Deterministic where it can be.
Intelligent where it must be.
Most AI email tools run a model over every message and hope. Aldyn runs a layered engine where the AI is the last resort, not the first. That's why it's consistent, explainable, and a fraction of the cost.
The engine, in order
Sender, domain, subject, regex, list: your rules always win, before anything else runs. Import your existing Gmail filters in one click.
Every sender is judged once, then filed identically forever. Your corrections repin instantly and permanently.
Newsletters, notifications, receipts, calendar invites, and cold-outreach platforms identified from mail headers: deterministic, instant, free.
Only genuinely ambiguous mail (usually first contact from a real human) reaches the model. Its verdict is pinned so it never re-decides the same sender.
Below the confidence bar, Aldyn leaves the email in your inbox and says so. A wrong guess costs your trust; we don't gamble with it.
A decision log no one else will show you
Every filed email carries its reason: the rule that matched, the pin you created, the header that gave it away, or the AI's one-line rationale with a confidence score. When something looks wrong, you don't wonder. You look, then fix it with one drag, permanently.
Competitors call their sorting a smart black box. When it misfiles something, you get no explanation and no guarantee it won't happen again tomorrow. Aldyn's whole architecture is the opposite bet: trust comes from showing your work.
Categories that are actually yours
Create categories in plain language ("stuff from my property managers about the rentals") and Aldyn sharpens the description for its classifier, suggests example emails from your history for you to confirm, and proposes deterministic rules from the patterns in what you approved.
New categories start deterministic-only: the AI earns the right to file into them only after you've confirmed real examples. And if you make a label directly in Gmail, Aldyn notices and asks whether to manage it; it never bulldozes a filing system you already have.
Already have years of labeled email? Adopt a label and Aldyn learns from its entire history in about a minute: every sender pinned, every message a training example. Your past effort becomes your assistant's expertise.
Drafts in your voice, only when wanted
When an email genuinely needs a reply, a draft appears in your real Drafts folder, threaded into the conversation. You edit, you send; Aldyn never sends anything. And when a reply isn't clearly needed, there's no draft at all. A missing draft costs you two seconds; a wrong-sounding one costs you a relationship.
The voice behind those drafts is learned, documented, and measured. Aldyn distills your sent mail into a readable profile (greetings, sign-offs, habits, typical length) plus a per-recipient register: how you write to your cofounder is not how you write to a vendor, and Aldyn knows the difference. The profile retrains automatically as you send.
Then it grades itself: every time you send one of its drafts, Aldyn measures how much survived your edits. That voice-accuracy score is the number that should climb week over week, and you can watch it. Your standing instructions (banned phrases, sign-off rules, bluntness settings) override everything the profile has learned; the learned document itself is read-only, because a profile you can hand-edit is fiction. It remains yours to clear at any time.
Cold outreach quarantined. Newsletters respected.
Sales-sequence platforms disguise their mail with the same compliance headers real newsletters use. Aldyn identifies them deterministically and quarantines the pitch away from the newsletters you actually chose. Unsubscribe from anything in one click.
Privacy as architecture, not a promise
Aldyn processes your email in memory and stores decisions, rules, and sender pins. Your email bodies are never stored. What it learns about your preferences is stored as readable rows you can inspect, edit, or delete, never as invisible model weights.
Your data trains your assistant only. No shared models, no 'improving our AI' clause, no training on customer mail. That is structural, not just contractual. See the privacy policy for the full architecture.
the data pledge
Your email is yours. Full stop.
Never sold. Your data is not for sale, rent, or 'sharing with partners.' There is no version of Aldyn where your inbox is the product.
Never used for ads. No ad targeting, no profiling, no analytics resale. Ever.
Never trains shared AI. Your email and voice profile train your assistant only. Our model providers process your data under paid API terms that prohibit training on it.
Never handed to anyone else. No third party touches your email or your voice training except the infrastructure required to run Aldyn: hosting and model APIs under strict data-processing agreements. The only other exception is a valid legal order we are compelled to comply with.
Never keeps what you delete. Aldyn stores organization records, not copies of your email. Permanently delete a message in your inbox and every trace of it on our servers (its decision record, drafts, example sketches) is purged automatically. Forever.
Not marketing copy: architecture. Aldyn doesn't store your email bodies at all, and everything it learns about you is readable and deletable in the app. Read the full privacy policy.
What a voice profile looks like
This is the actual screen in Aldyn (with example data). No other email assistant will show you what it thinks your voice is, let alone score itself against your real sends.
Writes short, warm, direct emails: straight to the ask in the first line, no formal openers entirely. One idea per paragraph, one question per email. Friendly but never fluffy; uses a dash where most people would start a new sentence.
The full comparison
| Compare | Aldyn | Fyxer AI | SaneBox | Superhuman | Gemini / Copilot |
|---|---|---|---|---|---|
| Starting price / user / mo SaneBox's $7 tier is rules-based filing only, no AI understanding, no drafts. Native = Gemini in Workspace / Copilot in M365 plans. | $14 · $10 billed yearly | $30 | $7 | $33+ | bundled |
| AI-grade categorization (understands content, not just rules) | ✓ | ✓ | ✗ | ◐ | ◐ |
| Custom categories you define, trained by your own examples Fyxer's five categories are fixed: no renaming, creating, or deleting. SaneBox allows custom folders but files by sender behavior, not meaning. | ✓ | ✗ | ◐ | ✗ | ✗ |
| Shows why every email was filed (decision log) | ✓ | ✗ | ✗ | ✗ | ✗ |
| Consistent by design, same email, same result “Mislabeled emails remain the most consistent complaint” across Fyxer reviews. Aldyn decides each sender once, then files deterministically. | ✓ | ✗ | ✓ | ✗ | ✗ |
| One drag = permanent correction (visible, reversible) | ✓ | ◐ | ✓ | ✗ | ✗ |
| Learns your existing labels & filing history at onboarding | ✓ | ✗ | ✗ | ✗ | ✗ |
| AI reply drafts in your voice | ✓ | ✓ | ✗ | ✓ | ◐ |
| A voice profile you can read, with steering you control Fyxer accepts a custom tone prompt, but what it has learned about your voice is invisible and unauditable. Aldyn shows you the profile, measures drafts against what you actually send, and lets your instructions override it. | ✓ | ◐ | ✗ | ✗ | ✗ |
| Cold-outreach quarantine (separate from real newsletters) Fyxer routes cold pitches into the same bucket as marketing you subscribed to. | ✓ | ✗ | ◐ | ✗ | ✗ |
| Meeting notetaker | Pro tier | ✓ | ✗ | ✗ | ◐ |
| Keep your existing inbox app (no switching) | ✓ | ✓ | ✓ | ✗ | ✓ |
| Multiple inboxes on the entry tier Fyxer's $30 tier covers one inbox; a second forces the $50 tier. | 2 included | ✗ | ✗ | ✗ | ✗ |
| Trial that never auto-converts to paid Fyxer's trial auto-converts to paid; surprise post-trial charges are its most common 1-star review theme. Aldyn never charges a card without explicit confirmation, and Organize trials collect no card at all. | ✓ | ✗ | ✓ | ◐ | ✓ |
| No hidden overage fees Multiple independent reviews report undisclosed Fyxer volume overage fees that appear on no public pricing page. | ✓ | ✗ | ✓ | ✓ | ✓ |
| Your data trains only your assistant, never a shared model | ✓ | ◐ | ✓ | ◐ | ◐ |
Based on vendors' public pricing pages, documentation, and independent reviews as of July 2026. ◐ = partial or limited.