Tetrate
routes the rest.

01Who They Are, Right Now

A Wednesday morning at a Fortune 100 bank. A developer pushes a small change - a new chatbot that summarizes loan documents. The request leaves her laptop, hits an internal gateway, runs through six microservices, gets stamped with a workload identity, and ends up - eventually - asking GPT-5 to read a PDF. No one at the bank built most of that path. A company in Milpitas, California did.

Tetrate is that company. It is not famous in the way consumer software is famous. It will not show up in a Super Bowl ad. But if your bank is moving its mainframe workloads to Kubernetes, or your government agency is trying to standardize on Istio, or your engineering team is comparing the cost of Claude versus Llama versus Grok in real time - there is a very good chance Tetrate is in the room, holding the wiring diagram.

The pitch, stripped of jargon, is this: modern software is a conversation between thousands of small services, and somebody has to make those conversations safe, fast, observable, and - in 2026 - aware of which AI model they're routing to. Tetrate wants to be that somebody. Quietly. Reliably. At Fortune 500 scale.

02The Problem They Saw

Back in 2018, a curious thing was happening at large enterprises. They had spent a decade breaking up monoliths into microservices. The promise was agility. The reality was a thicket. Every team built its own retries, its own auth, its own logging. Security was inconsistent. Latency was unpredictable. Engineers spent more time writing plumbing than features.

Google had a name for the answer: service mesh. The idea was elegant - take all that plumbing out of the application and push it into a sidecar that lives next to every service. Two open source projects, Istio and Envoy, became the de facto implementation. Both were extraordinarily powerful. Both were extraordinarily annoying to run in production.

That gap - between "Istio is the future" and "Istio is on fire in our staging cluster at 2 a.m." - is the problem Tetrate exists to solve. Not by selling something better than Istio. By selling Istio the way enterprises actually need to buy it: hardened, FIPS-validated, supported around the clock, and wrapped in a management plane that an operations team can use without a PhD in Envoy filter chains.

03The Founders' Bet

Varun Talwar co-created Istio at Google. He also co-created gRPC, the protocol your microservices probably talk in. Jeyappragash Jeyakeerthi - JJ to almost everyone - ran the cloud infrastructure platform at Twitter back when Twitter still scared people for technical reasons rather than political ones. In 2018 they bet that the open source projects they had helped create were going to win, and that the real money would be made not by inventing a competitor but by being the most trustworthy company to operate the thing in regulated environments.

It was an unfashionable bet at the time. The fashionable bet, in 2018, was to be the next consumer unicorn. Talwar and JJ went the other direction. They raised $12.5M from Dell Technologies Capital, 8VC, Intel Capital and Samsung NEXT. They hired Envoy maintainers. They got obsessive about U.S. federal compliance, which is approximately as glamorous as it sounds. Three years later, in March 2021, Sapphire Ventures led a $40M Series B, bringing total funding to $52.5M.

That capital bought time, which is what infrastructure companies actually need. You cannot will an enterprise sales cycle to be shorter than a Fortune 500 procurement cycle. Tetrate, refreshingly, did not pretend otherwise.

04The Product, Or Rather, Products

There are roughly two Tetrates today. The first sells service mesh. The second sells AI routing. They share a control plane, a code lineage, and a worldview, which is that traffic is traffic and someone needs to govern it.

The Agent Router is the headline act of the moment. It is, in essence, a fleet of Envoy AI Gateways with a credit card on file. Developers point their OpenAI-compatible client at a single Tetrate endpoint. Tetrate handles the keys, the failover, the traffic splits between models, the cost telemetry. You pay the model provider's cost plus five percent. That number - 5% - is doing a lot of strategic work. It is small enough to be easy to defend in procurement, and large enough to compound interestingly if every enterprise developer in the world starts routing AI calls through one pipe.

05The Proof

Infrastructure companies live or die by their reference logos. Tetrate's are the kind that take eight months of paperwork to land and then never leave: FICO, the U.S. Department of Defense, a parade of large financial institutions that Tetrate is generally too polite to name in marketing. In early 2023 the company reported a 50% quarter-over-quarter jump in customer logos - a number that sounds modest until you remember that each one represents a multi-cluster Kubernetes rollout at an organization that did not used to allow developers to make decisions on their own.

Partnerships tell a similar story. Tetrate is in the AWS Marketplace, on the Red Hat OpenShift catalog, and distributed to U.S. federal customers through Carahsoft. In February 2025 it co-released the Envoy AI Gateway with Bloomberg - a collaboration which, translated, means "Bloomberg's engineers thought this was important enough to put their name on a CNCF project." That is not a small endorsement.

06The Mission

Tetrate's stated mission is to make application networking and security boring. In a category that loves the word "revolutionary," that is a refreshing thing to say out loud. Boring, in this context, means "standard." It means a developer at a regional insurer can pick up the same control plane a federal agency uses, configure it the same way, and trust that the people behind it have skin in the upstream open source community.

The open-core thesis is unfussy: contribute aggressively to Istio, Envoy and Envoy Gateway; sell the hardened distribution, the management plane, the support contract, and now the AI routing service to anyone who would rather not run any of it themselves. Nothing about the strategy is novel. Plenty about the execution is.

07Why It Matters Tomorrow

Every large company is about to become, whether it likes it or not, a model shop. It will use OpenAI for one thing, Anthropic for another, an open-weight model for the third because the lawyers said so. Somebody has to route that traffic. Somebody has to log it, govern it, cache it, fail it over, and explain to a finance committee why the inference bill doubled in March.

Tetrate is making a credible argument that the somebody is them - and it is credible because the same company already does the equivalent job for non-AI traffic at organizations that do not change vendors casually. The Agent Router is not a pivot. It is the same idea, applied to a new kind of packet.

Back to the Wednesday morning. The developer's chatbot returns its summary. The bank's compliance team has a log of which model was used, how much it cost, and which workload identity made the call. The developer never wrote that part. She probably does not know the word "Envoy." Tetrate is fine with that. Quiet plumbing is the whole point. The water shows up, the lights stay on, and somebody in Milpitas keeps tightening the joints.

08Links & Receipts