For teams building with agents

Let the agent build. Keep production yours.

Connect Claude Code, Codex, Cursor or any MCP client once. It gets computers to work in, sandboxes with a deadline, databases it can branch and apps it can deploy. Its key carries a class and an app list the platform enforces; the app list keeps a sandbox key away from production apps.

Join the waitlist For coding agents
Your job
Let coding agents do real work on real infrastructure, not only write code in a chat.
What gets in the way
A key in an agent's hands is a key to everything. A prompt that says "only read" is one sentence from forgetting it.
What Rytiv does
Gives the agent places to work and puts the limits in the key, where the platform checks them on every request.
What you get

Six things that do the job.

01

A computer per agent

A machine with a home disk at /data that outlives it, Node.js, git, the GitHub CLI, Python and a desktop. Watch it or take over with ry computer view and ry computer control.

3 sizes, up to 4 vCPU Docs: A computer per agent →
02

A sandbox per task

Started from the image your app runs, removed at its deadline. Outbound traffic: everything, nothing, or an allowlist of domains.

100 running at once Docs: A sandbox per task →
03

A branch per migration

The agent branches the database in seconds and tries the migration there. Without a branch, execute_sql runs in a read-only transaction.

Branch expiry 60 s to 30 days Docs: A branch per migration →
04

Keys with a class

sandbox, twin or production. A production key only reads. A list of apps narrows any key. No agent key can mint, reveal or revoke a credential.

Enforced on every request Docs: Keys with a class →
05

An audit before every change

Every refusal and every allowed request beyond a plain read is written to an audit before it runs. A change that cannot be recorded does not run.

GET /v1/agent-actions Docs: An audit before every change →
06

One server, three doors

The MCP server, the ry CLI and the REST API use the same keys and see the same apps. Hosted connectors sign in with OAuth, no key pasted.

OAuth 2.1 with PKCE Docs: One server, three doors →
A typical setup

Connect once. Then ask in plain words.

A key of its own for the agent, the MCP server added to Claude Code, and a first task.

  1. ry keys create claude-code --agent sandbox --app web-try

    A key that reaches web-try and nothing else, shown once

  2. claude mcp add --transport http rytiv https://api.rytiv.com/v1/mcp --header "Authorization: Bearer <your key>"

    Any MCP client over Streamable HTTP works the same way

  3. "Create a computer on Rytiv, clone the repository
 into it and run the tests."

    The agent uses computer_create and computer_exec

  4. curl https://api.rytiv.com/v1/agent-actions?limit=100 \
  -H "Authorization: Bearer $RYTIV_ADMIN_TOKEN"

    Read what the agent did. A key narrowed with scopes cannot read it

Why the key, not the prompt

The platform holds the line, so you do not have to watch.

Credentials are out of reach

Creating, revealing or revoking keys, SSH keys, webhooks and Git installations is refused for every agent class.

Production only reads

Status, releases and logs are open to a production key. Files and query results need an explicit flag. A shell and any change are refused.

Copies, not originals

Branches for databases, sandboxes for code, previews for apps: the agent works on copies you can throw away.

Building blocks

Everything on one API.

AppsFrom GitHub, a folder or an image. Rolling deploys, previews, rollbacks. Static sites without a machine.Serverless PostgresBranch in seconds, fork or restore from any moment in its history. Private, TLS only.ComputersA machine per coding agent, a home disk that stays, a desktop you can watch or take over.SandboxesMachines with a deadline from your app's image. Egress allowlist. 100 at a time.StorageS3 buckets with presigned URLs and CORS. Volumes that outlive machines, resizable.MailTransactional mail from your domains, inbound, a delivery log. No advertising or newsletters. Private beta.Cron and jobsCall a path on a schedule or run a command on its own machine. Exit code and output kept.Domains and TLSSet the records the answer names; certificates issue and renew themselves. Included.Network and protectionPrivate app links, password or share-link previews, signed access, egress caps.
Also for Freelancers, startups and product teams →Enterprises and agencies in Europe →

Give your agent a cloud.

One line to connect.

Join the waitlist Quickstart 
claude mcp add --transport http rytiv https://api.rytiv.com/v1/mcp --header "Authorization: Bearer <your key>"

For businesses and freelancers, from a solo developer to a team. Pricing