How it works
Chat with Kelp to tell it what’s not working in your inbox. Kelp finds real examples, creates a plan, and takes action when you’re ready. Here’s how it works.
Works with Gmail and anything on top of it, including Superhuman.
First, the rule that makes it safe
Your contacts, anyone you correspond with, and anyone you choose to trust. The people who matter are never filtered, never interrupted. Kelp reads nothing of theirs.
Kelp focuses on cold inbound from people you don’t know. It’s the only mail Kelp reads. AI judges each message against your instructions and in context: what it says, not which keywords it contains.
Senders and domains you’ve blocked are quietly removed before they reach you. Kelp automatically blocks your worst offenders. One click to block; one click to undo.
Start with a conversation
Describe the problem in plain language. Kelp searches your inbox for real examples, drafts the filter, and dry-runs it in front of you, so you see exactly what it would have done before anything is live.
I went through the last month and found nine like that. Here are two. Tell me if either should still get through:
I’d file these under Recruiting and archive them: in Gmail if you ever want them, out of your way otherwise.
Quietly agentic
After setup you mostly don’t open settings. You just use your email. Mark a message as spam, peel off a label Kelp applied: it notices. Overnight, it reviews what changed, forms its own ideas, and backtests them against your recent mail.
Suggested by Kelp · backtested on your own mail
You moved to Rockridge last year, but Nextdoor still sends daily digests for your old Santa Monica neighborhood. I’d filter those out and keep anything tied to where you live now.
Tested against your last 30 days: it caught all 23 Santa Monica digests and left your direct messages, security alerts, and anything local to Rockridge alone.
Memory you can read
Like any good curator, Kelp keeps notes: what your work is, what matters to you, what you’ve already declined. Those notes shape its judgment about your mail.
Every note is visible in your settings, editable and deletable. Kelp learns quietly, but never secretly.
Message bodies are handled in memory just long enough to classify, and never written to our database.
For mail Kelp filters, we keep the sender, subject, and first few words for about 30 days so you can review its work. Turn it off in Settings and it’s purged immediately.
We store one-way hashes of your contacts: enough to recognize someone you know, never the addresses themselves.
Google’s consent screen mentions “send” because Gmail’s permission bundles it, but Kelp has no sending code at all. It only sorts what arrives.
Your mail never trains a model (ours or anyone’s) and is never sold or shared.
Kelp passed Google’s CASA security assessment, run by accredited assessors through the App Defense Alliance, before Google granted production Gmail access.
Learn more at Security & Compliance (written for security teams) or our Privacy Policy.
Sign in with Google, tell Kelp what’s bothering you, and let it take things from there.
By signing in, you agree to our Terms of Service and Privacy Policy.