The learning path: Advanced
The security model, RPA, MCP, guardrails, budgets and releases.
- 1
The security model: classification, visibility and requirements
Flow does not secure per screen but per item: a table knows how sensitive its data is, a workflow says what a caller must bring, and an endpoint may only make security stricter. In this guide you create a table with personal data, a workflow that reads it, and you see an endpoint with too few rights being refused — and how to put it right.
- 2
Operating a website with browser steps (RPA)
Not every system has an API. A supplier portal where you look up invoices every week, Flow can operate like a person too: fill in, click, read a table. You build a browser session on a demo portal (if operating websites is still off, you switch it on at the step itself) and look at the screenshots Flow keeps with the run.
- 3
Building from Claude Desktop and Claude Code (MCP)
With MCP access you let an AI app build in your development environment: modules, tables, workflows, agents. The app acts as you and can never do more than your role, it cannot delete, and keys and security policies remain in human hands. You create a key with limited rights, see what the app may and may not do, and revoke the access again.
- 4
Guard rails and budget for an agent
An agent in Flow always runs in a box: a budget for tokens, tool calls and time, and guard rails that run before the model and cannot be got around with a clever prompt. You create an agent, set its budget and guard rails, and watch an attempt at prompt injection fail before the model sees anything.
- 5
Export a module as a package and import it elsewhere
A module — tables, workflows, APIs, agents — travels as a package with a version from one Flow to another: to your personal Flow to try things out, or to another organisation. The package is signed by your environment, so nobody can change it on the way unnoticed. Secrets, connections and table rows never travel. Whoever imports it first sees what is new, what is updated and what is still missing; a next version updates without overwriting their own changes.
- 6
Start a workflow at fixed times (schedule)
Not everything starts with a call. A start of the day at eight, a check every quarter of an hour, an overview every Monday: for that you replace the trigger with a Schedule. You write the moment as cron, choose the time zone and test the workflow by hand.
- 7
Encryption: your environment's key and renewing it
Passwords, tokens and keys you store in Flow — of connections, API policies, signing and dashboards — are kept encrypted in the database. For that every environment has a key of its own, made and kept by the portal and never shown to anyone. In this guide you see where that key is, renew it, and see how the environment re-encrypts everything without anyone having to set anything again.
- 8
Your own key vault (Azure, AWS or Google Cloud)
Passwords and keys Flow uses are kept encrypted in Flow's own vault by default. If you want to keep them in your own cloud, in Azure Key Vault, AWS Secrets Manager or Google Cloud Secret Manager, you choose that per environment in the portal. Workflows refer to a secret by its name; which vault looks that name up is up to the environment. So Test reads Test's vault and Production reads Production's, with the same workflow. Another vault is an access change: an owner approves it, and Flow never copies values — you put them in the new vault yourself.