Talavine
Sweep removes the backlog
Mail you trash before enabling the pipeline is never enriched. Sweeping a sender with 3,000 messages of history is 3,000 model calls that never happen.
Tutorial
Make room for the messages that matter.
Your inbox is one part of your data hub. Teach Talavine how to handle routine mail so newsletters and alerts are easier to manage and useful conversations are easier to find.
Why clean first
Every message the pipeline enriches is a model call. Junk you remove first is money you never spend.
The cleanup moment
After an account finishes downloading, Talavine pauses before processing so you can triage.
Sweep bulk junk
The Top senders helper finds safe candidates and clears years of noise in a few clicks.
Teach the keepers
One taught sender becomes a permanent, deterministic rule. No model call needed ever again.
Why clean first
The cheapest model call is the one that never happens.
Talavine's message pipeline makes one calibrated AI call per message to categorize, summarize, and route it. That's efficient, but in a typical mailbox most of the volume is newsletters, promotions, and notifications. Cleanup attacks the cost from three directions.
Teach removes the future
A taught sender is classified by your rule, not by a model. Newsletters, promotions, alerts, low-priority mail, and junk default to skipping the AI pipeline entirely: routed, filed, and labeled deterministically at zero cost.
Spend Guard catches the rest
Whatever still reaches the pipeline runs under a spend ceiling, with a full per-call cost history. Cleanup shrinks the bill; Spend Guard makes sure it can never run away.
Step 1 · Clean up
Talavine pauses before spending. That pause is your moment.
When you connect a mailbox, Talavine downloads the full history first and then waits. AI processing doesn't start until you say so, which means you get one perfect window to clear junk before a single model call runs.
The cleanup banner
When an account's initial download completes, a banner appears at the top of the inbox: “your account is ready for cleanup.” It offers “Review cleanup helper”, which opens that account's dashboard, and “Start processing now” if you'd rather skip straight to AI processing.
With several accounts ready at once the banner offers “Start all now”, but running the helper first is where the savings are.
The cleanup gate
Open any account from the sidebar and the dashboard shows the same choice: “Cleanup done — start processing” once you've swept, or “Skip cleanup, start now” if the mailbox is already tidy. Until you pick one, the AI pipeline waits.
Nothing is lost by taking your time here. The mailbox is fully downloaded and searchable; only AI enrichment is on hold.
Step 2 · Sweep
The Top senders helper: years of noise, gone in a few clicks.
The account dashboard's Top senders card ranks the mailbox's highest-volume senders and shows only safe cleanup candidates. VIPs, address-book contacts, and anyone you've written to are filtered out. Check the noisy ones, pick an action, and let it run in the background.
Sweep checked
Moves every message from the selected senders in this account to Trash. It's a one-time action: the sender isn't blocked, so future mail still arrives.
Block + sweep checked
Sweeps the history and registers a permanent block, so future mail from those senders is stopped deterministically, with no model call needed to reject it.
Archive checked
Keeps the mail but moves it out of the inbox. Good for senders you might search later but never need to read.
Mark checked read
Clears the unread noise without moving anything, for senders that are harmless but pile up.
A progress dialog tracks the sweep: messages affected so far, then a short “Syncing to mailboxes” phase while the moves flow back to your mail server over IMAP. Before sweeping a sender you're unsure about, expand their row to preview sample messages; double-click any sample to read it in full.
Step 3 · Teach
Teach a sender once; never pay to classify them again.
Teach writes a deterministic routing rule for a sender: which category their mail belongs to, whether the AI pipeline should run for them, and how long their messages should live. It's available everywhere you meet a message.
The inbox toolbar
Select any message and click “Teach”. A “Should be” menu lists the categories, so you can pick where mail like this belongs.
Right-click any message
The context menu has the same Teach submenu, plus one-click “Teach: Junk / Block Sender” and “Teach: Phishing” for the obvious cases.
New Mail Review
On the Unread view, “Teach senders” opens a triage panel of unclassified senders. “Preview AI suggestions” proposes categories with confidence scores; “Apply safe suggestions” accepts the non-destructive ones in bulk.
Top senders on the dashboard
Each sender row has a Teach toggle that expands sample messages and category chips, so you can classify a sender while you're deciding whether to sweep them.
Tell it what the mail is
Inbox / important, Financial, Order tracking, Appointments / travel, Security, Health / legal / government, Work / client, Operational alert, Newsletter, Promotion, Low priority, plus one-click Junk and Phishing. Categories like Newsletter and Promotion route to their own AI-sorted stream instead of your inbox.
Streams are smart views, not new folders. Your underlying mailbox structure is untouched and stays compatible with every other mail client.
Control the cost per sender
Each taught sender has an AI mode: Auto (the category decides), Full (always run the pipeline), or Skip (never spend a model call on them). There's also an Expire setting that auto-deletes their mail after 3, 7, or 30 days.
Newsletters, promotions, alerts, low-priority mail, junk, and phishing default to Skip. Pick Junk or Phishing and safe defaults (Skip + 3-day expire) apply in one click.
Putting it together
The 15-minute cleanup routine for a new mailbox.
Run this once when an account finishes downloading, and your AI budget goes to the mail that deserves it from day one.
01
step
Open the cleanup helper before starting the pipeline
When the “ready for cleanup” banner appears, click “Review cleanup helper” instead of “Start processing now”. The Top senders table loads the account's highest-volume senders, already filtered to safe cleanup candidates.
Why it matters: The list deliberately excludes VIPs, address-book contacts, anyone you've emailed or replied to, and your own agents. People you actually correspond with never show up as sweep candidates.
02
step
Sweep the obvious junk in bulk
Check the senders you never want to hear from again and click “Block + sweep checked”. For borderline senders, plain “Sweep checked” trashes the history without blocking. A progress dialog counts affected messages, then syncs the moves back to your mail server.
Why it matters: Sweeps go to Trash through standard IMAP, so your other mail clients stay in step and your provider's normal trash-recovery window applies.
03
step
Teach the senders you're keeping
For the high-volume senders that survive, expand the Teach row and click a category: Newsletter, Promotion, Order tracking, Financial, Alert, and more. Newsletters, promotions, alerts, and low-priority mail default to skipping the AI pipeline entirely.
Why it matters: Teach is a routing rule for similar future mail, not a one-off folder move. Classify a sender once and every future message from them is handled deterministically, at zero model cost for skip-mode categories.
04
step
Set Expire on mail with a shelf life
In the same Teach control, the Expire chips (3, 7, or 30 days) auto-delete a sender's messages after they've gone stale. Delivery notifications, one-time codes, and daily deal mail clean themselves up from now on.
Why it matters: Choosing Junk or Phishing applies safe defaults automatically (skip the pipeline, expire in 3 days), so the worst mail needs exactly one click.
05
step
Start the pipeline on what's left
Click “Cleanup done — start processing”. AI enrichment now runs only over mail that earned a place in your inbox: the junk is in Trash, the blocked senders are gone, and the taught senders route themselves without model calls.
Why it matters: You can pause and resume processing from the dashboard at any time, and re-run a sweep whenever a new noisy sender shows up.
06
step
Keep teaching as you go, and watch the savings
From now on, when a message lands in the wrong place, select it and hit “Teach”. Each correction becomes a permanent rule. Spend Guard caps total AI spend, and the LLM call history shows exactly what each model call cost.
Why it matters: A taught inbox converges fast: after the first week or two, most of your routine mail is classified by your own rules instead of a paid model.
Cleanup FAQ
Common questions about cleanup, Sweep, and Teach.
How does cleaning up email in Talavine save on LLM costs?
Talavine enriches each message with a calibrated AI model call. Cleaning up before enabling the pipeline means swept and blocked mail is never enriched, and teaching a sender writes a deterministic routing rule. Categories like newsletters, promotions, alerts, and junk skip the AI pipeline entirely, so that sender's mail is classified and filed at zero model cost forever.
What is the difference between Sweep and Teach?
Sweep is a one-time bulk action over existing mail: it moves everything from the selected senders to Trash (or archives it, or marks it read) in the current account. Teach is a rule for the future: it records how mail from a sender should be categorized, whether the AI pipeline should run for them, and whether their messages should auto-expire. Block + sweep combines both: it clears the history and blocks the sender permanently.
Can Sweep delete mail from someone important by mistake?
The cleanup helper only lists safe candidates. Senders marked VIP, senders in your address book, anyone you have emailed or replied to, and your own agents are excluded from the Top senders cleanup table. Sweeps also go to Trash via standard IMAP rather than deleting permanently, so your provider's normal recovery window applies.
Do I have to clean up before Talavine starts processing?
No. After an account downloads, the pipeline waits so you have the chance to clean up first, but “Skip cleanup, start now” begins processing immediately. You can run sweeps and teach senders at any time later. Cleaning first just means the AI never spends money on mail you were going to delete anyway.
What does teaching a sender as Junk or Phishing do?
Junk and Phishing are one-click: they route the sender's mail to Junk, skip the AI pipeline for them, and set a short auto-expire (3 days) so the mail cleans itself up. Phishing additionally feeds Talavine's phishing detection. Both are per-sender rules, so future mail is handled instantly and deterministically.
Does Talavine have other cost controls besides cleanup?
Yes. Spend Guard sets a hard ceiling on AI spend, the LLM call history shows every model call with tokens, cost, and latency, and the model router prefers free local models where they're proven good enough. Cleanup multiplies all of it by shrinking the number of messages that need a model call at all.
Your inbox, your rules, your budget
Connect a mailbox, sweep the junk, teach the keepers.
The first cleanup pays for itself immediately, and every sender you teach afterward makes the AI cheaper and the inbox calmer. Spend Guard and the per-call cost history keep the whole thing honest.