OOz by Warp

Oz Runs Hundreds of Warp Coding Agents in Parallel

More than 700,000 developers use Warp, and on February 10, 2026 the company gave them Oz, a cloud platform that runs hundreds of coding agents at once. Each agent gets its own Docker sandbox, can change several repositories and leaves an audit trail anyone on the team can open from a link.

Oz suits engineering and platform teams with recurring code chores: feature-flag cleanup, documentation refreshes, security fixes and first-pass incident work. The Free plan includes limited cloud agent access, and Build costs $20 a month with $20 of usage at API rates. The main open question is scale, since Warp does not publish Oz concurrency limits per plan. Oz scores 9.0/10 on our scale.

Warp’s Users, Backers and Oz Launch Figures

The launch release, Warp’s pricing page and Product Hunt supply these facts.

Zach Lloyd said at launch that Oz “moves agents into the cloud so companies can run hundreds of agents safely and with full oversight.” The word to weigh there is oversight. Every Oz run has a link and an audit trail, which matters to a platform lead who has to explain to security what an agent changed and why.

Warp’s user base gives Oz a head start. Teams already on the Warp terminal can start cloud agents from the same app without adopting a new vendor.

Inside an Oz Agent Run on Warp

Oz separates where an agent is started from where it runs. A developer can start it from the Warp desktop app, the web, a phone, the CLI or Warp’s SDK. The work itself happens in a cloud sandbox.

Isolated Docker Sandboxes With Team Context

Each agent runs in a sandboxed cloud environment powered by Docker. The environment can index several GitHub repositories, so one agent can make a change that spans a service and its client library. Agents also share team context, such as coding conventions, across runs.

Interactive Runs and Scheduled Jobs

A developer can start an agent by hand to fix one bug. A program can start agents through the API, for example when a new issue arrives. Recurring jobs run on a schedule. Warp names feature-flag cleanup, documentation updates, security checks and incident response as typical scheduled work.

Links and Audit Trails for Every Run

Every agent run gets a shareable link and a full audit trail, open to teammates with access through the CLI or API. A reviewer can read what the agent did before approving its pull request. This record is the part of Oz most useful to security and compliance teams.

Apps Built on Top of Agents

Warp encourages teams to build their own tools on Oz. Two examples from the launch are a bug triage system and an incident response system. In each case, an outside event starts an agent, the agent investigates in a sandbox and a person reviews what it found.

Hosted or Self-Hosted

Oz runs on Warp’s infrastructure by default. Enterprise customers with strict network or security rules can host cloud agents on their own infrastructure and route model calls through their own cloud.

Where a Person Steps In

Oz agents propose changes. A developer still reads the pull request, checks the audit trail and decides whether to merge. For incident work, the agent gathers logs and suggests a cause, and the on-call engineer decides on the fix. Teams can set branch protection in GitHub so agents can only open pull requests.

A Weekly Feature-Flag Cleanup, Step by Step

A platform team writes one instruction: find flags fully rolled out for 30 days, remove the dead code paths and open a pull request per repository. They schedule it for Monday mornings in Oz. Each week, Oz starts an agent per repository in parallel. Each agent opens a pull request with its changes, and the team lead reviews the audit trail and merges the ones that pass tests.

Warp Free, Build, Max, Business and Enterprise

Warp sells one set of plans that covers the terminal, local agents and Oz cloud agents. Paid usage is charged at model API rates.

What a 10-Person Platform Team Would Pay

Ten developers on Business with annual billing pay $45 each, or $450 a month. That includes $200 of pooled agent usage at API rates, plus SAML SSO and bring-your-own keys.

A team that brings its own model keys pays the model provider directly. In that case the Warp seat cost of $5,400 a year covers the platform, and model spend sits on the team’s own cloud bill.

A single developer who runs agents heavily can compare Build and Max. Build at $18 a month on annual billing gives $20 of usage. Max at $180 gives $200 of usage, so it only pays off for someone who would otherwise top up Build by more than $160 a month.

Spend Caps for Large Agent Batches

Parallel agents can burn through usage fast. If 100 agents each spend 50 cents of model time on a cleanup job, one run costs $50, and a weekly schedule turns that into about $200 a month. Build and Max include a team-wide spend cap and auto-reload, so admins can set a ceiling before scheduling large jobs. We suggest starting any new scheduled job on a few repositories and reading its cost per run before widening it.

Warp Factories, a Separate Price List

Warp’s pricing page also lists Warp Factories, billed separately for inference, compute and agent usage. Its pay-as-you-go tier charges 20% above API rates, while its $20 Build tier charges at API rates and adds Claude Code and Codex harnesses. Buyers should confirm which product a quote covers.

Oz Next to Devin, GitHub Copilot Cloud Agent and OpenHands

All four run coding agents that work on their own and return pull requests. They differ in where agents run and how they are billed.

Oz’s difference is fleet management. Devin and Copilot focus on one agent per task. Oz is designed to start many agents at once, on a schedule, with a record of each run. OpenHands offers the most control for teams that want to own the whole stack, at the cost of running it. More choices sit in our coding agents and IT and DevOps agents categories.

Security Questions to Put to Warp Before Rollout

Warp states that it is SOC 2 compliant and keeps zero data retention with all of its contracted model providers. Business adds SAML SSO and bring-your-own API keys. Enterprise adds self-hosted agents and routing of model calls through the customer’s own cloud.

A security review should still cover a few points the public pages leave open. Ask which secrets an agent sandbox can reach, how long sandbox data is kept after a run, who can open audit-trail links, and whether agents can push to protected branches or only open pull requests. Written answers to those four questions will settle most internal approvals.

Teams Ready to Run Coding Agents at Fleet Scale

Oz fits platform teams that own dozens of repositories and a backlog of small, repetitive changes. Dependency bumps, flag cleanup and doc updates are dull for engineers and well suited to agents running in parallel.

It also fits SRE and security teams that want agents to take a first look at incidents or scanner findings. The audit trail gives them a record to review. Companies already on the Warp terminal have the shortest path, since Oz is built into the app they use.

Who Might Wait Before Adopting Oz

A small team with a handful of repositories may not need fleet-scale agents. A single coding agent in the IDE may cover their needs at lower cost. Teams larger than 25 seats need Enterprise, so they should budget time for a sales process.

Teams that want a mature review history may also wait a few months. Warp’s Product Hunt reviews mostly cover the terminal, and Oz-specific feedback is still thin. Public case studies with numbers, such as pull requests merged per week by Oz agents at a named customer, would give buyers more to go on.

Teams that need firm numbers on concurrency, run time or sandbox size should ask Warp for them in writing. The pricing page describes access levels as “limited” and “extended” and does not list Oz quotas per plan.

What Product Hunt Reviewers Say About Warp

Warp holds 4.8 stars from 79 Product Hunt reviews. Reviewers praise speed, polish and block-based output, and many say Warp replaced their old terminal. Several value the AI help for writing commands and say the team responds to feedback.

Complaints are minor: a sign-in step and survey during onboarding, settings search that does not always work, some keybindings and display issues with TTY-heavy programs. Most of these reviews predate Oz. The Oz launch thread had questions about recovering from mid-run failures and joining live sessions, which Warp could answer in its docs.

Oz and Warp Agent Milestones in 2026

• February 10, 2026: Warp launches Oz, its orchestration platform for cloud coding agents.
• February 2026: Oz launches on Product Hunt, hunted by Chris Messina.
• 2026: Warp’s pricing page adds Max, Business seat limits and a separate Warp Factories price list.

Oz by Warp Earns Its Score on Oversight

Oz scores 9.0/10. It scores well on autonomy and security: agents run in isolated sandboxes on schedules, every run leaves an audit trail, and Enterprise customers can self-host. Pricing is published with usage at API rates. Integrations and sentiment score lower because Oz-specific reviews and named connectors are still few.

For a platform team that wants to hand recurring code chores to many agents and keep a record of each one, Oz is a sound pick. Start on Build, schedule one chore and review a few weeks of audit trails before moving the team to Business.

Platform Leads Ask These Questions About Oz

How Many Agents Can Oz Run at Once?

Warp says Oz runs hundreds of agents in parallel. It does not publish a per-plan concurrency number, so teams planning large batches should ask Warp to confirm their limit.

Is Oz a Separate Product From the Warp Terminal?

Oz is built into the Warp terminal but also works standalone. Teams can start agents through the Oz CLI, the API or the SDK without using the terminal.

Can Oz Agents Run Inside Our Own Network?

Yes, on Enterprise. Warp offers self-hosted cloud agents on the customer’s infrastructure and lets Enterprise customers route model calls through their own cloud.

How Is Oz Agent Usage Billed?

Paid plans include a monthly usage amount charged at model API rates: $20 on Build, $200 on Max and $20 per seat on Business. Build and Max add a team-wide spend cap and auto-reload.

Can One Oz Agent Change Several Repositories?

Yes. Oz sandboxes can index many GitHub repositories, and Warp says agents can make multi-repo changes in a single run, such as updating a service and its client library together.

Does Warp Train Models on Our Code?

Warp says it is SOC 2 compliant and has zero data retention policies with all of its contracted model providers.

What Kinds of Jobs Suit Oz?

Warp names feature-flag cleanup, documentation updates, security checks and incident response. Any repeatable code change across many repositories fits the same pattern.