Breaking
YC W24 — OneGrep launches out of Y Combinator's Winter 2024 batch DEVOPS — AI agent automates incident response across a team's existing tools COST — Observability can eat ~30% of cloud spend; OneGrep targets the bill MCP — Ships SDK + gateway so agents discover and run the right tool PEDIGREE — Founders from Disney, Amazon and Coinbase YC W24 — OneGrep launches out of Y Combinator's Winter 2024 batch DEVOPS — AI agent automates incident response across a team's existing tools COST — Observability can eat ~30% of cloud spend; OneGrep targets the bill MCP — Ships SDK + gateway so agents discover and run the right tool PEDIGREE — Founders from Disney, Amazon and Coinbase
Company Profile · Developer Tools

OneGrep wants your runbooks to run themselves

A two-person Y Combinator startup out of New York is teaching AI agents to do the parts of DevOps nobody enjoys: chasing incidents at 2 a.m. and paying the observability bill.

Every engineer who has carried a pager knows the ritual. An alert fires. You wake up, squint at a dashboard, and start hunting through logs, dashboards, past tickets, and a wiki page someone wrote two reorgs ago. Most of the time, the outage is not new. The fix lives in a senior engineer's head or in a runbook that has quietly rotted. OneGrep, a startup that came out of Y Combinator's Winter 2024 batch, is built on a plain observation about that ritual: if the problem repeats, the response should not be manual.

OneGrep is a DevOps agent. In its own framing, it works across the tools an engineering team already runs - observability, CI/CD, infrastructure - and helps them search, correlate, and automate the operational work that usually requires a human who knows where everything is buried. It can investigate an incident on its own, including during off-hours, and turn a static runbook into something interactive that an agent can actually follow.

~25%
Engineer time spent resolving incidents
~30%
Share of cloud spend going to observability
2
People on the founding team

The ProblemTwo expensive problems, one root cause

OneGrep goes after two pains that most engineering leaders will recognize on sight. The first is time. Engineers spend a large slice of their week - roughly a quarter, by the company's estimate - fighting incidents instead of building. The second is money. Observability, the tooling meant to make production legible, has a habit of quietly consuming close to a third of a company's cloud bill through logs, metrics, and traces that nobody reads.

The root cause underneath both is the same: knowledge and signal are scattered. The context needed to fix a problem is spread across a dozen SaaS tools and a handful of people. OneGrep's answer is to filter the telemetry before it becomes an expensive haystack, and to put an agent on top of what remains.

Where the DevOps dollar and hour go
Observability
~30%
Incident time
~25%
Building product
rest
The pitch in one chart: observability cost and incident time are the two lines OneGrep is trying to bend. Figures are the company's own approximations.

How It WorksFilter the noise, then let the agent operate

OneGrep fine-tunes language models on two things at once: real-time telemetry, and the knowledge bases engineers actually use to run production systems. On the cost side, it filters logs, metrics, and traces with AI before they hit the observability platform, aiming to keep the signal while dropping the noise that inflates the bill. On the incident side, it correlates data and surfaces the essential facts fast, so an on-call engineer - or the agent itself - is not starting from a blank dashboard.

Telemetry AI filter Correlate Interactive runbook Resolve

The best DevOps engineer on your team is also a bottleneck. All the knowledge lives in one head. OneGrep's whole bet is that this should be a system, not a person.

— The thesis, in plain terms

The PivotFrom copilot to the tool layer agents need

OneGrep started life described as a "DevOps copilot." Over time its center of gravity has shifted toward a less glamorous but arguably larger problem: how AI agents find and safely use tools at all. The company now publishes a TypeScript SDK and an MCP gateway - "mcp-express" - that let agents author what OneGrep calls Toolprints and operate services that follow the Model Context Protocol, the emerging standard for connecting AI to external tools. There is a CLI explorer that lets a developer find tools by describing a goal in natural language, and an integration with Smithery, a hub for MCP-compatible services.

It is a logical move. Everyone is building agents; far fewer have solved the boring middle - how an agent picks the right tool and runs it without breaking production. That middle is exactly where a team of DevOps veterans has an edge.

Two products, one thesis
DevOps agent — automate incidents & runbooks
Agent tooling — SDK & MCP gateway for tool discovery
Both point at the same idea: agents that operate real tools, safely.
Not two companies: the agent and the tooling layer are the same bet seen from two ends of the stack.

The FoundersPeople who have carried the pager at scale

The credibility here is in the resumes. Achintya Ashok, co-founder and CEO, led the search team at Disney Streaming and has said he helped Disney+ absorb a two-week surge to 70 million users; he has also worked at Coinbase and Goldman Sachs. Brian Cripe, co-founder and CTO, was a founding engineer on the incident management platform at Amazon that tens of thousands of engineers reach for every day. Both have spent their careers on the operations side of very large systems, which is to say they have been the person paged.

Scaling Disney+ to 70 million users in two weeks teaches you which parts of operations are worth automating - and which parts still need a human.

— On the founders' background

Business & MarketSold on outcomes, run on your own cloud

OneGrep sells to engineering, DevOps, and SRE teams as B2B software, and it leans on outcomes rather than features: resolve incidents faster, spend less on observability. Notably, teams can deploy it on their own infrastructure or cloud - a detail that matters when the product needs access to production telemetry and internal tooling, and when data security is a gating concern for buyers.

The market it sits in is crowded at the edges and thin in the middle. Observability incumbents like Datadog and Grafana, and on-call platforms like PagerDuty, are all bolting AI onto existing products. A newer wave of AI-incident startups is chasing the same on-call pain. OneGrep's distinction is that it is not trying to be the dashboard or the pager - it sits across the tools a team already has and puts an agent to work on top of them. On telemetry cost specifically, it lines up against observability-pipeline tools that filter data before it is billed.

Where OneGrep sits
Datadog / Grafana
see
PagerDuty
alert
OneGrep
act
The wedge: incumbents help you see and get alerted; OneGrep is aimed at the step after - acting across the tools you already run.

What's NextA small team, an open question

OneGrep is still early - a seed-stage, two-person company with a modest disclosed raise and a product that is finding its shape. The open question is whether the DevOps agent or the MCP tooling layer becomes the main event. Both rest on the same wager: that operating real production tools is the job AI agents are best positioned to take on first, and that the team who built incident platforms at Amazon and search at Disney knows where the sharp edges are.

For now, the name says the plan out loud. "Grep" is the Unix command engineers use to search through everything. OneGrep is the pitch for one search - and, increasingly, one agent - across all of it.

#onegrep#devops#ai-agents#mcp #observability#incident-management#yc-w24 #developer-tools#runbook-automation#new-york