Email automations

Your agent can do a lot with email on its own — classify every incoming message, flag the ones that matter, draft replies, and eventually send them without you. This page is the single reference for how all of that is configured.

Prerequisite: Connect your email first. Automations only fire against mailboxes you’ve connected.

Where to configure it

Settings → Email automations (/dashboard/settings?tab=email-automations).

Everything on this page is controlled from there. The tab has two cards:

Email automation — the master switch, the 8-category matrix, the daily spend cap.
Contacts — the list your agent uses to decide who counts as “a contact” for scope filters.

The email account connection itself lives in Settings → Integrations → Email Accounts — that’s a prerequisite, not an automation setting.

The master switch

At the top of the automation card: Email automations: On / Paused.

One click pauses every automation without touching your per-category config. Use it when:

You’re on vacation and don’t want the agent replying on your behalf
You’re in a stretch of focused work and want zero surprise sends
You’re investigating something weird and want a clean slate

While paused:

Classification keeps running. Incoming mail is still sorted into categories so your insights stay current and the 48h retroactive window works the moment you flip back on.
Nothing auto-drafts or auto-sends. Every category behaves as if it were set to Observe.
Your settings are preserved. Flipping back to On restores every per-category authority/scope/state exactly as you left it.

The eight categories

Every incoming email is classified into one of eight categories:

Category	What it’s for
Inbound Leads	People reaching out about your product or service
Calendar	Meeting invites, RSVPs, reschedules
Receipts	Purchase confirmations, invoices, payment status
Personal	One-on-one threads with known contacts
Newsletters	Subscribed content, digests, marketing
Support	Customer service, helpdesk, complaints
Automated	System notifications, OTPs, shipping updates
Unclassified	Everything else — cold outreach, ambiguous mail

Each category has its own row in the matrix with three controls: state, authority, and scope.

State: off / observe / active

Off — The agent doesn’t look at messages in this category at all.
Observe — The agent reads and classifies, but takes no action. You can see insights in the dashboard. This is the default for every category on a fresh account — safe and zero-risk.
Active — The agent applies whatever authority level you’ve set for this category.

Authority: what “active” means

When a category is Active, authority decides what the agent actually does:

Authority	Behavior
Observe	Same as the state above — no action, just visibility
Draft only	Writes a draft and saves it to your email Drafts folder. You send it yourself when (and if) you want.
Draft + review queue	Writes a draft and puts it in Approvals. You approve/edit/reject.
Auto-send	Writes and sends without asking. Gated by the unlock criteria below.

Calendar has its own ladder (Observe → Analyze & flag → Draft RSVP → Auto-respond) because “send a reply” doesn’t map cleanly to RSVP behavior.

Scope: who counts

Scope filters the senders the agent will act on:

All mail — everyone
Contacts only — addresses in your Contacts list (see below)
CRM contacts only — contacts you’ve flagged as lead, client, or prospect
Custom rule — allowlist/blocklist of domains or senders

Defaults are conservative: Inbound Leads, Personal, and Support start in Contacts only when you activate them. Calendar, Receipts, Newsletters, Automated, and Unclassified default to All mail.

The 48-hour retroactive window

When you flip a category from Observe to Active, the agent looks back 48 hours and acts on any message from that window that still needs a reply. Older messages stay in the activity log but don’t trigger drafts.

This is deliberate. Without the window, activating a category could fire off drafts for a six-month-old thread that’s no longer relevant. With the window, you get immediate payoff for turning a category on, without surprises.

Your agent tells you how many messages the window will cover before you flip the switch.

Contacts

The Contacts card is the source of truth for scope filters.

Sources (in order of strength):

Sent-folder scan — addresses you’ve emailed twice or more (auto-populated)
Manual mark-as-contact — click “Mark as contact” on any inbox message
CSV upload — bulk import (columns: email, name, organization, crm_status)
CRM sync — if you have a CRM integration configured

Each contact row has an optional CRM status (lead, client, prospect) that feeds the CRM contacts only scope. Addresses you’ve only emailed once show up as a greyed-out “suggestion” — click to promote, or ignore.

Removing a contact doesn’t delete their messages, it just removes them from scope allowlists.

Daily spend cap

The cap is a hard stop on agent-initiated drafting and sending per day. If the agent hits the cap mid-day, it pauses all automations until the UTC day rolls over.

What each action costs

Action	Cost
Draft reply	~1¢ per draft, occasionally 2¢ on long threads. Capped at 600 output tokens.
Classify	~0.03¢ averaged. Most incoming mail is free — thread inheritance and regex rules handle common cases. Only ambiguous messages hit the LLM.
Send	Free. Pure SMTP, no LLM.

Rough math: at $5/day, the agent can generate roughly 500 drafts before the hard stop. At a Pro-tier max of $15/day, roughly 1,500 drafts.

The cap only bounds drafting and sending (the actions the agent takes for you). Classification cost rolls up into your overall usage budget, not into this cap — see Usage for actuals.

Tier maximums

Tier	Default cap	Max cap
Trial	$0.30	$1.00
Standard	$0.50	$3.00
Pro	$1.50	$15.00

You can raise or lower within your tier at any time.

The auto-send unlock gate

Authority level Auto-send is the only setting that lets the agent put mail in your Sent folder without you clicking approve. Because that’s a genuinely scary action, it’s gated by a five-criterion check that every draft must pass individually:

≥ 50 approved drafts in this category — shows the agent understands your voice
< 5% edit rate on your recent approvals — you’re accepting drafts mostly as-is
Sender is a known contact — never auto-send to strangers
Classifier confidence ≥ 0.90 — high confidence the category is right
Explicit opt-in — you set this category’s authority to Auto-send

Until the criteria pass, the draft auto-demotes to Draft + review queue. You don’t need to do anything — the agent quietly funnels these to Approvals until it has earned the trust to go direct.

”Show me a sample” before you commit

Every category row has a Show me a sample button. Click it and the agent picks a recent message in that category, runs the real drafter against it, and shows you what the draft would look like — without creating an approval or sending anything.

Use this to sanity-check a category before turning it on. If the sample draft is off-tone or confused, the agent isn’t ready for that category yet.

When activating a category, you’ll see a confirm dialog

If you’re moving a category from Observe to Active and there are messages in the 48-hour window, you’ll get a confirmation like:

Activating Inbound Leads

~14 inbound_leads emails in last 14d. Only messages from the last 48h will trigger drafts; older mail will only be visible in the activity log.

Projected cost at Draft authority: ~$0.14/mo.

Continue?

Click through or cancel.

Same for Auto-send — you’ll see a breakdown of the unlock criteria and how close you are to each.

Per-account exceptions

If you’ve connected multiple mailboxes, you may want a category on for one and off for another (e.g., auto-draft Inbound Leads on your work mailbox but not on your personal one). The backend supports this as a per-account exception layer.

Common patterns

”I just want drafts, nothing sent automatically”

Master switch: On
Every category: Active · Draft + review queue · All mail

Everything lands in Approvals. Nothing leaves your outbox without your click.

”Fully hands-off for calendar only”

Calendar: Active · Auto-respond · Contacts only (once the unlock criteria pass)
Everything else: Observe or Draft + review queue

Low-risk because calendar RSVPs are structurally simple.

”I’m on vacation”

Master switch: Paused

Preserves everything. Flip back on when you return.

Newsletters: Active · Observe · All mail

The agent classifies them into the Newsletter bucket so your weekly digest includes them, but never drafts a reply.

Troubleshooting

“I activated a category but no drafts are appearing.”

Check, in order:

Master switch is On (not paused)
Category state is Active (not Observe)
Category authority is Draft or higher (not Observe)
Scope matches the sender (if set to Contacts only, sender must be in your Contacts list)
Message arrived after the activation timestamp minus 48 hours
Your daily cap hasn’t been hit (/dashboard/usage shows daily spend)
The email has been classified into that category (check the activity log; classification can take a minute on a fresh inbox)

“My draft looks generic / off-tone.”

The agent learns from your approvals. Edit the draft before you approve — the edit delta feeds back into the style model for that category, and the next draft will land closer. Consistent approvals with no edits also contribute to the edit-rate metric for Auto-send.

“The daily cap keeps getting hit.”

Either raise the cap (up to your tier max) or switch some categories from Draft authority back to Observe until your mix settles. High-volume categories like Newsletters and Automated rarely need drafting — set them to Observe to save budget.

“I want to turn off automation for one mailbox but keep it on for another.”

Right now that requires the API workaround noted above. The per-account UI is on the roadmap.

Privacy

Nothing about the automation pipeline changes the privacy posture:

Credentials (app passwords) are encrypted per-agent
Classification and drafting happen in your agent’s container
Email bodies are not logged to our central servers
No AI model is trained on your email

See Privacy & security for the full statement.

Email triage & drafts — the capability from the user’s perspective
Managing approvals — what happens when a draft lands in your queue
Trust levels & approvals — the underlying trust system that authority levels build on
Connecting your email — the prerequisite