Oximy Wants to Be the System of Record for How Companies Use AI
Enterprises are running dozens of AI tools with no idea what they cost or where the data goes. Oximy sits on the network and turns that fog into a live map you can actually govern.
Every company now runs on AI it cannot fully see. A designer expenses a subscription to one image tool. An analyst pastes a spreadsheet into a chatbot. An engineering team wires an API key into a script and forgets about it. Multiply that by a few thousand employees and you get the quiet crisis Oximy was built for: no single person can name every AI tool their own company uses, what it costs, or what data walked out the door with it.
Oximy, a San Francisco startup in Y Combinator's Winter 2026 batch, calls itself the system of record for how a company uses AI. In plainer terms, it is the layer that watches. It sits on the network - on firewalls and proxies, wired into sign-in and security tooling - and produces one live map of every AI tool, account, model, and dollar the organization touches. No SDK to install. No vendor-by-vendor integration project. According to the company, it can be running in under a day.
01Shadow IT took a decade. Shadow AI took a year.
Security teams spent the 2010s learning to inventory the unsanctioned software employees dragged into work. AI compressed that timeline. Where a new SaaS app might take budget approval and an IT ticket, a new AI tool takes a browser tab and a corporate email. The result is what Oximy calls AI sprawl - dozens of tools in daily use, most of them invisible to the people accountable for cost and compliance.
The stakes are highest in regulated industries, which is where Oximy has started. When an employee at a bank or a hospital pastes a record into a chatbot, that is not a productivity story; it is a data-governance event. Oximy's pitch to those buyers is blunt: you cannot govern what you cannot see, and right now you cannot see it.
02See it, run it, route it
Oximy is three products that feed one another. Visibility is the map: a live inventory of AI tools, spend, and risk that discovers and categorizes usage across the org, sanctioned or not, with real-time spend analytics broken out by team, vendor, and workflow. Sidekick is a company-controlled workspace - multiplayer threads where people and agents collaborate across any model, using the company's own keys and connectors. Relay is the traffic cop: it routes each request to the best model by cost, quality, and latency.
The design choice underneath all three is that Oximy is model-agnostic on purpose. It does not want to sell you the model. It wants to be the map of every model you already use - and it insists your data, artifacts, workflows, and configuration stay inspectable and portable, even if you decide to leave.
Where it fits03Betting on the map, not the model
Plenty of startups are racing to build a better chatbot. Oximy is making a different wager: in a world of a thousand models, the durable enterprise business is the control plane that governs all of them. That puts it in conversation with AI-security and data-loss-prevention tools on one side, SaaS-discovery and CASB vendors on another, and LLM gateways like the open-source routing projects on a third. Oximy's argument is that discovery, spend, data protection, a workspace, and routing belong in one system of record rather than five dashboards.
04A third act at twenty-three
Oximy is the work of Naman Ambavi, who founded the company solo in 2025. He is 23, and this is not his first company. He co-founded Easocare, a pharma supply-chain business he says reached hundreds of thousands of lives across a dozen products, and he was on the core team at Induced, an AI browser-automation startup. The through-line is enterprise plumbing - the unglamorous systems that decide whether a big organization runs smoothly or not.
That track record is what pulled in his first check. The angel investor Sudarshan Sridharan, who backed Ambavi's previous startup, wrote the first check into Oximy - calling it the first governance and observability layer for AI usage in the enterprise, and citing three years of watching Ambavi's execution and product velocity.
05Quiet infrastructure, live customers
Oximy sells to enterprises as B2B software, and it is already past the demo stage - the company says it processes millions of AI requests a day, with early customers in financial services and healthcare. The team is small, four to eight people depending on the day, and hiring founding engineers and a founder's-office go-to-market role. The category is young enough that the biggest competitor is still inertia: companies that have not yet realized they need a map.
The tagline the company keeps returning to captures the posture better than any feature list. It does not promise to invent your AI strategy. It promises to hand you the controls to the AI you are already running - on your keys, on your network, on your terms.