Company
Security and trust
Collabre is in private beta. This page lists the controls the product runs today, and each one is checked against the module that implements it.
How this page is built
The controls below are read from a data file at build time. Every row in that file names the module that implements the control, so a maintainer can reopen the code behind a claim and check it before this page is deployed again. A control whose module has moved, or whose behavior has changed, is corrected or removed. The page names no certification and no availability figure, because no evidence for either exists in the repository.
Authentication
- SAML single sign-on
- Configured per Workspace against the identity provider's own endpoint and certificate, with the certificate held encrypted and never returned by the API.
- Required single sign-on
- A Workspace can require every member to sign in through its identity provider, so no one signs in around it.
- Two-factor authentication
- Time-based one-time codes from an authenticator app, with recovery codes issued when it is switched on.
- Recovery codes
- Issued once, stored as hashes, and each one works a single time.
- Magic link sign-in
- A single-use link sent by email, valid for fifteen minutes, stored only as a hash.
- Password sign-in
- Passwords are stored as bcrypt hashes and never in readable form.
- Login activity
- Every sign-in attempt is recorded with its method, address, browser, and whether it succeeded, and the member can read their own history.
Access
- Workspace roles
- Three roles decide who runs the Workspace: an owner with billing and deletion, an administrator, and a member.
- Space roles
- Inside a Workspace, a space has its own administrator, contributor and viewer roles.
- Team roles
- A team has its own administrator, editor and viewer roles over the Brands it holds.
- Campaign roles
- A member joins a Campaign in a named specialist role, the same roles an Agent can fill.
- Content Slot roles
- Each Content Slot is assigned to a writer, a designer or a reviewer.
- Approval roles
- An approval stage requires an author, a reviewer or a publisher.
- Platform roles
- Collabre staff accounts are separated into ordinary, support and administrator roles.
- Support access
- Collabre support cannot open a Workspace on its own. Access is granted from the Workspace, it carries an expiry date that is always set, and the default level is read-only. A grant that goes beyond read-only is a separate level that the Workspace owner has to choose. Granting access and revoking it are both written to the audit log.
Audit
- Per-entity history
- The log can be filtered to one object, so a Campaign, a post or a task shows its own history.
- Workspace-wide log
- One searchable log per Workspace, filterable by action and by person, newest first, open to Workspace members.
- Redaction of log detail
- Every entry's detail is run through the redactor before it is written, so personal data does not land in the log.
- Agent actions in the log
- Agent runs and tool calls write their own audit entries.
- Compliance export
- An owner or an administrator can export a compliance report for any date range. It gathers the tool allow and deny rules in force at the end of the range, the decisions taken on destructive actions, Agent run cost and every kill-switch activation, the most frequent audit event types, the redaction setting, and how much of the spending caps was used. It comes out as a PDF file or a CSV file, and the export itself is written to the audit log.
AI governance
- Spending caps
- Caps apply at three levels: per Agent run, per Agent per day, and per Workspace per month.
- Kill switches
- Halt one Agent, or every Agent in the Workspace, at once. Work in flight is canceled.
- Tool approval rules
- Allow, deny or require confirmation for any tool, scoped to the Workspace, an Agent, a Department or a Campaign.
- The default answer
- When no rule matches a tool call, the answer is to ask a person.
- Destructive action registry
- Actions that publish, pause an ad, send email, change a budget, swap a creative or delete a resource are classified and governed. Deleting a resource is denied unless a Workspace widens it.
- Autopilot level
- Set per Campaign, and it defaults to Suggest Only.
- PII redaction
- On by default, and it scrubs personal data from Agent inputs and from logs. When the setting cannot be read the service treats redaction as on.
- Agent rate limits
- Limits on how many runs an Agent and a Workspace can have in flight, how many runs start in an hour, and how often one tool is called.
- Spend forecasting
- A projection of Workspace AI spend from its own recent history, so a cap can be set before it is reached.
Data
- Encryption of credentials
- Platform access tokens, integration credentials and two-factor secrets are encrypted with AES-256-GCM before they are written to the database. Each record carries its own salt and initialization vector.
- The encryption key
- The key is derived from an environment variable that every deployed environment has to set. The API refuses to start in a deployed environment when that variable is missing.
- Transport
- The API sets the standard security headers, including strict transport security.
- Object storage
- Files are stored in S3-compatible object storage that the deployment configures. The endpoint, the bucket and the credentials are required environment variables, so there is no quiet fallback to a local disk, and the API does not start in a deployed environment without them.
- Storage quota
- Each Workspace has a storage quota that comes from its plan, and usage is visible in the product.
- Version history
- Replacing an asset keeps the previous version and moves an active-version pointer.
- Agent memory retention
- Agent memory entries expire after 30 days by default.
- Snapshot retention
- Agent memory snapshots expire after 7 days.
- Tool analytics retention
- Tool call analytics are pruned after 30 days.
- Getting data out
- Analytics and reports export to CSV and PDF, and a REST API with scoped keys reads Workspace data.
Subprocessors
Collabre uses a small set of vendors to host the product, store files, take payments, send email and run AI models. Each of them, what it is used for and the data it receives, is listed on the subprocessors page, with the date the list was last reviewed. The data processing agreement sets out the terms Collabre handles customer data under.
Reporting a vulnerability
Send security reports to [email protected]. Include enough detail to reproduce the issue, and leave customer data out of the report. Please give us a chance to fix the issue before describing it publicly.
See the controls in a Workspace
Request a demo and we will walk through the roles, the approval gates and the Agent spending caps with you.
Last verified against develop @ 12887a33 on 2026-09-13.