Today, TensorFoundry officially launches its website and opens its doors to the community. We're a Melbourne-based company with a straightforward mission: make it practical for organisations to run large language models on their own infrastructure - no cloud dependency, no data leaving your premises, no vendor lock-in.
Why We Built This
The past few years have seen extraordinary advances in large language models, yet most deployment options funnel your data through third-party cloud providers. For many organisations - particularly those in regulated industries, government, healthcare, and finance - that simply isn't acceptable.
We believe the future of AI is private-first: intelligent systems that run where your data already lives, under your control, with performance that rivals hosted solutions. That conviction is the foundation everything we build rests upon.
Introducing Our Product Suite
Olla
An open-source LLM router and proxy that intelligently distributes inference requests across your GPU fleet. Already available on GitHub with early adopters downloading and deploying it in production environments. Olla handles load balancing, failover, and model routing so your applications don't have to.
Learn more about Olla →FoundryOS
Our enterprise-grade orchestration platform for managing AI workloads at scale. FoundryOS brings fleet management, health monitoring, and intelligent scheduling to your on-premises LLM infrastructure. Currently in active development, with early access available for qualifying enterprise customers.
Learn more about FoundryOS →AgentOS
An agentic workflow system designed to coordinate AI agents and integrate with your enterprise software. AgentOS is currently in development and will bring multi-agent orchestration, enterprise integrations, and workflow automation to the TensorFoundry platform.
Learn more about AgentOS →Privacy First, From the Ground Up
Every product decision we make is filtered through a simple question: does this keep our customers in control of their data? We don't offer hosted inference. We don't aggregate telemetry. Our tools are designed to operate entirely within your network perimeter, which means your prompts, your outputs, and your models stay yours.
For Olla, this philosophy is baked into the architecture from day one - it runs as a local proxy, routing traffic between your applications and your LLM backends without ever touching external services.
Based in Melbourne, Building Globally
We're headquartered in Melbourne, Victoria, and proud to be building world-class AI infrastructure from Australia. Our engineering team has deep experience in distributed systems, GPU infrastructure, and enterprise software - and we're growing.
The Australian and Asia-Pacific markets are where we're starting, but the problem we're solving is global. Organisations everywhere are asking how they can benefit from modern AI without surrendering control of their most sensitive data.
What's Next
Over the coming months we'll be releasing Olla updates on a regular cadence, opening the FoundryOS waitlist, and sharing more detail about our roadmap. Follow us on GitHub and subscribe to our newsletter to stay informed.
If you're evaluating on-premises LLM infrastructure, we'd love to hear from you. Reach out at hello@tensorfoundry.io.
Get Started Today
Olla is open source and ready to deploy. FoundryOS and AgentOS are in active development - join the conversation on GitHub or sign up for updates.