About
We give on-call engineers their night back
Every incident starts the same way: an alert fires at 3am and an engineer opens six tabs to find out why. That hour of blind digging, not the fix, is the problem Ketl exists to solve.
Built by engineers who have been paged
The founders ran on-call rotations at a payments platform. The alerts were real and the tools were fine. But finding the cause still took most of an hour of reading logs in four places, cross-referencing a deploy, and pulling in a colleague who knew the service.
On-call incidents are not usually hard. They are messy. The signal is scattered across logs, traces, recent deploys and past incidents. Reading all of it together and naming a cause is reading and correlation work, not deep expertise work.
Modern language models are good at exactly that work. They can read raw, unstructured telemetry, hold the context of a deploy and a log window and a trace at once, and name a cause in plain language. That is why Ketl is AI-native: the reasoning is the engine, not a feature bolted onto a dashboard.
Ketl is not autopilot. It reads and correlates so the on-call engineer starts at the hypothesis, not the raw logs. They still decide. They still act. They get the hour back.
The loop before Ketl
- 01Alert fires at 3:14am
- 02Open logs, traces, dashboard, deploy feed, past incidents
- 03Spend 47 minutes correlating across six tabs
- 04Name a hypothesis, propose a fix
- 05Act, verify, close
Ketl reads steps 2 and 3. The engineer starts at step 4.
How we build
Four principles that guide every decision, from the prompt design to the product.
Minutes, not hours
The point of Ketl is the time it gives back during an incident. Everything we build is measured against that.
Show your work
Every cause cites the lines behind it. A copilot you cannot check is one you cannot trust at 3am.
Propose, do not seize
Ketl advises; a human acts. We earn the right to act automatically one careful, opt-in step at a time.
Quiet with your data
Telemetry answers your incidents only. It never trains a shared model. Read-only by default.
11 people building the copilot for on-call
Marisol and Aaron started Ketl and build it full time. The rest of the team brings infrastructure, product and reliability expertise from environments where incidents are expensive and sleep is scarce.
Marisol Reyes
Co-founder & CEO
Marisol ran the on-call rotation for a payments platform and spent too many nights doing the blind digging Ketl now does in seconds. She started Ketl to give that hour back to every on-call engineer. She builds Ketl full time.
San Francisco, CA
Aaron Feld
Co-founder & CTO
Aaron built observability pipelines and model-routing infrastructure before Ketl. He owns the reasoning engine: how telemetry is read, how models are routed, and how every claim stays tied to evidence. He builds Ketl full time.
San Francisco, CA
Devin Cho
Founding Engineer, Inference & Routing
Devin makes the model-agnostic router fast and cheap, deciding which model reads which signal, and keeping a lab outage from ever reaching the pager.
Seattle, WA
Hannah Weiss
Head of Product
Hannah shapes how a diagnosis reads at 3am: ranked, evidence-first, and honest about confidence. Her rule is that a tired engineer should trust it or overrule it in seconds.
San Francisco, CA
Theo Marchetti
Founding Engineer, Integrations
Theo connects Ketl to the tools teams already run, from PagerDuty and Datadog to GitHub deploys, so the signal arrives the moment an alert fires.
Austin, TX
Priya Nair
Reliability Engineer, Applied ML
Priya turns real incidents into the evaluations Ketl is measured against, so the engine keeps clearing the bar as models and failure modes change.
Remote, US
Legal entity
Ketl, Inc.
Incorporated
Delaware, United States
Founded
March 2023
Stage
Pre-Seed
Headquarters
535 Mission Street, Suite 1400San Francisco, CA 94105United StatesPhone
+1 (415) 555-0134
Team size
11 people
Come build the copilot for on-call
We are a small team shipping fast on a problem that keeps engineers up at night. If that sounds like a good use of your time, we want to hear from you.