Oz

Oz provides a comprehensive cloud agent infrastructure with four key components: Host, Connect to Context, Track, and Launch. The Host component determines where agents run—either self-hosted in your own Docker or Kubernetes sandboxes or managed by Warp. Connect to Context integrates agents with MCP servers, skills, memory, and secrets, ensuring each agent has the right information to work effectively. Track offers full logging and auditing of agent sessions, which can be shared across the team for greater transparency. Finally, Launch enables starting cloud agents from existing tools like terminals or chat platforms. This infrastructure is designed to make agents truly cloud-native, running in isolated, scalable environments that can be spun up on demand. The combination gives teams full control over agent execution environments while maintaining security and compliance.

Beyond harness flexibility, Oz features automatic multi-model routing that spans frontier and open-weight models. Rather than hardcoding a single AI model, the platform dynamically routes each task to the most appropriate model based on quality and cost considerations. For instance, a simple code formatting request might go to a cheaper, faster open-weight model, while a complex architecture design could be routed to a frontier model like GPT-4. This intelligence is built into the Warp Agent but also available when using other harnesses. Additionally, Oz integrates with any tool through a programmable platform with primitives for building custom triggers. Agents can be launched from within any tool via specific events, such as an @warp mention in Slack or a CI/CD pipeline hook. This open integration model ensures Oz fits into existing workflows without forcing teams to change their tools.

Oz operates as a cloud-based agent orchestration layer that turns your development pipeline into a software factory. The platform allows you to define and manage agents that run every stage of the SDLC—from code review to bug investigation, refactoring, and incident response. Each agent runs in an isolated cloud environment, connected to your codebase, MCP servers, and internal tools. A central dashboard provides a single pane of glass where you can see all agents running across your team, join ongoing sessions with one click, and audit agent actions in real time. The platform measures agent performance centrally, allowing the system to improve with each run. Teams retain control over decisions that matter—approving changes, escalating issues, and setting policies. The workflow is designed for seamless handoff between cloud and local development, with the Warp Terminal providing a unified experience for developers who want to interact with agents directly.

Oz enables specific, high-impact use cases across the SDLC. In code review, teams can run a first-pass review on every pull request using agents, catching style issues, potential bugs, and security vulnerabilities before human reviewers look at the code. Bug investigation becomes more systematic: agents can reproduce bugs, analyze logs, and suggest fixes, routing the most likely solution to the appropriate developer. Refactors and migrations benefit from agents that scope changes, migrate code, and validate results automatically, reducing the risk of breaking existing functionality. During incident response, agents can investigate alerts, summarize root causes, and propose next steps, accelerating mean time to resolution. Real-world outcomes include Rectangle Health using Oz to build an AI teammate that writes its own code, demonstrating the platform’s potential for self-improving automation. Teams report faster shipping, fewer manual errors, and better traceability across the development process.

Oz is designed for software developers, engineering teams, and enterprise organizations that want to scale AI agent usage beyond individual coding assistants. It is particularly valuable for DevOps, SREs, and AI/ML engineers who need to automate multi-stage SDLC workflows. The platform integrates with Docker and Kubernetes for hosting, and supports MCP servers for extensibility. Oz can be used standalone or in combination with Warp Terminal and Warp Agent for a fully integrated experience. Pricing is available on request via contacting sales, with enterprise plans that include security and compliance features for regulated industries like financial services, insurance, and telecommunications. The platform is SOC 2 certified. In summary, Oz provides the most flexible and scalable way to orchestrate cloud agents, turning ad-hoc AI coding aid into a programmable, auditable software factory that improves continuously.