Security at TINIUN
Workforce data describes people's working lives. Here is how TINIUN keeps it separated, access-controlled and accountable.
Every record belongs to one organization. Postgres row-level security enforces that boundary on every query the application makes, so a bug in a single screen can't expose another customer's data. The application connects with a restricted database role that can't switch these policies off; cross-organization access happens only on a few explicit system paths, such as sign-in and background jobs.
Passwords are stored only as salted scrypt hashes. Sessions are server-side, stored as hashes, carried in HTTP-only secure cookies, and can be revoked at any time. Sign-in attempts are rate-limited, and state-changing requests require an anti-CSRF header.
Organizations can sign in through their identity provider using SAML 2.0 or OpenID Connect (authorization code flow with PKCE), require authenticator-app multi-factor authentication, and provision people automatically with SCIM.
Administrators can limit sign-in and time-in to approved networks and enrolled computers. Enrolled computers hold a random token in an HTTP-only cookie (only its hash is stored), and every decision, including time-limited exceptions, is logged.
Applications authenticate with OAuth 2.1 (authorization code with PKCE required, single-use codes, rotating refresh tokens that revoke the grant if reused) or with scoped, expiring API keys. Tokens and keys are stored only as hashes, can be revoked by administrators at any time, and per-organization rate limits apply. OAuth and the MCP server act as the person who approved them and never exceed their own permissions.
Outgoing webhooks are signed with HMAC-SHA256 and a timestamp, retried with backoff, and logged. Destinations must be HTTPS and are checked each time TINIUN connects, so webhooks can't be pointed at private or internal addresses. Inbound integration events are verified against each sender's signature or shared secret before anything is recorded.
The assistant and the Workforce advisor chat are available only where an organization turns them on, read data with the asker's own scope, can't change anything, have no tools, and don't store the conversation. Every message is screened for prompt injection and attempts to reach secrets or other organizations before any AI is called, the topic is limited to workforce analysis, and replies are screened for secrets. Questions and the data needed to answer them go to Anthropic to generate the answer, as described in the Privacy Policy. Connected third-party AI tools see only what each person approved.
Every app lists the exact scopes and data it uses before you buy and again before you connect, and cannot use anything it did not declare. Nothing is active until an administrator has both bought (or tried) and connected it, and both steps are audited. An app's settings, secrets, routes, linked people, activity log and subscription belong to one organization only; cancelling or suspending one organization never affects another, which is covered by isolation tests. Entitlement is checked every time an app acts.
Third-party tokens, signing secrets and webhook addresses are encrypted by the application before they are stored, are write-only, and are never returned by the API or shown again. The activity log scrubs credential-like text before saving a message.
Inbound requests from Slack, Microsoft Teams, Zendesk, Intercom, HubSpot, phone systems and TINIUN webhooks are verified against the sender's signature on the exact bytes received, must be recent (five minutes) and cannot be replayed; the organization is taken from the verified request. Outbound calls use an SSRF-guarded client: HTTPS only, private and internal addresses refused after name resolution, no redirects, and size and time limits. OAuth state is single-use, tied to the user and organization, and expires in 10 minutes, with PKCE where the service supports it.
Purchases, trials, cancellations, connects, disconnects, credential changes, people links and every action taken from Slack or Teams are written to the audit log with who did them. MCP tool calls are logged with the tool and credential, and only the names of the arguments, never their values. Blocked attempts against the Workforce advisor chat are audited for administrators.
Roles are scoped to the organization, a department, a team or the individual. A team leader sees their team; an agent sees their own data. Every API request is checked against those permissions on the server.
Activity events, attendance corrections, schedule versions and audit logs are append-only: the database rejects edits and deletions. Administrative and sensitive actions are recorded with who, what, when and where.
All traffic to TINIUN is served over HTTPS, and the application's connections to its database are encrypted with TLS. Stored data is encrypted at rest by our database provider. Secrets we must be able to reuse, such as webhook signing secrets and integration credentials, are additionally encrypted by the application before they are stored.
Chat is for colleagues, not customers' data. Messages are checked before they're stored: card, bank account and government ID numbers are blocked, and customers' email addresses and phone numbers are removed. Chat messages are deleted automatically after the period each organization sets — 24 hours by default.
TINIUN runs on Vercel, with application data in a managed Supabase Postgres database in Singapore. Security headers are set on every response, and production secrets are kept out of source code.
Agent actions are idempotent and check the expected current state, so duplicate taps, retries and stale screens can't corrupt a timeline or create duplicate exceptions.
If you believe you've found a security issue, email [email protected] with the details and steps to reproduce. Please give us reasonable time to fix it before sharing it publicly, and don't access data that isn't yours or degrade the service while testing. We'll acknowledge your report and keep you updated.
What we collect, where it's stored, which providers process it and how long we keep it is covered in our Privacy Policy. For security questionnaires or a data processing agreement, contact [email protected].