Security
Your agent uses your accounts, never your passwords.
Every promise on this page has a concrete platform mechanism behind it. If we do not have it, we do not promise it.
We support providers outside the EU, all listed in the subprocessor register, and on Enterprise, at the client's request, we limit your workspace to providers that process its data in the European Economic Area.
Our promises
What we never do, and what you can always stop.
Six commitments, each with the mechanism that keeps it inside the product.
Your passwords never reach the agent
The agent holds a decoy shaped like each password, key or card. The real value goes in only on the way to the site it belongs to.
✓ Decoys in the sandbox · swapped at the destination
Each credential stays on its sites
A Vault credential is used on the sites you set for it, and on any other only with an approval.
✓ Credentials bound to their sites
You decide what gets approved
Require approval for everything an agent writes, only for some tools or destinations, and keep your answer for the next times.
✓ Approval per action · saved decisions
You see what it does, as it does it
The agent's browser can be watched live, you can take over or stop the run, and every action is recorded with who approved it.
✓ Live browser · action record
Every workspace kept apart
The database's access rules and the server's checks separate every workspace, and every run executes on its own isolated machine.
✓ Row level security · one sandbox per run
Your data is yours
We train no models on it. It is kept for twelve months, or what you set on Enterprise, and deleting a workspace erases it from the backups too.
✓ Retention per workspace · erasure register
Under the promises
Everything behind them.
The concrete measures, grouped by where they act.
Identity and access
- One account per person, with a verified email. Sign in with Google or Microsoft, and SAML with SCIM on Enterprise.
- A TOTP second factor for anyone who turns it on, required across the workspace when its owner enforces it on Enterprise.
- Roles per workspace and permissions per resource.
- API keys shown once, stored as a hash, with optional expiry and narrowed scopes.
- Lockout after failed attempts, an email on any change of password, email or factors, and every sign in logged for twelve months.
Credentials and secrets
- Customer credentials are encrypted under a root key that only exists encrypted by AWS KMS.
- Passwords, keys and cards enter the agent's sandbox as decoys, swapped only on the way out to their destination or when filling the site's field.
- When an agent uses another person's credential, the browser switches to a protected mode: no JavaScript or debugging for the agent, and fields hidden in every capture.
- Server images are built without credentials, and the platform's secrets are delivered at startup.
Agents and execution
- One isolated machine per run, never shared.
- Process, file and network isolation on every command.
- With a network policy, internet traffic leaves through a proxy that reaches only the allowed destinations.
- Customer and contact files are analysed inside the sandbox, never on our servers.
- Addresses flagged by Google Safe Browsing are blocked, except in workspaces limited to European providers.
- Spend limits and loop detection on every run.
Infrastructure
- Database, API and web app hosted in Frankfurt.
- TLS 1.2 or higher on every external connection, and HSTS on the site, the app and the API.
- Encryption at rest for the database, storage and disks.
- The API is published through Cloudflare, with denial of service mitigation at the edge.
- No SSH from the internet: servers are administered through AWS Systems Manager.
- AWS account audit log kept for twelve months and threat detection that alerts the security team.
- Daily database backups in the same EU region.
Records and response
- Every configuration, security or permission change is recorded with its author and its previous state.
- Every tool an agent uses is recorded with its arguments, its result and the approval it received.
- Alerts on availability, authentication, the database, servers and sandboxes.
- A public status page at status.vibetasking.com, also shown in the app.
- We tell you of a breach that affects your data without undue delay, and always within 48 hours.
Development and dependencies
- Dependencies locked by hash, and the base image pinned by digest.
- Vulnerability scanning of dependencies every week and on every backend change, plus malicious package analysis.
- A software bill of materials (SBOM) on every API deploy.
- A deploy is blocked when the image still ships a vulnerability past the window its severity sets.
In writing
What you sign, to read.
We provide the service under the BÁSICO profile of its System Categorisation under the Spanish National Security Framework (ENS), with a declaration of conformity in the BÁSICA category, and under the MEDIO profile on Enterprise by agreement.
Responsible disclosure
Found a flaw?
Write to us with the steps to reproduce it, including Content Credentials we signed that do not conform. Do not test against other customers' data or the availability of the service. Report abuse here too: phishing, malware or spam on a site or preview under vibetasking.app, vibetasking.site or vibetasking.dev. We review every report and tell you what we did.