Fair questions, straight answers

01 / getting started

What is Rollout?

A workspace where your team's work lives (projects, tasks, documents and custom records), workflows that run on that work with AI steps, tools and human approvals, and apps your AI assistant builds on the same data. Every run keeps its steps, inputs, outputs and approvals. The examples show what that looks like on real jobs.

What should I put on it first?

One recurring job that already has a queue and a decision in it: a daily check of your cluster, a release that needs a sign-off, a review that collects evidence before someone approves it. Start there, keep your other tools, and move more over once the first job runs on its own.

Who is it for?

Technical teams that run recurring work across software tools, AI and people: platform and DevOps engineers, engineering leads at small software teams and agencies, and anyone building agents that need somewhere accountable to run.

Can I get in today?

Rollout is in alpha. Cloud Pro workspaces open from the waitlist; Business is sold after a scoping call with the founder. The pricing page has the current entry path for each.

Do I have to write code?

No. Workflows are built on a visual canvas, and an AI assistant connected over MCP can build and edit them for you. When you want code there is a sandboxed JavaScript step (Deno) and shell steps on the runner of your choice.

02 / your existing tools

How is this different from n8n or Zapier?

Those are built to move events between apps, and they do that well. Rollout is built around the work item: the task or record a run starts from, the approvals it waits on, and the history it leaves. Workflows publish as immutable releases, a run records which release it executed, and a step can run on our cloud, your cluster, an SSH host or a registered desktop.

How is this different from LangGraph or another agent framework?

A framework gives you the model and tool loop as a library, and you build persistence, approvals, tenancy, audit and a UI around it. Rollout ships those parts. AI steps call only the tools you attach, a tool can require approval, and every call is recorded with the run. External agents such as Claude Code and Codex run as steps on a registered device instead of being rebuilt inside.

Do I have to move off Jira or GitHub?

No. Workflows read and write Jira, GitHub, Gmail, Google Drive and Sheets, Telegram and any HTTP API, so a first process can sit next to the tools you keep. The built-in tracker is there for work you want agents to pick up with its context attached.

Can Claude Code or other assistants connect?

Yes. Rollout's MCP server gives Claude, Cursor and any MCP client the same workspace over OAuth 2.1, with read, write and admin scopes. An assistant can query records, edit and release workflows, and build apps. Setup is in the MCP docs .

03 / where it runs

Where do steps actually run?

Each step picks its place. On Cloud Pro, code and shell steps run in ephemeral gVisor-sandboxed pods we operate. On Business they run on your Kubernetes or Docker. On either plan a step can run on an SSH host you connect or on a desktop running the Rollout client, where it can use a local shell, browser, files and Claude Code or Codex on your own subscription.

Which AI models can I use?

On Cloud Pro, a managed catalog billed at provider cost plus 10%, or your own provider keys at no Rollout charge. On Business, your own keys. The model is set per step, so two steps in one workflow can use two different models.

What data leaves the platform?

What your workflows send: prompts to the model provider a step uses, requests to the APIs and connections you configure, email and messages you send, and commands to the devices you register. Cloud steps reach the internet through policy-controlled egress. On Business the control plane and runners are yours, so those are the only paths out.

04 / security

How are credentials handled?

Secrets are envelope-encrypted, cannot be displayed after saving, and can be replaced by an authorized manager. Workflows reference them by name. Credentials the engine injects are scrubbed from step output before it is stored or streamed, and if redaction is unavailable, text output is suppressed.

What does the run history record?

Each step's inputs, outputs, streamed events, retries, duration and model usage, plus who responded to an approval and when. Record changes carry an actor and a timestamp. What an approval gates is up to the workflow: a rejection is a value the workflow branches on, so design the branch you want. Details are in the security overview .

05 / pricing and licensing

What does Cloud Pro cost?

A per-workspace subscription with unlimited members and an included monthly usage credit. Runner time, platform operations, storage beyond the allowance and managed AI are metered against prepaid credit with a hard cap. Every rate and a worked month are on the pricing page .

Can I run Rollout on my own infrastructure?

Yes. Business is the annual licensed self-hosted product for Kubernetes or Docker. Design Partner implementation is optional if you want the founder to build the first process with your team.

Is the Rollout repository public?

No. The repository and self-hosted distribution are private. Business customers receive deployment materials and a commercial license under their agreement.