# Security & Trust > How Syncon One protects what customers tell a business, described as it works today. - Canonical page: https://synconone.com/enterprise/security - Product: Syncon One, by SynconAI - Updated: 2026-10-22 ## Data protection Your conversations, customer records and files belong to your workspace, and you can take them out or remove them. - Conversations, people, companies, leads and files are stored against one workspace. - Records can be exported as CSV by people whose role allows it. - A person, visitor, conversation or the whole workspace can be deleted on request. - Every conversation is kept in full, agent and teammates together, so you can see exactly what a customer was told. **How deletion runs.** A deletion request names one subject: a person, a visitor, a conversation or the workspace. A background worker carries it out after a 24-hour grace period, so a mistaken request can be withdrawn the same day. ## Authentication and permissions People sign in to their own accounts, can add a second step, and see only what their role allows. - Two-step sign-in with an authenticator app, with backup codes stored encrypted. - Sign-in, sign-up and password reset are rate limited for each IP address. - Five roles in every workspace: Owner, Admin, Manager, Teammate and Viewer. - Custom roles with field-level permissions on Pro and Enterprise. - Permissions are checked on the server for every action, never by hiding a button. **Sign-in limits.** In production, sign-in is limited to 8 attempts a minute, sign-up to 5 every five minutes and password reset to 3 every five minutes. The counters are kept in the database, so every server instance shares them. ## Workspace isolation Each business is its own workspace, and one workspace cannot read another's conversations or records. - Every query for workspace data is scoped to the workspace on the server. - The workspace is taken from the signed-in session, not from anything the browser sends. - Public API keys belong to one workspace and reach only that workspace's data. ## AI data handling The agent answers from material you give it, records where each answer came from, and sends conversation text to a model provider to write its replies. - Conversation text and the relevant passages from your material are sent to the AI model provider to produce each reply. - On Free, and on Enterprise if you choose, the AI runs on your own provider key (OpenAI, Anthropic, Google or Groq), under your own contract with that provider. - The source of each answer is recorded with it, so anything the agent said can be checked against what you published. - Questions it could not answer are collected for your team rather than improvised. - Agent policy sets subjects that always go to a person and things it may never promise. ## Encryption Everything travels over TLS, and the secrets we hold for you are encrypted before they are stored. - All traffic between visitors, your site, your team and the platform uses HTTPS. - Browsers are told to use HTTPS only (HSTS, two years, including subdomains). - Provider keys, calendar and mailbox tokens, channel credentials and webhook secrets are encrypted with AES-256-GCM before they reach the database. - Public API keys are stored only as a hash; the full key is shown once, when it is created. **What is and is not encrypted by the application.** Secrets are encrypted in the application with AES-256-GCM, a fresh random IV for each value, and an authentication tag checked on every read. Conversation text and customer records are protected by access control and transport encryption; we do not describe them as application-level encrypted, because they are not. ## Integrations and credentials Connections use each provider's own sign-in, and what they hand back is stored encrypted. - Google and Microsoft calendars and mailboxes connect through OAuth; we never see the account password. - The OAuth state is signed and checked on return, so a connection cannot be completed by someone else's link. - Outgoing webhooks are signed with HMAC-SHA256 over a timestamp and the body, so receivers can reject forged or replayed deliveries. **Verifying a webhook.** Each delivery carries an x-syncon-signature header of the form t=,v1=. Reject a delivery whose timestamp is more than 300 seconds old or whose signature does not match, compared in constant time. ## Data retention and deletion You choose how long closed conversations and uploads are kept; by default they are kept until you remove them. - Closed conversations can be removed automatically after 30, 90, 180, 365 or 730 days. - Uploaded files can be removed on their own schedule, from the same choices. - Workspace exports are stored for seven days and then removed. - Deletion requests run after a 24-hour grace period. **Where this is set.** Settings → Data in the workspace. A policy is two numbers of days, or nothing, which means keep, and the worker applies it every hour. Anything outside 30 to 3,650 days is clamped into that range. ## Infrastructure and hosting The application and its database run on servers SynconAI operates; files are kept in private object storage. - The web application, background worker and PostgreSQL database run in containers on servers SynconAI operates. - The database is backed up daily. - Uploaded files and exports are stored in private Cloudflare R2 buckets with no public access. - Email is sent through Resend; payments are taken by Creem, our merchant of record. - A hosting region cannot be chosen per workspace today. ## Auditability and governance Consequential actions are recorded with who did them and when, and every answer can be traced to its source. - Changes to connections, channels, knowledge, teammates and conversation ownership are written to an audit record with the person, the time, and the IP address and browser. - Every handover to a person carries the reason and a brief. - Changes to the agent are published deliberately rather than edited live in front of visitors. - A browsable audit log in the workspace is not available today; records can be provided on request. ## Operational security The site and the widget are hardened against the common web attacks, and input is checked at every boundary. - A content security policy with a fresh nonce on every request. - Pages cannot be framed by other sites, and browsers are told not to guess content types. - The website widget renders inside a shadow root, so it cannot read or restyle your page. - Everything that arrives from outside, from forms to the public API, is validated before it is used. ## Not offered today - Single sign-on (SAML or OIDC) - SCIM user provisioning - A browsable audit log in the workspace - A choice of hosting region - SOC 2, ISO 27001 or other third-party certification of Syncon One - A contractual uptime commitment ## Policies - Privacy policy: https://synconone.com/privacy - Terms of service: https://synconone.com/terms - Security questions: contact@synconai.com