Product
Home Core Platform AI Intelligence Documentation Blog & Insights Pricing
Company
Security System Status Contact
Legal
Privacy Policy Terms of Service Refund & Cancellation
Trust

Security

This page describes the security controls that are implemented in the Flowprise codebase and deployed service today. We list what is built, not what we would like to build, and we do not use absolute security claims.

Revised: November 2026 Covers: flowprise.tech

No system is perfectly secure. Flowprise makes no absolute, unqualified, or industry-superlative security guarantees, and no software should be treated as incapable of being compromised. Software defects and attacks are always possible. If you believe you have found a vulnerability, please report it to support@flowprise.tech rather than exploiting it.

1. Authentication

Flowprise implements the following authentication controls:

  • Passwords are stored as salted one-way hashes using Werkzeug's password hashing helpers. Plaintext passwords are never written to the database or logged.
  • Sessions use signed tokens (HMAC-SHA256) rather than a client-trusted session identifier. Token signatures are verified with a constant-time comparison to resist timing attacks.
  • Token identifiers include a high-entropy value generated from a cryptographically secure random source.
  • Email verification is required before workspace access is granted.
  • Password reset and one-time login codes are stored only as hashes, are time-limited, and are consumed on use.
  • Every issued token has a revocation record, so a session or credential can be invalidated before its natural expiry rather than only by waiting for timeout.
  • Expired revocation artifacts are cleaned up on a recurring basis instead of accumulating.

2. Authorization and role-based access control

Authorization is enforced on the server. Every protected endpoint validates the session before touching data, and role permissions are checked in backend code rather than relied upon from the interface.

  • A central permission map assigns granular actions — view, create, edit, delete, export, manage, approve — to each role.
  • Reusable guards verify authentication, allowed roles, and specific required permissions.
  • Account status is re-checked on request, so a suspended or deactivated user loses access immediately rather than at token expiry.
  • Current membership is resolved server-side on every request, so removing a user from an organization takes effect without waiting for token expiry.
  • AI endpoints apply the same permission model as the rest of the platform. An AI request cannot retrieve or act on data the requesting user is not permitted to see or change.

3. Multi-tenant isolation

Flowprise is built as a partitioned multi-tenant application, and tenant separation is enforced in the data layer rather than by convention.

  • Business records carry organization and workspace keys, and queries are partitioned on those keys.
  • Tenant identifiers are derived from the verified server-side session. They are never accepted from a request body or query parameter, so a caller cannot ask for another organization's data by editing a URL.
  • Persisted datasets, generated reports, and integration credentials are namespaced to the owning organization and workspace.
  • AI context, conversational history, and embeddings are bounded by organization.
  • Database-level foreign key constraints prevent records from referencing a customer or SKU that does not exist.

4. Data in transit and security headers

  • All client and API traffic is served over HTTPS, and the deployment sets HTTP Strict Transport Security to enforce encrypted transport.
  • A Content Security Policy restricts which script, image, style, and connection sources the browser will load.
  • X-Content-Type-Options prevents MIME-type sniffing.
  • X-Frame-Options and frame-ancestor restrictions prevent the application being framed by third-party sites.
  • X-XSS-Protection is set for legacy browser XSS filtering.
  • Referrer-Policy restricts how much referrer information is sent off-site.
  • HTML documents are served with no-store caching so authenticated pages are not retained by shared caches.

5. Secrets management

  • Secrets are read from environment variables and are never hard-coded in the source tree or shipped in browser bundles.
  • Where a signing key is not configured, the application refuses to use a shared fallback and generates an unpredictable per-process value instead of a fixed default.
  • API keys and webhook secrets are stored as hashes so they are not readable back out of the database.
  • Credentials for connected integrations are held server-side and are not exposed to the browser.

6. Input validation and injection defence

  • Request payloads are validated before use, with explicit type, length, and required-field checks on sensitive endpoints.
  • Database access is parameterised rather than built from concatenated input.
  • Uploaded file names are sanitised before being used as filenames or written to disk, preventing path traversal.
  • Natural-language analysis endpoints screen input for code-injection indicators and reject it.
  • Analysis operations are executed from a fixed allow-list of aggregations, chart types, outlier methods, models, and cleaning operations. Arbitrary code is never evaluated on user input.
  • Dataset size is bounded on multiple axes — file size, row count, column count, and total cell count — to prevent resource exhaustion from oversized or malformed input.

7. Upload validation

  • Uploads are restricted to an allow-list of spreadsheet formats: CSV, TSV, XLSX, and XLS.
  • Uploads are capped at 50 MB per request at the framework level, before the body is read into memory.
  • File content is checked against the declared extension, and parsing failures are rejected rather than partially accepted.
  • File type and size are re-validated server-side at the point of use, not only at the form.
  • Uploads are stored within the requesting organization's dataset namespace, so a dataset identifier cannot be used to reach another tenant's file.

8. Spreadsheet formula-injection protection

Exported data is a genuine risk: a cell value beginning with =, +, -, or @ is executed as a formula when the file is opened in a spreadsheet application. Flowprise neutralises these leading characters when serialising cells for export, so exported files cannot execute attacker-controlled formulas when opened.

This protection is applied on the export path for cleaned datasets and report output.

9. Rate limiting and abuse prevention

  • A rate-limiting gate applies per user and per client IP to authentication routes, file uploads, and AI query endpoints.
  • Limits scale with the platform's configured rate-limiting policy, so an operator can tighten or relax thresholds centrally.
  • Exceeding a limit returns HTTP 429 rather than processing the request.

10. Concurrency and data integrity

  • Multi-record operations that must succeed together execute inside a single database transaction, so a partial failure rolls back rather than leaving half-written state.
  • Inventory mutations are protected against lost updates under concurrent requests.
  • Foreign keys, unique constraints, and check constraints are enforced by the database, not only by application code.
  • State transitions follow defined paths, and monetary totals are derived from persisted line items rather than trusted from client input.

11. Database and schema management

  • The relational schema is managed through version-controlled migration files, so schema changes are reviewable and repeatable rather than applied as ad-hoc DDL.
  • Production PostgreSQL is the system of record, with foreign keys, uniqueness constraints, and indexing on tenant and lookup keys.

12. Audit logging

Flowprise maintains an append-only audit log covering the events that matter after an incident:

  • Authentication events, including sign-in, sign-out, and failed attempts.
  • User, membership, and role changes, including invitations and access revocations.
  • Data mutations such as bulk deletion, dataset export, and manual stock overrides.
  • Financial state changes such as invoice status transitions and payment recording.
  • Configuration changes such as API key generation, webhook creation, and automation toggles.
  • Automation executions and AI-initiated activity.

Each entry records the acting user, the organization, the action, the target resource, a UTC timestamp, the client IP, and the outcome. Secrets are not written into audit parameters.

13. Security regression testing

Security-relevant behaviour is covered by automated tests in the repository, including suites covering the core security model, authentication freshness, token revocation, migration safety, and inventory concurrency. These run as part of the project's test suite so that a regression in authorization or isolation is caught before release.

14. Reporting a vulnerability

If you believe you have found a security issue, please email support@flowprise.tech with a description, the affected area, and steps to reproduce if you are comfortable sharing them. Please avoid accessing other customers' data, and do not run denial-of-service testing against the live service.

We aim to acknowledge good-faith reports promptly. Flowprise does not currently operate a public bug-bounty programme or publish a formal safe-harbour policy; please do not treat this section as one.

15. Platform status

Current live service state is published on the System Status page. Flowprise does not publish an uptime percentage, because we do not operate a third-party monitoring service whose historical data we can substantiate.

Flowprise Security · Revised November 2026 · Report an issue: support@flowprise.tech