Run Claude Code in the cloud for your whole team

Quick answer

Nori runs native Claude Code on isolated cloud machines that your whole team shares. Start a session from Slack, the browser or the terminal. Each session gets its own machine with your repositories, your organization's CLAUDE.md, skills, subagents and shared memories already in place, and it runs on the Claude subscription or API key you already pay for. Sessions keep working when laptops are closed and return pull requests for review.

Why move Claude Code off the laptop

Claude Code on a laptop is a strong tool for one engineer at a keyboard. It becomes harder to live with as a team standard, for four reasons.

• It stops when the laptop sleeps. A long refactor or an overnight test run needs a machine that stays on.
• It runs as you. A local agent can reach everything your user account can reach: every repository, every credential file, every cloud profile.
• Every setup is different. Ten engineers have ten CLAUDE.md files, ten sets of skills and ten ideas of how a pull request should look.
• Only engineers can start it. A product manager or support lead cannot ask for a small fix without finding someone with a terminal.

Running the same Claude Code on a shared cloud platform fixes all four without changing the agent your team already trusts. For the longer comparison, see Nori vs Claude Code.

How it works

Nori does not replace Claude Code with a look-alike. Each session runs the real Claude Code program on a cloud machine, with the same tools, hooks, skills, subagents and MCP servers it has locally.

1. Connect once. An admin connects GitHub, picks the repositories every session starts with, writes a setup script for the toolchain, and connects a Claude subscription or API key.
2. Start from anywhere. Mention @nori in Slack, open a new chat in the browser, or hand off a running local session from the terminal with the Nori CLI.
3. Get an isolated machine. Nori claims a fresh machine for the session, with the repositories cloned, the setup script run, and the organization skillset and memories installed.
4. Work as usual. Claude Code plans, edits, runs tests and uses your MCP servers. You watch and steer from the Slack thread or the browser, and you can open a shell on the machine when you want to look yourself.
5. Ship and release. The session pushes a branch and opens a pull request. When the work is done, the machine is released. The conversation, the branch and anything saved to shared memory remain.

Three ways in, one session history

Every session, whatever started it, appears in the same dashboard with its transcript. A teammate can open a chat that began in Slack and continue it in the browser.

Your subscription, not a second token bill

Nori runs Claude Code on the Claude subscription your team already has. You can also use Anthropic API keys or route requests through your own LLM gateway. Nori never charges for tokens and adds no markup on model usage.

You pay Nori a flat plan price for machines: a 30-day free trial with no card, Developer at $99 per month for one person and three machines, Team from $250 per month for five machines and unlimited members, and Enterprise with custom terms. For background on how other pricing models work, see do cloud coding agents mark up token costs.

Shared skillsets and memories

The largest difference between ten laptops and one platform is shared context. On Nori, the things Claude Code reads at startup are managed for the whole organization.

• Org skillset. One CLAUDE.md, one set of skills and one set of subagents, installed on every machine. Change the test-driven development skill once and every session uses the new version.
• Shared memories. When someone tells the agent "we never deploy on Fridays" or "the staging database is refreshed at 02:00 UTC", it is saved for every later session, not for one laptop.
• Custom agents. A team can pair its own skillset with a model and integrations. The data team's Claude Code knows dbt; the platform team's knows Terraform.
• Integrations with scope. Connect an MCP server for the whole organization, for one person or for one agent. A session receives only the credentials its requester and agent are allowed to use.

Repository-level files still work. A CLAUDE.md or .claude/skills/ directory checked into a repository is read as it is locally, and it layers on top of the organization defaults.

A worked example: one afternoon, three sessions

Here is how a six-person team uses the same Claude Code setup from three entry points in one afternoon. The repository names and details are illustrative.

1. 13:05, Slack. A support lead writes in #bugs: @nori password reset emails link to the old domain. Find where and fix it. A machine starts with web-app and notifications cloned. Claude Code finds the hard-coded host in notifications/templates/reset.html, replaces it with the APP_BASE_URL setting, adds a test, and replies with a pull request link at 13:19.
2. 13:30, terminal. An engineer is halfway through a refactor of the orders module in a local Claude Code session and has a meeting at 14:00. She hands the session off to the cloud from the Nori CLI. Her workspace and transcript move to a cloud machine, and she gets a browser link. Claude Code keeps running the test suite while she is away.
3. 14:45, browser. She opens the link from her phone after the meeting, sees two failing tests, and types: The fixture in tests/orders/conftest.py still uses the old status enum. Update it. The session finishes and opens a pull request.
4. 15:10, browser. A second engineer starts a new chat: Review the open PR on web-app that changes the reset email. Check it against our review skill. A separate machine starts, so the review does not share state with the session that wrote the code. It leaves comments on the pull request.
5. 15:20, Slack. The first session, still attached to the #bugs thread, picks up the review comments, pushes a follow-up commit, and replies in the thread.

All three sessions ran the same Claude Code with the same organization skillset, on the same subscription. None of them depended on a laptop staying awake, and each had its own machine and branch.

# The shared CLAUDE.md every session above started with (excerpt)

## Rules
- Never push to main. Open a pull request from a feature branch.
- Write a failing test before the fix. Run `make test` before pushing.
- Branch names: <type>/<short-description>, for example fix/reset-email-host.
- Pull request descriptions list what changed and how it was tested.

## Repositories
- web-app: Next.js front end. Run `pnpm test`.
- notifications: Python email service. Run `make test`.
- orders: Python order service. Fixtures live in tests/orders/conftest.py.

Isolation and enterprise deployment

Each session runs in its own sandbox. It can install packages, start servers and run tests without touching another session or anyone's laptop, and it holds only the credentials it was given. Admins can see every session the organization has run.

For teams that need everything inside their own boundary, 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. Hosted customers get a dedicated, single-tenant control plane and data store per organization.

Claude Code is not the only harness Nori runs. The same machines run Jiro, Nori's own agent, as well as Codex, Gemini CLI, Cursor, OpenCode, Droid, Amp and any ACP agent, so a team can standardize on Claude Code today without being locked to it. Sessions can also start without a person: see scheduled coding agents.

Frequently asked questions