trust / security
Security
Most software asks you to trust a company with your data. work.flow is built so that, for the most part, you don't have to: the architecture keeps your data where it already is — on your machine. This page describes that model honestly, including what it does not cover, and explains how to report a vulnerability to us.
1. The local-first model
- Your files never leave your computer. Workflows execute locally. There is no upload step, no sync service, no work.flow cloud. A workflow that renames 3,000 photos touches those photos exactly where they live.
- Your AI keys are yours. When a workflow calls an AI model, the application talks to the provider you configured — with your API key, directly from your machine. Keys are stored locally on your computer, are sent only to the provider they belong to, and never transit our infrastructure. Point the app at a local model and nothing leaves your machine at all.
- No account, no server-side profile. The beta requires no login. There is no work.flow server holding your workflows, your prompts or your history.
- Readable code. Every built-in step is real Python you can open and read inside the application, and everything the agent generates is shown to you as inspectable code before it becomes part of your workflow. The system is designed so that "what will this do?" is always answerable by reading, not trusting.
- Preview before execution. Before a workflow runs, you get a plain-English preview of what will happen. Destructive operations are explicit, not incidental.
2. Honest boundaries
Security claims fail when they overreach, so here are ours precisely. The local-first model means most cloud-provider risks — server breaches of your content, insider access, subpoenaed hosted data — simply do not apply, because the data is not with us. It does not make your computer itself secure: the application runs with the permissions you give it, on your machine, alongside your other software, and workflows you build can do the things you tell them to do (that is the product). Third-party AI providers you connect process whatever your workflow sends them under their own terms. We currently hold no formal certifications such as SOC 2 or ISO 27001; for a product with no cloud component the audit surface is small, and this page — plus code you can read — is our substitute until a formal audit is warranted.
3. Website security
This website is a static site: no accounts, no database of yours to breach, no third-party scripts, no cookies (verifiably). It is served over TLS. The only personal data connected to it — waitlist emails — is described in the Privacy Policy and is deliberately minimal: an email address is the entire prize an attacker could win, and we would still treat that breach with full seriousness, including notification duties.
4. Responsible disclosure
If you believe you have found a vulnerability in the work.flow application or this website, we want to hear from you, and we will treat you as a colleague, not a threat.
How to report
- Email [SECURITY-EMAIL] with a description of the issue, steps to reproduce, affected version or URL, and — if you have one — a proof of concept. Plain text is fine; so is imperfect English.
- We acknowledge reports within 72 hours and aim to give you a substantive assessment within 7 days.
- We ask for up to 90 days of coordinated disclosure before you publish, less if we fix it sooner — and we will credit you by name or handle if you wish.
Safe harbour
Good-faith research conducted within this policy — without harming users, without accessing data that isn't yours, without degrading the service — will not be met with legal action from us. If in doubt whether something is in scope, ask first at the same address.
Out of scope
- Denial-of-service and volumetric attacks, spam, and social engineering of people connected with the project.
- Reports that a third-party AI provider you configured has the data your workflow sent it — that is how the product works, under your agreement with that provider.
- Automated scanner output without a demonstrated impact.
There is no paid bounty program during the beta; we say so here rather than letting you find out after the report.
5. A note for businesses
Because the application processes your data entirely on your machines and we never receive it, work.flow the company is ordinarily neither a controller nor a processor of the content your teams run through it — which means there is usually no data-processing agreement (DPA) needed with us, no sub-processor list to watch, and no cross-border transfer of your operational data to analyse. Your data protection assessment mostly collapses to: the software runs where your files already are. Security questionnaires can be sent to [CONTACT-EMAIL].
6. Contact
Security reports: [SECURITY-EMAIL] · everything else: [CONTACT-EMAIL]