Built by the people who carry the pager.
oneinfra exists because the team building it spent years switching between five tools to understand a single alert — and decided it was easier to write the AI agent that should have lived in the middle than to keep tabbing between dashboards.
Why this exists
If you've worked on a platform team in the last five years, you've probably built some version of this on a Sunday afternoon: a script that takes the firing Prometheus alert, pulls the recent commits, greps the logs, and posts a Slack message that might point at the cause. It works for a week, breaks, and never gets touched again.
oneinfra is what happens when you decide the script should be a product, the script should call an LLM, the script should have access to your knowledge graph, your cost data, your SLOs and your code repo — and the whole thing should still run inside your VPC because nobody wants to send their production telemetry to a vendor's cloud.
We're building the platform we wished we'd had at every previous company. The AI-SRE category is suddenly crowded with SaaS startups; we think the right answer is open source, self-hosted, and bring-your-own model. None of them are.
Who's behind it
A small, deliberately senior team — SREs, platform engineers, and AI/ML practitioners who have shipped infrastructure software at scale. We are intentionally heads-down on the product right now rather than fundraising; we will introduce ourselves properly when the GA repo opens. If you want to talk in the meantime, hello@oneinfra.tech reaches a human within 4 business hours.
How we work
Public changelog, public comparison matrix, public pricing. No sales gate to see the live product — it's open at oneinfra.tech/oneinfra right now. Design-partner conversations happen by email, not via marketing automation. We will publish named case studies as design partners go live, not before.
What we'll defend in a design review.
Your data stays where you put it
We will never build a SaaS data plane. The day we ask you to forward your telemetry to our cloud is the day we have lost the plot. Self-hosted is not a tier — it is the only tier.
Specialized agents, not "an AI"
"AI" is not a product. A team of agents — each with explicit scope, tools and handoffs — is. We will keep naming them, scoping them, and measuring them in public.
Shipping > positioning
We keep the changelog honest, the comparison matrix honest, and pricing on a public page. If we can't justify a feature in plain English, we shouldn't have built it.
On-call should be quiet, not loud
Every alert that wakes someone up without a root cause attached is a bug in our product. We treat alert fatigue as a P0, not a "soft metric".
Become a design partner.
We're working with the first 10 teams who want this built right. There's a real ask in both directions — see the program page for what you get and what we ask in return.