Email triage & drafts

Email is where most knowledge workers lose the most time to low-value activity. Your agent can take the triage work off your plate so you only look at the emails that actually need you.

Prerequisite: Connect your email first.

The three things your agent can do

1. Inbox summary

On demand or on a schedule, your agent reads your inbox and produces a summary like this:

Overnight activity (12 unread, 3 need attention)

🔴 Needs your response today:

• Sarah @ Acme — asking about Q2 discrepancy, 2 questions, needs answer before their board meeting tomorrow
• Jorge @ ReliableLaw — wants to reschedule Friday’s call, offered 3 time slots

🟡 Informational (FYI, no response needed):

• Weekly newsletter from Product Hunt (featured 3 AI tools)
• Stripe payout notification — $2,340 deposited

⚪ Low priority / can wait:

• 7 marketing emails, 1 GitHub notification, 1 LinkedIn connection request

You can ask for this summary whenever — “Catch me up on my inbox” — or schedule it as part of a morning briefing.

2. Draft replies

Your agent can draft responses to emails. You can ask it explicitly:

“Draft a reply to Sarah’s email acknowledging the discrepancy. Tell her I’ll have it resolved by end of day, and suggest a 15-minute call to walk her through the fix. Professional tone.”

Or you can ask your agent to draft replies to everything that needs a response:

“Draft replies to every email in the ‘needs response today’ category. I’ll review them before sending.”

Drafts go into your Approvals queue by default. You review each one, edit if needed, and hit approve to send.

3. Smart organization

Your agent can apply labels, move to folders, archive, and flag emails based on your rules. Describe the rules in plain English:

“From now on, automatically move all Stripe payment notifications to the ‘Finance’ folder. Archive anything from LinkedIn unless it’s a direct message.”

Your agent will remember this and apply it going forward (and optionally retroactively if you ask).

The approval flow

By default, every send requires your approval. This is deliberate. Even the best AI sometimes writes a reply that misses a subtle context or has the wrong tone for a specific relationship.

When you approve a draft:

• The email is sent immediately via your own SMTP server
• A copy goes to your Sent folder (like any manual send)
• Your agent logs that the message was sent and remembers the context for future replies

If you reject a draft, your agent asks why and updates its understanding — “Too formal for this person” or “Don’t commit to that date, I’m not sure yet.”

Want per-category control?

If you want different behavior for leads vs. newsletters vs. calendar — and a master on/off switch for vacations — see Email automations. That’s the full configuration reference for the 8-category matrix, scope filters, daily spend cap, and auto-send unlock.

Changing the default to “auto send”

If you build up trust, you can change the trust level for “send email” to Auto + notify or Full auto in Settings → Agent → Agent Trust & Approvals.

The trust level is a single setting per action — not per-contact. If you set Full auto for “send email,” every send goes automatically. If you set Require approval, every send waits for you. Most users keep approval required for email until they’ve built up a track record with their agent’s drafts.

If you want “auto-send for replies, approval for new threads” or similar finer-grained behavior, tell your agent in plain language and it will hold the rule in memory — but the system-level enforcement is still the single trust setting. Treat the plain-language rule as a guideline, not a guarantee.

Smart search

A lot of inbox triage is remembering what you wrote last week. Your agent makes this instant:

“When did I last email Priya about the Q3 budget?”

“Find the thread where I discussed the NDA with Acme’s legal team.”

“Did I ever promise Sarah a discount? Search for any email where I mentioned pricing with her.”

Your agent searches your inbox semantically (not just keyword match) and can return the relevant threads with summaries.

Following up

Your agent can also proactively flag emails that didn’t get responses:

“Check which emails I’ve sent in the last 7 days that haven’t gotten replies. Draft follow-ups for the ones that still matter.”

It’ll use its judgment about what “still matters” — excluding auto-responses and informational broadcasts, focusing on conversations you actually need resolved.

Privacy

Your email credentials (app passwords) are encrypted per-agent. Only your agent can read them.

Your agent does NOT:

• Copy your emails to our central servers
• Log email contents outside of your own container
• Train any AI model on your email data
• Share email content with other users

Email summaries and drafts are generated on-demand and only persist in your agent’s session memory unless you explicitly save them.

Start with read-only

For the first week, tell your agent explicitly: “Don’t send anything. Only draft. I want to see what you’d write before you’re allowed to send it.” This builds trust fast.