Cloud coding agents compared (October 2026): Nori, Devin, Factory, Claude Tag, Cursor, and more

Nine cloud coding agents, four questions: which agent runs, whose subscription pays, who charges for tokens, and whose infrastructure it runs on.

Which cloud coding agent should you use?

It depends on which agent you want and who pays for its tokens. If your team is committed to Devin, Droid, Claude, Cursor, or Grok, that vendor’s cloud product is the shortest path. If you want open source, OpenHands; pipelines as code, Warp Factories. If you want a ready-to-work AI engineer, Jiro, or any other harness, on your own Claude or ChatGPT subscription, keys, or gateway, with no token charges and Slack, Discord, cron, and webhook automations, choose Nori.

Every product below runs coding agents on cloud machines and hands back pull requests. They differ on four questions: which agent runs, whose subscription pays for the model, who charges you for tokens, and whose infrastructure it runs on. Facts are as of October 2026 and come from each vendor’s own pages, linked under Sources.

Cloud coding agents at a glance

Cloud coding agents compared, as of October 2026
Product	Harness	Your subscription?	Token charges	Your infrastructure	Control plane	Best for
Nori	Jiro (Nori’s own agent) or any harness: Claude Code, Codex, Gemini CLI, Cursor, OpenCode, pi, ACP agents	Yes: Claude or ChatGPT, plus keys or any gateway	None; flat plans	Enterprise: your compute, cloud or on-prem	Your AWS VPC, Azure VNet, or on-prem (Enterprise); dedicated single-tenant per organization (hosted)	Teams that want a ready-to-work agent or any harness, their own models, and opinionated defaults
Devin	Devin only	ChatGPT Go/Plus/Pro can pay for GPT usage; no Claude	Token quotas, then on-demand credits	Outposts run session machines on your infra; the agent brain stays in Cognition’s cloud	Cognition’s cloud; Dedicated is a Cognition-hosted single-tenant VPC	A single managed agent with broad integrations
Factory	Droid only	Not documented	Rolling limits with per-model multipliers, then Extra Usage	OpenCompute in private preview; air-gapped Enterprise	On-prem or air-gapped on Enterprise; Factory-hosted otherwise	Enterprises standardizing on Droid
Claude Tag	Claude only	Claude Team or Enterprise plan	Org usage balance with a spend limit	Self-hosted runners for sessions	Anthropic, including transcripts and by default inference	Claude Team and Enterprise orgs that live in Slack
Cursor cloud agents	Cursor’s agent	Paid Cursor plan	Billed at API pricing	Tool execution on your workers; agent loop and inference stay with Cursor	Cursor, including the agent loop and inference	Teams already in the Cursor editor
Grok Bot	xAI models only	SuperGrok or Cursor paid plans	Own usage pool; no published rates	SaaS only	xAI (SaaS only)	SuperGrok users who want always-on agents
Amp Orbs	Amp’s harness	Yes: ChatGPT, Claude, and others, or BYOK	No Amp token fees; Orbs billed per minute	Your own runners	Amp; no self-hosted option documented	Teams already using Amp
Warp Factories	Any harness per stage: Warp Agent, Claude Code, Codex	Not documented	Pay as you go at 20% above API rates; paid tiers at API rates	Managed self-hosting	Warp’s backend (orchestration, transcripts, inference)	Pipelines defined as code (closed beta)
OpenHands	OpenHands agent; third-party agents on paid plans	Not documented	Bring your own LLM key	Self-host the MIT-licensed agent; Helm on paid plans	Self-hostable on paid plans (Helm)	Teams that want open source

Facts as of October 2026, from each vendor’s documentation and pricing pages. “Not documented” means the vendor pages we read do not say.

Enterprise deployment: who runs the control plane?

Most of these products let you host execution, the machines agents run on, but keep the control plane — the part that orchestrates sessions, holds transcripts, and runs the agent loop — in the vendor’s cloud. Devin, Claude Tag, Cursor, and Warp document it that way. Two exceptions self-host: Factory Enterprise offers on-prem and air-gapped deployment, and OpenHands is self-hostable on paid plans. Nori Enterprise deploys the full stack — control plane, agents, and data — into your own AWS VPC, Azure VNet, or on-prem environment. On hosted plans, Nori gives each organization a dedicated, single-tenant control plane and data store, and every session runs on its own isolated machine.

Nori

Nori ships Jiro, its own AI engineering teammate, and also runs any harness you prefer: Claude Code, Codex, Gemini CLI, Cursor, OpenCode, pi, and ACP agents such as Droid, Goose, Amp, and Copilot CLI. It never charges for tokens — you bring your Claude or ChatGPT subscription, API keys, or an LLM gateway — so our incentives are aligned with yours and you are not locked to one model provider. Sessions start with strong defaults (a workflow skillset for plans, TDD, worktrees, and review, plus org skillsets and memories) and run from Slack, Discord, the web UI, the CLI, cron, and webhooks. Plans: free 30-day trial, Developer $99/mo, Team $250/mo, Enterprise custom, where Enterprise brings its own compute, IdP (SAML, SCIM, OIDC), LLM gateway, token provider, and memory provider, in your cloud or on-prem.

Meet Jiro

Jiro is Nori’s own AI engineering teammate. It runs on Claude Code or Codex under the hood, on your own Claude or ChatGPT subscription or API key, so there are no token charges. It follows an opinionated engineering workflow: planning, test-driven development, review, and worktrees. You work with it in Slack or the browser, and on its first day it reads your repos, asks the team how you ship, and sets up the workspace. See jiro.bot.

The alternatives

Devin

Cognition’s Devin is a mature autonomous agent with automations on schedules, webhooks, Slack, GitHub, Linear, Jira, and PagerDuty. It runs only Devin, does not support third-party LLM API keys, and bills token quotas then on-demand credits; Outposts can run session machines on your infrastructure. Read Nori vs Devin.

Factory

Factory’s Droids run on rolling usage limits with per-model multipliers, and its Enterprise tier offers isolated EU, air-gapped, and hybrid deployments. Automations start Droid sessions only, BYOK is limited to the CLI and desktop app, and bring-your-own compute is in private preview. Read Nori vs Factory.

Claude Tag

Anthropic’s Claude Tag puts Claude in your Slack channels with admin-governed access, in public beta for Team and Enterprise plans. It runs Claude only, bills channel work to an org usage balance, and works in Slack but not Discord. Read Nori vs Claude Tag.

Cursor cloud agents

Cursor’s cloud agents run in isolated VMs on paid plans and are billed at API pricing. Self-hosted machines keep tool execution on your workers, but the agent loop and inference stay with Cursor. Read Nori vs Cursor cloud agents.

Grok Bot

xAI’s Grok Bot is a beta team of always-on agents, each with its own cloud computer, included with SuperGrok tiers and Cursor paid plans. It runs xAI models only, is SaaS only, and has no published usage rates. Read Nori vs Grok Bot.

Amp Orbs

Amp Orbs give each thread a fresh cloud VM billed per minute, with no Amp token fees and support for your ChatGPT, Claude, and other subscriptions — the closest to Nori on billing. The difference is that Orbs run Amp’s own harness. Read Nori vs Amp Orbs.

Warp Factories

Warp Factories defines agent pipelines as version-controlled code and lets each stage use any model and harness, including Claude Code and Codex. It is in closed beta by application, and pay-as-you-go usage is billed at 20% above API rates. Read Nori vs Warp.

OpenHands

OpenHands is an MIT-licensed open-source agent you can run yourself, with a large community and paid team plans from $499 a month for 10 seats. Paid plans add self-hosting through Helm, your own LLM key, and third-party agents. Read Nori vs OpenHands.

What about OpenAI dots?

OpenAI dots were reportedly announced at DevDay on September 29, 2026 as always-on agents inside ChatGPT, each with its own cloud computer and browser (The New Stack). According to press coverage, they launched on ChatGPT Pro and Business Premium with one dot included, run OpenAI models only, are SaaS only, and can be triggered from ChatGPT, Slack, and Teams. Dots are a general-purpose ChatGPT agent rather than a coding runtime, so we have not written a head-to-head page.

What happened to Google Jules?

If you arrived from our old Jules page: Jules is Google’s asynchronous coding agent. It runs Gemini models only, works against GitHub repositories, and is priced as daily task quotas inside Google AI plans — 15 tasks a day on the base tier, 100 on Google AI Pro, and 300 on Google AI Ultra, according to its usage-limits page (Jules usage limits). Its changelog shows recent additions such as a CI fixer for failed GitHub Actions checks and newer Gemini models (Jules changelog).

How to choose a cloud coding agent

Which agent runs? Most products run only their own agent; Nori ships its own, Jiro, and also runs any harness. If your team already trusts Claude Code or Codex, or wants to switch as models improve, pick a product that runs any harness.
Whose subscription pays? Using the Claude or ChatGPT subscriptions you already pay for can be far cheaper than buying the same model again at API rates.
Who charges for tokens? A vendor that bills or marks up tokens earns more when your agents use more. Flat plans keep incentives aligned.
Whose infrastructure? Check what actually runs on your machines: the tools, the agent loop, the inference, and the transcripts.
Where does the team work? Slack, Discord, Teams, Linear, Jira, GitHub — and whether cron schedules and webhooks are built in.
Defaults or blank slate? Opinionated workflow defaults get you value on day one; a pipeline language gives you control if you have time to design one.
Available now? Several products are in beta or private preview. Check before you plan a rollout.

Frequently asked questions

What is a cloud coding agent?

A cloud coding agent is an AI coding agent that runs on a remote machine rather than your laptop, so it can work unattended, in parallel, and from triggers such as Slack messages, schedules, or webhooks, usually ending in a pull request for human review.

Which cloud coding agents let me use my own Claude or ChatGPT subscription?

As of October 2026, Nori and Amp Orbs document using your own Claude and ChatGPT subscriptions. Devin lets a ChatGPT Go, Plus, or Pro plan pay for GPT usage but does not support Claude subscriptions. The other products we compared do not document subscription sign-in.

Which cloud coding agents run any harness?

Nori runs its own agent, Jiro, plus Claude Code, Codex, Gemini CLI, Cursor, OpenCode, pi, and ACP agents. Warp Factories lets each pipeline stage use Warp's agent, Claude Code, or Codex, and paid OpenHands plans can use third-party agents. Devin, Factory, Claude Tag, Cursor, Grok Bot, and Amp run their own agents.

Which cloud coding agents do not charge for tokens?

Nori never charges for tokens; you bring your own subscription, keys, or gateway and pay a flat plan. Amp Orbs advertise no Amp token fees and bill Orbs per minute. OpenHands has you bring your own LLM key. Devin, Factory, Cursor, Claude Tag, Grok Bot, and Warp meter or bill model usage in some form.

What is the best Jules alternative?

Jules runs Gemini models only with daily task quotas inside Google AI plans. If you want to keep Gemini but also run other agents, Nori runs Gemini CLI alongside Claude Code, Codex, and others on your own subscriptions or keys, with Slack, Discord, cron, and webhook automations.

Sources

Competitor facts were read from the vendor pages below on October 19, 2026.