, large enough to save meaningful time compared to manual writing.

In the Manus chat, paste the contact list with enough detail for personalization:

"For each of these 5 contacts, draft a personalized intro email mentioning their company and role: Sarah Chen (VP Engineering, Acme Corp), James Park (CTO, Bolt Labs), ..."

Manus generates one smart-compose call per contact, embedding the recipient's name, title, and company into each prompt. A 20-contact batch typically completes in 3-4 minutes including review pauses. For lists over 50 contacts, split them into separate batches to stay within Manus credit limits and keep review quality high.

Review and send

Sending happens only after you approve a draft in the Manus chat. Manus shows the full email body, subject line, and recipient, then waits for explicit confirmation. Once approved, Manus executes the send command through the Nylas CLI. The median time from approval to delivery is under 2 seconds because the CLI connects directly to the provider's API.

The --yes flag on the send command skips the CLI's built-in interactive confirmation prompt, which would hang in the Manus sandbox environment. Human review happens at the Manus level instead — you see the draft and type "yes" before Manus runs the command:

nylas email send --to "sarah.chen@acme.com" --subject "Quick intro — Nylas + Acme" --body "Hi Sarah, ..." --yes

The --to, --subject, and --body flags are all populated by Manus from the approved draft. The CLI delivers through whatever provider the authenticated grant uses — Gmail, Outlook, Yahoo, or IMAP — without any provider-specific code.

Templates

Templates let you define a reusable email structure with placeholder variables like {{FIRST_NAME}} and {{COMPANY}}, and Manus fills them in per contact. Unlike simple mail-merge tools, Manus also adjusts phrasing and tone around the placeholders so each email reads naturally. A template with 4-5 placeholders typically produces output that varies by 30-40% between recipients after Manus fills in context-specific details.

Paste the template directly in the Manus chat or reference a file. The placeholder names are freeform — Manus maps them to whichever contact fields you provide:

Subject: {{FIRST_NAME}}, quick question about {{COMPANY}}

Hi {{FIRST_NAME}},

I noticed {{COMPANY}} recently {{RECENT_EVENT}}. We help teams like yours
{{VALUE_PROP}}.

Would you be open to a 15-minute call this week?

Best,
[Your name]

Manus replaces each placeholder based on the contact details you provide and rewrites connecting sentences so the emails don't read like mail-merge output. The result is closer to a hand-written email than a find-and-replace fill.

Safety

Safety in this workflow depends on one rule: Manus must never send an email without explicit user confirmation. The SKILL.md file — which defines Manus's behavior — should contain an instruction like "Always show the draft and wait for approval before sending." According to Anthropic's 2024 responsible AI guidelines, human-in-the-loop review is the single most effective safeguard for AI-assisted communication. Three controls enforce this in the Manus email workflow.

  • Human-in-the-loop: Manus shows each draft and waits for explicit "yes" before sending. If you say "no" or request changes, Manus revises the draft and presents it again.
  • The --yes flag: This flag bypasses the CLI's built-in interactive confirmation, not the user's approval in Manus. The flag exists because interactive terminal prompts hang in the Manus sandbox environment. User approval at the Manus level still gates every send.
  • Batch safety: For batch sends, review at least the first 2-3 drafts carefully to catch tone or accuracy issues. If quality is consistent across the first few, subsequent approvals can be faster.

FAQ

These are the most common questions about using Manus AI for email drafting with the Nylas CLI. The answers cover batch limits, personalization quality, and the safety model that prevents unsupervised sends.

How many emails can Manus draft in one batch?

There is no hard limit on drafts per batch. Manus processes each contact sequentially, calling nylas email smart-compose once per recipient. Practical limits depend on the user's Manus credit balance and Nylas API rate limits (typically 5 requests per second per grant). For lists over 50 contacts, splitting into batches of 10-20 keeps review quality high and avoids credit exhaustion.

How good is the personalization quality?

Personalization quality depends directly on the context provided. Giving Manus access to previous email threads, company details, or a detailed template produces measurably better output than a one-line generic prompt. The smart-composecommand uses the Nylas AI layer, which is tuned for professional email tone and typically produces drafts that need fewer than 2 edits per message.

Will Manus send emails without my approval?

Not if the SKILL.md file includes an explicit instruction to always confirm before sending. Manus follows the rules defined in its Skill file. The review step is built into the workflow — Manus shows each draft and waits for approval before running the nylas email send command. Without a "yes" response, no send command is executed.

Next steps