There is a genre of enterprise software problem that goes like this: a company generates a huge amount of data about itself, pays a vendor to store that data somewhere else, and then pays the same vendor again, at a rate that scales with success, for the privilege of asking questions about it. This is roughly the business model of observability - the category of tools that tell you whether your software is on fire - and it has produced some very large public companies. It has also produced a lot of engineers staring at an invoice and wondering how the bill to watch the app got bigger than the app.
Kloudfuse, a company in Cupertino with about twenty-two people, is a bet that those engineers are annoyed enough to do something about it. The pitch, which the company renders as “Your Cloud. Your Control.”, is that observability data should not leave your cloud at all. Instead of shipping your metrics and logs off to a vendor's SaaS, you deploy Kloudfuse's platform inside your own virtual private cloud, and Kloudfuse operates it for you. They call this “Self-SaaS,” which is a slightly awkward name for an idea that is not awkward at all: you host, they manage.
“The second time you build hard infrastructure, you build it for control - over where your data lives and what it costs.”
The reason this is interesting is that the people behind it have done the hard-infrastructure thing before, and it worked out. Kloudfuse was founded in 2022 by Krishna Yadappanavar, Pankaj Thakkar, and Ashish Hanwadikar. Yadappanavar and Hanwadikar previously built Springpath, a hyperconverged storage startup that Cisco bought in 2017 for $320 million. Thakkar was an early engineer at Nicira, the software-defined networking company that essentially invented a category and then sold to VMware for $1.3 billion. This is a founding team whose resume is, more or less, “we have turned unglamorous infrastructure into large acquisitions twice.” Investors noticed.
The unification argument
The other half of the Kloudfuse pitch is about sprawl. Modern observability tends to arrive as a pile of separate tools: one for metrics, one for logs, one for distributed traces, one for real user monitoring, one for profiling. Each has its own storage, its own query language, its own bill, and its own dashboard, and the on-call engineer at 3 a.m. is the human glue expected to correlate across all of them. Kloudfuse's answer is a single observability data lake that ingests metrics, logs, traces, events, real user monitoring, continuous profiling, and - as of recently - LLM monitoring, all in one place, correlated by default.
Technically, the interesting move here is that Kloudfuse is OpenTelemetry-native. OpenTelemetry, or OTel, is the open-source standard for how software emits telemetry, and its whole point is that you should be able to instrument once and send the data anywhere. Kloudfuse leans on this hard: keep your existing OTel pipelines, they say, and just point them at a backend that costs a lot less and that you own. The cost claim - 60 to 80 percent savings versus incumbents - is the kind of number you should treat as a vendor's number, but the mechanism is real enough. Log fingerprinting clusters repetitive log lines so you store patterns instead of duplicates; the data lake architecture avoids paying premium SaaS rates on high-cardinality data.
“Unified Observability. Your Cloud. Your Control.”
Then the models showed up
Observability had a fairly stable definition for a decade - you watch CPU, memory, latency, error rates - and then generative AI arrived and quietly added a new thing to watch. When your application calls a large language model, you now care about token usage, prompt failures, model-level errors, and cost per call, none of which fit neatly into the old dashboards. In December 2025, Kloudfuse shipped version 3.5, which folds this in: LLM monitoring lives directly inside the APM traces, capturing prompt and output tracing as events attached to distributed traces, and tracking token usage across OpenAI, Anthropic, Google, AWS, and Azure.
The same release includes a genuinely of-the-moment feature: a Model Context Protocol server. MCP is the emerging standard for letting AI agents talk to tools, and Kloudfuse used it to let systems like Claude or ChatGPT query your observability data in plain English. You can ask, roughly, “why is checkout slow,” and the platform translates that into its own query language, FuseQL, and returns correlated insights across metrics, logs, traces, and dependencies. Whether the future of on-call is an engineer asking an AI to read the dashboards is a real question; Kloudfuse has decided the answer is yes and shipped accordingly.
Who actually buys this
The in-your-cloud pitch lands hardest with companies for whom data leaving the building is a compliance problem, not a preference. That is why Kloudfuse's named customers skew toward regulated and large-scale: GE HealthCare, Zscaler, Tata 1mg, Automation Anywhere, Innovaccer, Eltropy. The 3.5 release also added FIPS 140-2/3 validated cryptography and expanded data scrubbing across every telemetry stream, plus audit trails aimed at GDPR's right-to-deletion and HIPAA's data-minimization rules - the unsexy checklist items that decide enterprise deals.
The company launched out of stealth in November 2023 with $23 million, including a $17 million Series A led by Newlands with Blumberg Capital, Aspenwood Ventures, HighSage Ventures, and Exponent participating; total reported funding sits around $29 million. In 2025 it earned an Honorable Mention in Gartner's Magic Quadrant for Observability Platforms, which for a company this small is less a coronation than a foot in the door of a market dominated by names with far bigger sales teams.
None of this guarantees anything. Kloudfuse is competing against Datadog, New Relic, Grafana, Splunk, and Dynatrace - well-funded incumbents with entrenched contracts and their own AI roadmaps. The self-hosted model that is a selling point to a bank is an operational chore to someone who just wants a URL and a login. But the underlying observation is hard to argue with: a lot of companies are uncomfortable with how much they pay to watch themselves, and how little of that data they truly control. Kloudfuse built a company around taking both of those complaints literally. The values it lists for itself, fittingly, all start with the same letter - Humble, Hungry, Honest - which is either a coincidence or exactly the kind of thing a team of repeat infrastructure builders would find satisfying.