Built for your work. Run on your terms.
Map the workflow before choosing a model. We build agents around the systems your team uses and the decisions people still need to make. Run them in your environment or ours. The agents we build are yours.
From production: Auxi Labs moved from rigid workflow definitions to workflows created in plain language. Read the case study →
From manual step to managed workflow
An invoice arrives with a missing purchase order.
The awkward cases decide whether an agent is useful. We build around the exceptions and approval rules your team deals with every day.
Explore the workflows we buildThe agent reads incoming documents against your rules. It flags missing fields for review before writing to your system of record.
For example: Invoices and purchase orders · Contract and form review · Email attachment processing
A useful answer shows its source. We connect agents to the documents and databases your team trusts, then apply the same access rules people already follow.
For example: Operational questions · Recurring reports · Cross-system research
The agent moves work between your existing tools. You choose which actions need a person to approve them. Every step leaves a trace.
For example: Customer onboarding · Approval routing · System updates and reconciliation
How we work
We follow the work from start to finish with the people doing it. The handoffs and workarounds tell us what to build.
We decide what the agent does and where it runs. Before coding, we set approval rules and pick a measure your team can check after launch, such as time saved.
Our engineers connect the agent to your systems and test it with real examples, including the awkward cases. Your team sees the failures and fixes as we go.
We launch in the environment you choose and watch how the agent performs. We make changes based on real runs. The agent is yours, and we agree on support terms with you.
Your deployment choice sets where the runtime and its records live. The model choice is a separate decision: a hosted model sends requests to its provider; a model in your environment keeps those calls there.
Inside your environment
We deploy into your AWS, Azure or GCP account under your network and access controls. Your team reviews the data flow and monitors the runtime with its existing tools.
On-premise, including environments with no internet connection at all. For when the answer to "can this data leave the building" is simply no.
Operated by Koru
We host and operate the agent for you. We document the architecture and data flow so your team knows what Koru operates.
The agent we build is yours. Choose its model, including one hosted on your own hardware. We document the agent and its connections so your team has the details it needs to operate it.
See how deployment worksKoru Runtime records each model call, tool call and handoff. We test changes against representative cases before release and set approval rules for live actions. Your team sees the result in the execution trace.
About us
The engineers you meet in your first session stay involved through launch. They work with your operations lead on the workflow and with your technical team on the architecture.
Co-founder & CEO
Former Principal Engineer at Okta & Expedia.
Co-founder & CTO
Former Engineer at AWS and 2x founding engineer.
Bring one workflow, including the workarounds. We follow it with the people doing the work. Together we mark where approval is needed and choose a measure for success. A clean demo tells us less than a real exception.
Run agents in your cloud account, your data centre or an environment managed by Koru. We document where the runtime and its records live. If you choose a hosted model, its provider receives the data sent in model calls. A model running inside your environment keeps those calls there.
Choose a deployment on your infrastructure with no internet egress and an open-weight model inside your network. We map the data flow with your team before settling on that design.
You own the agents we build for you. Koru Runtime is the platform that runs them. Your agreement explains where the platform runs, who can access it and what support Koru provides.
You choose the model for each job. Koru Runtime supports hosted providers such as OpenAI, Anthropic and Google. It also supports open-weight models on your own hardware. We test how well each option works for your task and review its cost and data path with you.
We map the workflow and pick a result to measure, such as time saved per case. Then we test the build with real examples, including exceptions. The schedule depends on the integrations and where the agent runs.
We agree on access and data flow before connecting systems. For a customer-hosted deployment, the runtime stays in your environment and uses scoped credentials. For a managed deployment, the agreed data is processed in the Koru-managed environment. Hosted model calls still go to the provider you choose.
You retain ownership of the agents we built for you. Your agreement sets the terms for Koru Runtime and ongoing support. We hand over documentation for the agent and its deployment so your team has a clear record of how it runs.
Most engagements are scoped in two-to-four-week increments rather than as one big project. If priorities change, we re-scope at the next checkpoint. You're not locked into a six-month plan that no longer matches your business.
A useful first conversation
Bring the workflow people actually follow, including the workarounds. We will trace where it gets stuck. Then we will agree on the approvals and one result to measure after launch.
Talk directly with the engineers who build the agent.