Oz is an orchestration platform for cloud agents. Launch hundreds of cloud agents in minutes, from Warp, CLI or even your phone. Wake up to production-ready PRs.
Zach here - Founder and CEO at Warp. Our team is excited to launch Oz, the orchestration platform for cloud agents.
Oz makes it easy to scale up to hundreds of cloud agents, keep tasks running when you step away from your laptop, and turn agent skills into agent automations.
Why we built Oz and what it does
Developers today are running 3-5 local agents to fix bugs but once you try to go beyond that, things start to break down: your laptop hits capacity, you can’t see what your agents are doing, and agents start to be more trouble than they’re worth.
Deploy agents from anywhere: From the Warp desktop app, on the web, your phone, using Warp’s SDK, or even the CLI.
Isolated cloud environments: Set up isolated cloud environments for agents to run that can index as many GitHub repositories as you want.
Build apps on top of agents. Use CLI and API access to build bug triage systems, incident response tools, or any app that needs an agent backend.
If you're looking for a starting point, try automating something small - a recurring task that's tedious but straightforward and watch your productivity compound.
We can’t wait to see what you build and I’d be curious to hear the community’s thoughts on whether they think cloud agents are going to be the future?
Feel free to comment any questions/feedback you have below and one of our engineers will respond. ✌️
been a Warp fan since early beta — Oz looks like a massive step up for agent orchestration. 🪄 the parallel cloud execution is a game changer for long-running dev tasks. what's the most common automation you've seen devs build on top of Oz so far? 🚀
We've seen so many great things built on top of Oz, but some of the most common ones that people are starting with are around automating simple but time-consuming chores: auto-updating docs, triaging GitHub issues, running reports on a schedule. These are well-defined & understood problems that you can easily hand off to an agent to do on it's own.
And then the cool stuff is starting to show as well... OpenClaw alternatives, agents talking to each other to get stuff done, orchestrating third-party agents, and a bunch of others. Super excited to see what everyone is building and creating.
Report
@adam_lab@petradonka You have my attention at "OpenClaw alternatives." I have a major project on which I'd like to develop and deploy some AI agents, but I have had no idea how to get started. I've been looking at OpenClaw but hesitant to pull the trigger. It just seems like overkill and a security risk.
Report
@adam_lab@jay_bienvenu might want to take a look at https://moltbookagents.net which has a simple setup guide plus a guide to securing the agent in a self-contained NAS. Might be a quick solution
@shreya_chaurasia19 agreed! Oz agents run in isolated environments, only have access to what you give them access to, and normally don't make changes directly in prod (though they can), so that you remain in the loop.
Plus, these agents can also serve as the guardrails, so you can enforce best practices better with less effort.
Super proud of the team for rethinking how we start, drive, and finish tasks with agents from first principles. Warp built the right core abstractions early on, and it's paying off! Oz to the 🌙
When you say “hundreds of agents,” what are the core primitives you built to keep that manageable (visibility, run state, approvals, budgets, retries)—and which one ended up being the real unlock versus just adding more compute?
@curiouskitty This is a good q! In a nutshell, all of the above. I don't have all of the context, but I'd break this down into 2 core areas:
The agentic development "experience". These are UI bells and whistles meant to support managing several agents. Some neat things I've been using:
/conversations slash command to cycle through past conversations
Warp's lightweight code editor (and new global search panel!) to gather context around code and guide agents more intelligently
Warp's in-built notifications and agent management view when agents need your attention
Intelligent conversation forking, our code review panel (try adding comments on changes and sending them to an agent! it's magical), etc.
Our harness. We launched support for agent skills in this release, which uses progressive disclosure to expand an agent's capabilities without bloating its context window. We did something similar for MCP servers, see here. Re: approvals/budgets/retries, we give users a lot of control here. You can configure your command allowlist in settings, and turn YOLO mode on/off when you desire. I've personally liked using /plan mode and then letting an agent cook on YOLO mode once I'm satisfied :)
We have a bunch of content on our blog that dives deeper into how we built our harness; take a look if you're curious!
This is an opinion - but I don't think there's any one true unlock. We focused on building the best platform to code with agents from anywhere, and that includes a mixed bag of changes - from improving the in-app experience, to integrating cloud agents, and ofc to improving our harness. Let us know what you think!
Report
Congrats on the launch! Moving agents to isolated cloud environments makes sense once local setups hit limits. How do you handle coordination and state management across hundreds of agents so tasks don’t conflict, duplicate work, or drift from the original objective when running asynchronously?
Replies
Warp
Hey Product Hunt 👋
Zach here - Founder and CEO at Warp. Our team is excited to launch Oz, the orchestration platform for cloud agents.
Oz makes it easy to scale up to hundreds of cloud agents, keep tasks running when you step away from your laptop, and turn agent skills into agent automations.
Why we built Oz and what it does
Developers today are running 3-5 local agents to fix bugs but once you try to go beyond that, things start to break down: your laptop hits capacity, you can’t see what your agents are doing, and agents start to be more trouble than they’re worth.
Deploy agents from anywhere: From the Warp desktop app, on the web, your phone, using Warp’s SDK, or even the CLI.
Isolated cloud environments: Set up isolated cloud environments for agents to run that can index as many GitHub repositories as you want.
Build apps on top of agents. Use CLI and API access to build bug triage systems, incident response tools, or any app that needs an agent backend.
If you're looking for a starting point, try automating something small - a recurring task that's tedious but straightforward and watch your productivity compound.
We can’t wait to see what you build and I’d be curious to hear the community’s thoughts on whether they think cloud agents are going to be the future?
Feel free to comment any questions/feedback you have below and one of our engineers will respond. ✌️
Migma AI
been a Warp fan since early beta — Oz looks like a massive step up for agent orchestration. 🪄 the parallel cloud execution is a game changer for long-running dev tasks. what's the most common automation you've seen devs build on top of Oz so far? 🚀
Warp
Thanks for the kind words @adam_lab 🙌
We've seen so many great things built on top of Oz, but some of the most common ones that people are starting with are around automating simple but time-consuming chores: auto-updating docs, triaging GitHub issues, running reports on a schedule. These are well-defined & understood problems that you can easily hand off to an agent to do on it's own.
And then the cool stuff is starting to show as well... OpenClaw alternatives, agents talking to each other to get stuff done, orchestrating third-party agents, and a bunch of others. Super excited to see what everyone is building and creating.
@adam_lab @petradonka You have my attention at "OpenClaw alternatives." I have a major project on which I'd like to develop and deploy some AI agents, but I have had no idea how to get started. I've been looking at OpenClaw but hesitant to pull the trigger. It just seems like overkill and a security risk.
@adam_lab @jay_bienvenu might want to take a look at https://moltbookagents.net which has a simple setup guide plus a guide to securing the agent in a self-contained NAS. Might be a quick solution
Warp
Cloud agents feel inevitable. Curious how you balance flexibility with guardrails so things don’t spiral in production.
Warp
@shreya_chaurasia19 agreed! Oz agents run in isolated environments, only have access to what you give them access to, and normally don't make changes directly in prod (though they can), so that you remain in the loop.
Plus, these agents can also serve as the guardrails, so you can enforce best practices better with less effort.
Warp
Super proud of the team for rethinking how we start, drive, and finish tasks with agents from first principles. Warp built the right core abstractions early on, and it's paying off! Oz to the 🌙
Product Hunt
Warp
@curiouskitty This is a good q! In a nutshell, all of the above. I don't have all of the context, but I'd break this down into 2 core areas:
The agentic development "experience". These are UI bells and whistles meant to support managing several agents. Some neat things I've been using:
/conversations slash command to cycle through past conversations
Warp's lightweight code editor (and new global search panel!) to gather context around code and guide agents more intelligently
Warp's in-built notifications and agent management view when agents need your attention
Intelligent conversation forking, our code review panel (try adding comments on changes and sending them to an agent! it's magical), etc.
Our harness. We launched support for agent skills in this release, which uses progressive disclosure to expand an agent's capabilities without bloating its context window. We did something similar for MCP servers, see here. Re: approvals/budgets/retries, we give users a lot of control here. You can configure your command allowlist in settings, and turn YOLO mode on/off when you desire. I've personally liked using /plan mode and then letting an agent cook on YOLO mode once I'm satisfied :)
We have a bunch of content on our blog that dives deeper into how we built our harness; take a look if you're curious!
This is an opinion - but I don't think there's any one true unlock. We focused on building the best platform to code with agents from anywhere, and that includes a mixed bag of changes - from improving the in-app experience, to integrating cloud agents, and ofc to improving our harness. Let us know what you think!
Congrats on the launch! Moving agents to isolated cloud environments makes sense once local setups hit limits. How do you handle coordination and state management across hundreds of agents so tasks don’t conflict, duplicate work, or drift from the original objective when running asynchronously?
Warp
@vik_sh thank you! The basics of working with git repos applies to most of these use cases right now, with merge conflict resolutions, PRs, etc.
And we're also looking at more defined protocols for the agents to work together more productively.
Congratulations on the new launch! I’ve heard a lot of good things about you)
Warp
Thanks @mykyta_semenov_ !
The `shell` code is not working.
Warp
@itsmeblackops could you tell us more about which code you mean specifically? We'll get it fixed.