> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.cloudraker.com/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.cloudraker.com/_mcp/server. # Quickstart This walkthrough takes you from a blank screen to your first approved records in one sitting. You'll create a project, start a playbook, do a task, let Florent draft some records, and approve them — the full RakerOne loop in miniature. Each step here gives you just enough to act, then links to the page that covers the detail. If you only read one thing first, read this rule: **AI drafts, humans approve.** Florent and automated steps never commit data on their own. They produce drafts — proposed records, pre-filled forms, generated documents — and you review, edit, approve, or reject before anything becomes real data. You stay in control at every gate. ## What you'll need * An account in a RakerOne **organization** (your sign-in is handled by your company's identity provider). * Permission to create a project, or an existing project someone has shared with you. * A **Published** playbook to run, or a few documents and a sentence describing what to do with them (Florent can shape the rest). New to the terms? Skim [Concepts](/get-started/concepts) for the glossary, or [Introduction](/get-started/introduction) for the big picture and a tour of the sidebar. This page assumes you've signed in and can see the left sidebar. ## The happy path A **project** is the workspace for one piece of work — it holds the files, records, runs, and people for that job. In the sidebar, find the **Projects** group and click the **New project** button (the **+** next to the group label). Choose **Blank project** to start from scratch, give it a **name** and an **icon**, then click **Create project**. You land on the new project's page. If your team has set up **Project templates**, pick one instead of Blank project — it pre-loads the playbooks and record types you'll use, so you can skip straight to running. See [Create a project](/projects/create-a-project). A **playbook** is a reusable process — the steps, their order, who's assigned, and what needs approval. A **run** is one live execution of that playbook inside your project. Open the project's **Playbooks** tab. You have two ways to begin: In the **Playbooks** table, find one with the **Published** status and click its **Run** button. The run starts and you drop straight into the run workspace. Only **Published** playbooks show a **Run** button. If a playbook you want is still a **Draft**, it can't be run yet — someone with builder access has to publish it first. See [Playbooks overview](/playbooks/overview). Scroll to the **Ask Florent** card, describe the work in plain English (for example, "Import these invoice PDFs and extract them into the Invoice record type, then send to finance for approval"), and click **Create plan**. Florent shapes your description into a reviewable playbook plan. Florent never auto-runs anything — you review the plan and approve it before any work starts. Read more in [Run a playbook](/projects/run-a-playbook). The **run workspace** is where the work happens. It has three columns: * **Left** — the **task rail**: every step in the run, grouped **Open** (needs attention now), **Blocked** (shown as **Waiting** until earlier steps finish), and **Done**. * **Middle** — the **Timeline**: comments, Florent's messages, and run events in one stream. * **Right** — the task you've opened, or the **Run summary** once the run is finished. Tasks light up automatically as they become available. Human tasks (**Form**, **File upload**) wait for a person; the rest (**Document**, **Action**, **AI task**) run on their own. For the full layout, see [Run a playbook](/projects/run-a-playbook). Click the first task in the **Open** group of the rail. If it's a human task, you claim it first, then do the work. * **Form** task — click **Claim task**, fill in the **Fields**, then **Submit task**. * **File upload** task — click **Claim task**, **Upload** the requested files, then **Complete task**. Claiming makes the task yours so two people don't work the same step. Required fields and uploads are checked before you can submit. Step-by-step detail lives in [Completing tasks](/work/completing-tasks). If a task is marked **Waiting**, it's just not its turn yet — an earlier task has to finish first. It flips to **Available** on its own. For automated steps — a **Document** task reading files into records, an **Action** task, or an **AI task** — you don't start anything. Florent picks them up automatically when they become available. You'll see progress messages like **Reading document** while Florent works. You can leave the page; the run keeps going. When Florent has results, the task moves to **Pending approval** and waits for you. Florent drafts; it never commits. What each task type does is covered in [Task types](/playbooks/task-types). This is the gate that turns drafts into real data. Open the task that's **Pending approval** and you'll see the drafts Florent produced. For extracted records, each value carries a **citation** — a pointer back to the exact spot in the source document it came from. Click a value to open the source and confirm it matches before you approve. Then: * **Approve** the values you trust (the button reads **Approve values** or **Approve documents** depending on what's being approved). * **Reject** anything wrong — rejecting requires a reason, so type why you're sending it back, then **Confirm rejection**. Approving the task writes the records into your project. The full review experience — checking citations, locking values, and correcting fields — is in [Reviewing AI work](/work/reviewing-ai-work). Rejecting a task requires a reason. Once you reject, the doer (or Florent) is notified and the task comes back for another try — be specific so they know what to fix. When the run finishes, the right column shows the **Run summary** — what got done and how many **records** were created. To work with that data, open the project's **Data** tab. Your approved records are there as rows you can filter, sort, edit, and reuse. That's the payoff: a pile of documents turned into clean, structured data — with a human signature on every value. See [Project data](/projects/data). ## What just happened You ran the whole RakerOne loop: a **project** held the work, a **playbook run** broke it into **tasks**, you did the human steps, Florent drafted the rest, and your approval turned those drafts into committed **records**. Every consequential action — starting, submitting, approving — was yours. ## Where to go next See everything waiting on you across all your projects, grouped by urgency, on the **My work** page. Ask Florent to look things up or draft projects, playbooks, and record types — in plain language. Go deeper on starting runs, the run workspace, and the Pause / Resume / Cancel controls. Ready to design your own process? Assemble tasks, set approvals, and publish in the no-code editor.