Oz

What is Oz?

Oz is the orchestration platform for cloud agents. It coordinates and runs agent sessions at scale, with a programmable control layer that lets you steer how agents operate, track their activity, and run them across different triggers.

In practice, Oz lets you start cloud coding agents and manage their lifecycle through a unified control plane. Agents are described as being based on “Skills,” and Oz supports building custom agent automations that can be scheduled or triggered via APIs.

Key Features

• Unified orchestration for cloud agents: Spin up agents in the cloud and manage them through a single control plane that supports multiple entry points (web, CLI, and mobile).
• Programmable automation via Skills: Oz agents are based on Skills, and you can “turn” agent activity into automation flows that run with specific behaviors.
• Scheduling and triggers (cron-like): Schedule agents to run like cron jobs and have them report results back in the way you specify.
• Start and join agent sessions across surfaces: Start agents from the Warp app, the web, or your phone, and join sessions with one click from the CLI, web, mobile, or Warp.
• Flexible hosting: Run agents on your infrastructure or on Oz’s infrastructure (“our infra”).
• Multi-repo change coordination: Use a single agent to work across source code across repositories and make coordinated changes in one go.
• Programmable stack (CLI, SDK, API): Use the Oz CLI, SDK, and API to build and integrate custom automations.
• Multi-model support and standards: Oz includes models such as Claude, Codex, and Gemini, and supports industry standards like Skills for onboarding.

How to Use Oz

1. Set up an automation or agent workflow: Use the Oz CLI, SDK, or API to define how agents should run, based on Skills.
2. Choose a trigger method: Start agents from the Warp app, the web, or your phone, or configure them to run on schedules (cron-like) and other API-triggered flows.
3. Run agents on your chosen infrastructure: Decide whether to host agents on your infrastructure or Oz’s infrastructure.
4. Monitor and join sessions: Use the unified control plane to observe agent activity and join ongoing sessions from the CLI, web, mobile, or Warp.

Use Cases

• Scheduled coding or maintenance jobs: Set up Oz agent automations to run at specific times (cron-like) and receive reports on how the changes should be made.
• Event-driven workflows via API triggers: Start agent sessions when a webhook or API event occurs, then coordinate the resulting work in the cloud.
• Coordinated multi-repo changes: Use one agent to apply coordinated edits across multiple repositories rather than handling them repo-by-repo.
• Custom agent automation building: Implement bespoke agent workflows using the Oz CLI, SDK, and API, where your automation wraps one or more Skills.
• Team usage with shared session access: Have users start, track, and join agent sessions from different interfaces (CLI/web/mobile/Warp), supporting collaboration around the same agent activity.

FAQ

• What are “Skills” in Oz? The page states that Oz agents are based on Skills and that Oz supports Skills for quick onboarding.

• How do I start an Oz agent? Oz supports starting agents from the Warp app, the web, or your phone, and joining sessions via the CLI, web, mobile, or Warp.

• Can I schedule agents to run automatically? Yes. Oz can schedule agents to run like cron jobs and report back according to how you’d like.

• Where can Oz run my agents? Oz supports flexible hosting: you can run agents on your infrastructure or on Oz’s infrastructure.

• Can Oz make changes across multiple repositories? Yes. The page mentions one agent, multi-repo changes to make sweeping coordinated changes in one go.

Alternatives

• Generic workflow schedulers (cron + custom scripts): You can schedule scripts that call agent logic, but you would need to build orchestration, session tracking, and coordination yourself.
• Ad-hoc agent runners in a single environment (one-off cloud calls): Running agents directly via an API without a dedicated orchestration layer can limit how well you manage parallel runs and auditability across sessions.
• Other agent frameworks focused on prompt/skill execution: Frameworks that execute skills or agent steps can help with the “agent” part, but may not provide a unified control plane for coordinating cloud sessions, hosting options, and multi-surface session access.
• CI/CD pipelines for automated code changes: CI/CD can trigger automated changes and tests, but orchestration of interactive agent sessions and coordinating multi-repo edits is typically done with additional tooling.