Oz is the engine behind Warp's cloud agents. While Agent Mode handles interactive terminal sessions, Oz takes agentic development to the next level: parallel cloud agents running on remote infrastructure, triggered by webhooks, cron schedules, Slack messages, or CI/CD pipelines. It is the orchestration layer that turns Warp from a developer tool into an automation platform.

Released under the MIT license alongside Warp's open-source announcement, Oz is designed to be embedded in any workflow. This guide covers the platform architecture, agent lifecycle, environment configuration, trigger patterns, the Oz CLI, and production deployment strategies.

Whether you are running 4 agents on the free tier or 40 on the Max plan, understanding Oz's orchestration model is essential for getting the most out of Warp's cloud capabilities.

1What Oz Does and Why It Matters

Most AI coding agents run locally on your machine, one at a time, in an interactive session. Oz breaks all three of those constraints. It runs agents on cloud infrastructure, supports dozens of concurrent agents, and can be triggered programmatically without human interaction.

This unlocks use cases that local agents cannot handle:

• Parallel code review - Spin up 10 agents to review 10 PRs simultaneously
• Automated testing - Trigger agents on every push to run test suites and report results
• Scheduled maintenance - Cron-triggered agents that run daily dependency audits, security scans, or cost reports
• Incident response - Webhook-triggered agents that investigate alerts, gather diagnostics, and propose fixes
• Release automation - Agents that generate changelogs, update version numbers, create tags, and deploy

Oz is to AI agents what Kubernetes is to containers: an orchestration layer that handles scheduling, resource allocation, lifecycle management, and observability at scale.

2Oz Platform Architecture

The Oz orchestrator sits between triggers and agents. When a trigger fires (webhook, cron, Slack message, CLI command), the orchestrator creates a task, selects the appropriate AI model, allocates compute resources, and spins up an agent in a configured environment. The agent executes the task, and results flow back to the dashboard with full audit trails.

3Cloud Agent Lifecycle

Every cloud agent goes through a defined lifecycle:

1. Trigger - An event (webhook, cron, manual) initiates the agent
2. Environment setup - Oz provisions a container with your Docker image, clones repos, and runs setup commands
3. Execution - The agent processes the task using the selected AI model, running commands and calling MCP tools
4. Observation - The agent reads outputs, reasons about results, and decides next steps
5. Completion - The agent reports results, generates artifacts (PRs, reports, notifications), and shuts down
6. Cleanup - The container is destroyed, and logs are persisted for audit

Each agent runs in an isolated container, so there is no risk of one agent interfering with another. The environment is reproducible: the same Docker image, repos, and setup commands run every time, ensuring consistent behavior.

4Environment Configuration

An Oz environment defines everything an agent needs to run your code remotely. This includes:

• Docker image - Base image with your language runtime, tools, and dependencies
• Repositories - Git repos to clone into the agent's workspace
• Setup commands - Commands to run after cloning (install dependencies, build, etc.)
• Environment variables - API keys, configuration values, and secrets
• MCP servers - External tool connections available to the agent

Environments are reusable across agents and triggers. You might have a "Node.js Staging" environment for testing, a "Python ML" environment for data pipelines, and a "DevOps" environment with AWS and Kubernetes tools pre-installed.

5Trigger Patterns: Webhooks, Cron & Slack

Oz supports multiple trigger types for different automation patterns:

6The Oz CLI

The Oz CLI (oz) lets you interact with cloud agents from any terminal, not just Warp. This is essential for CI/CD integration and scripting.

The CLI supports all Oz operations: creating environments, triggering agents, viewing logs, and managing team access. It is the primary interface for integrating Oz into automated workflows.

7Multi-Agent Orchestration

Oz's real power emerges when you run multiple agents in parallel. The Max plan supports up to 40 concurrent agents, each with 8 vCPU and 16 GiB RAM. This enables patterns like:

• Fan-out testing - Run your test suite across 10 agents, each handling a different test group
• Parallel PR review - Assign one agent per open PR for simultaneous code review
• Multi-repo updates - Update a shared dependency across 20 repositories simultaneously
• A/B implementation - Have two agents implement the same feature with different approaches, then compare results

The orchestrator handles resource allocation and scheduling automatically. If you request more agents than your plan allows, tasks are queued and executed as slots become available.

8Monitoring & Audit Trails

Every cloud agent action is logged and visible in the Oz dashboard. This includes:

• Commands executed and their output
• Files created, modified, or deleted
• MCP tool calls and responses
• AI model interactions (prompts and completions)
• Resource usage (CPU, memory, duration)
• Credit consumption per task

For enterprise teams, audit trails are essential for compliance and security review. Oz's logging ensures that every agent action is traceable, reviewable, and attributable to the trigger that initiated it.

9Oz vs Other Agent Platforms

10Why Lushbinary for Cloud Agent Orchestration

At Lushbinary, we design and implement cloud agent orchestration systems for engineering teams. We have production experience with Oz, including environment configuration, trigger setup, multi-agent workflows, and CI/CD integration. Whether you need a simple cron-triggered code review agent or a complex multi-agent deployment pipeline, we deliver.

❓ Frequently Asked Questions

📚 Sources

• Oz Platform Documentation
• Oz CLI Quickstart
• Oz Product Page
• Warp Pricing

Content was rephrased for compliance with licensing restrictions. Pricing and feature data sourced from official Warp documentation as of April 2026. Pricing may change - always verify on the vendor's website.