Nori vs Warp Factories and Oz

Quick answer

It depends on how you want to describe the work. Choose Warp if you want agent pipelines defined as code in Warp Factories and Oz, and you accept usage billed at API rates with the control plane at Warp. Choose Nori if you want a team workspace: Jiro or any harness in Slack, schedules and webhooks, projects, shared memories and skillsets, on your own Claude or ChatGPT subscription, with flat plans and no token charges.

What is Warp?

Warp Factories and Oz are Warp's products for running coding agents in the cloud. The idea behind a factory is a pipeline defined as code: you describe the stages of the work and which agent handles each one, and the definition is reviewed and versioned like any other file in the repository.

As of November 2026, Warp's own pages establish these points.

• Harnesses. Factories can run Warp Agent, Claude Code or Codex.
• Inference path. When a stage uses Claude Code or Codex, inference goes directly to the model provider. Warp Agent is different: its inference does not take that direct path (Warp authentication docs).
• Pricing. Pay-as-you-go factory usage is billed at API rates plus 20%. Build is $20 per month and Business is $50 per user per month; both charge usage separately at API rates (Warp pricing).
• Storage and control plane. Customer-owned storage can hold selected data, and the control plane stays with Warp (Warp infrastructure and security docs).

Our October page also described a closed beta, a list of ticket and chat triggers, and SSO tiers. The November review did not re-establish those points, so this page does not state them. Check Warp's site for the current status.

What is Nori?

Nori is a cloud platform for coding agents, built for engineering teams. Work starts where the team already talks. Someone writes @nori flaky test in checkout_spec.rb, find the cause and open a PR in a Slack thread; Nori starts an isolated cloud sandbox with the team's repositories, skillset and scoped credentials, and the pull request link comes back in the thread. The same sessions start from Discord, the web, the CLI, cron schedules, one-time reminders and webhooks, and they share one run history with approval paths.

Longer work lives in a project: a coordinator that remembers the goal, shared docs, and subworkers that each take a piece. Every session starts with the organization's skillsets (CLAUDE.md or AGENTS.md, skills, subagents), shared memories and scoped integrations. Nori runs any harness (Claude Code, Codex, Gemini CLI, Cursor, OpenCode, Droid, Amp or any ACP agent) on your own Claude or ChatGPT subscription, API keys or LLM gateway, and never charges for tokens.

Meet Jiro

Jiro is the AI engineering teammate that Nori ships, a ready-made alternative to designing a pipeline first. It runs on Claude Code or Codex with your own subscription or key, and it follows an engineering workflow of planning, test-driven development and review. In its first days it reads your repositories, asks the team how you ship, and sets up the workspace. See jiro.bot.

Nori vs Warp Factories and Oz, head to head

"Not documented" means the Warp pages we read do not establish it. It does not mean Warp lacks it.

Inference paths and billing, with a worked example

Both products let you run Claude Code or Codex, so harness choice alone does not separate them. The bill does. The numbers below are an illustration, not a quote: assume a team's agents use $1,000 of model usage at API list rates in a month.

• Warp, pay as you go. Factory usage is billed at API rates plus 20%, so that month is $1,200 of usage.
• Warp Build or Business. Usage is charged separately at API rates, so it is $1,000 of usage plus $20 per month for Build or $50 per user per month for Business.
• Nori Team. $250 per month for five machines with unlimited members. Nori bills nothing for tokens. The model usage runs on the Claude or ChatGPT subscriptions the team already pays for, or on its own API keys or gateway at whatever rate the team has with that provider.

The practical difference is who the usage goes through. For how each vendor treats this, see do cloud coding agents mark up token costs.

Enterprise deployment: where does the control plane run?

The control plane is the service that decides who may start a session, which credentials it gets, where it runs and what is kept afterwards. Warp's infrastructure and security docs say you can keep selected data in storage you own, and that the control plane stays with Warp.

That split is common. As of November 2026, Cursor's self-hosted workers execute tools on your machines while Cursor retains planning and inference, and Claude Code's customer-run environments, a public beta for Team and Enterprise, retain Anthropic control. Moving storage or runners is not the same as moving the control plane; what an agent control plane is draws the line for each vendor.

Nori deploys the full stack, control plane included, into your own AWS VPC or Azure VNet, or on-prem, with your identity provider (SAML, SCIM, OIDC), your LLM gateway and your compute. On hosted plans, Nori gives each organization a dedicated, single-tenant control plane and data store, and every session runs in its own isolated sandbox.

When to choose Warp

• You want the pipeline itself in version control. If your goal is a fixed assembly line that you review and change like infrastructure code, that is what a factory is for. Nori automations start sessions from a schedule or a webhook and Nori projects coordinate subworkers; neither is a declarative pipeline file.
• Three harnesses are enough. Warp Agent, Claude Code and Codex cover what your team runs, and you want Warp Agent in particular.
• You prefer usage-based billing. Pay as you go bills usage at API rates plus 20%, and Build is $20 per month with usage on top. For light or irregular use that can cost less than a flat plan.
• Selected data in your own storage is enough. If your security review accepts a vendor-run control plane, Warp's storage option may cover what you need.

When to choose Nori

• You want the team workflow before you design a pipeline. Slack and Discord, schedules and webhooks, projects, shared memories and skillsets work together from the first session.
• You already pay for Claude or ChatGPT. Nori runs every harness on those subscriptions, on API keys or on your gateway, and never charges for tokens.
• You want a predictable bill. Flat plans by machine. Heavier agent use does not change what you pay Nori.
• You want a teammate to start with. Jiro plans, works test-first and reviews its own changes without any stage design.
• The control plane has to be yours. Nori Enterprise runs in your VPC or VNet with SAML and SCIM from your identity provider.

See how the other vendors line up in cloud coding agents compared, or read Nori vs Amp Orbs for another product that reuses your subscriptions.

How to switch from Warp to Nori

A factory definition is a good migration checklist, because it already names the stages, agents and tools.

1. Read your factory file stage by stage. For each stage write down the harness, the prompt or skill it uses, the tools it calls and what starts it.
2. Start the Nori trial. Thirty days, no card. Pick the Developer or Team plan.
3. Connect models. Sign in with your Claude or ChatGPT subscription, or add the API keys or gateway you use today.
4. Keep the same harness. Stages that run Claude Code or Codex run unchanged on Nori. Stages that ran Warp Agent move to Jiro or another harness.
5. Turn stages into skills. A triage stage becomes a triage-ticket skill in the org skillset; standing rules ("never touch migrations/ without a reviewer") go into AGENTS.md or shared memories.
6. Recreate the triggers. Scheduled runs become cron automations, for example 0 6 * * 1-5 for a weekday 06:00 triage. Event-driven runs become webhook automations. A multi-stage line becomes a project whose coordinator hands each stage to a subworker.
7. Run both for a week. Send the same tickets through both, compare the pull requests and the month's usage bill, then decide.

A recreated trigger looks like this:

name:      weekday ticket triage
schedule:  0 6 * * 1-5   (Europe/Berlin)
agent:     Jiro
repos:     acme/checkout, acme/web
prompt:    Use the triage-ticket skill on every Linear issue labelled
           needs-triage. Write a spec comment, then open a PR for any
           issue estimated under one day.
deliver:   Slack #eng-triage

Frequently asked questions

Sources

Competitor facts were read from the vendor pages below on November 11, 2026. Vendors change plans often; check the source before you decide.

• Warp pricing (pay as you go, Build, Business)
• Warp docs: Factories infrastructure and security
• Warp docs: harness authentication
• Cursor docs: self-hosted cloud agents
• Claude Code docs: self-hosted environments