How Definable keeps your data safe

Workspace isolation

Every request is bound to one workspace when it signs in, and every database read and write is filtered by that workspace, so one workspace can't read another's.

Encrypted credentials

Model provider keys, channel credentials, MCP credentials and agent secrets are encrypted with AES-256-GCM before they're stored, sealed to the workspace and record they belong to.

Hashed sessions

The sessions we issue are stored only as SHA-256 hashes, so a copy of the database can't be used to sign in.

App sign-ins held by Composio

The sign-ins for apps a workspace connects are held by Composio; our database keeps only a reference to each connection.

Secrets kept out of output

Agent secrets are replaced by their names in sandbox output, and secret values are masked wherever the application prints them.

No training on your data

We never train AI models on customer content, and model providers that run on our own accounts work under commercial terms that forbid it.

Retention limits

Platform telemetry is kept for 7 days and conversation sandboxes are deleted 30 days after their last use. Workspace content stays until the workspace deletes it.

Data location

Customer data is stored in Germany and the United States, and is never transferred to a country the Government of India restricts.

No passwords

Definable stores no passwords. People sign in with Google, GitHub or a one-time emailed link, handled by Stytch.

One-time sign-in codes

Codes for agent apps and the Chrome extension are stored hashed, expire after 10 minutes, allow 5 tries and are limited to 3 sends an hour.

Short-lived sessions

Dashboard session tokens last 5 minutes and renew in the background. Extension and agent-app access tokens last an hour, with refresh tokens that rotate.

Roles

Owners, admins and members. Only owners can change someone's role or delete a workspace.

Sandboxed code

Code an agent runs executes in a container of its own for each conversation, apart from our application and from other conversations, and stops after 15 idle minutes.

Browser allow-list

The API accepts browser requests only from an allow-list of our own origins.

Reviewed changes

Every change is reviewed in a pull request before it merges.

Approvals before actions

Customers choose which actions need a person's approval. The agent pauses until someone decides, and every decision is recorded.

Every step traced

Each run records its routing, model calls, tool calls and waits, so any answer can be traced back to what produced it.

Evals before publish

Agents can be required to pass evals, compared against the last good version, before a change goes live.

AI policy

An AI policy set by management governs how we build, provide and use AI, under our ISO/IEC 42001 certification.

Risk and impact assessment

AI features are assessed for risk and for their impact on people before release, and again when they change.

Model choice

Customers choose the model behind each agent and can use their own agreements with model providers.

Visible routing

When an agent has a fast and a main model, the trace shows which one took each message and why.

AI supplier review

Model providers are assessed before we use them, and those that run on our keys are listed as subprocessors.

Certified data centres

We run on Hetzner servers in Germany and store files in Amazon S3 in the United States, in data centres certified to ISO/IEC 27001.

Encryption in transit

Traffic to the app, the API and the Chrome extension travels over HTTPS, with TLS certificates that renew automatically.

Files encrypted at rest

Files are stored in Amazon S3, which encrypts every object at rest.

Separate environments

Development and production run as separate environments.

Central telemetry

Platform logs, traces and metrics are collected in one place, monitored and kept for 7 days.

DPA for every customer

Our Data Processing Addendum applies to every customer automatically, with Standard Contractual Clauses for transfers from Europe, the UK and Switzerland.

Data rights

People can see, correct, erase or move their data, withdraw consent, nominate someone to act for them and raise a grievance. We answer within 30 days.

Grievance Officer

A named Grievance Officer acknowledges complaints within 24 hours and aims to resolve them within 15 days.

Breach notice

Customers hear about a personal data breach affecting their data within 48 hours.

Subprocessor notice

Workspace owners get at least 30 days' notice by email before a new subprocessor handles their data, and can object.

No selling, no ads

We don't sell personal data or use it for advertising.

No trackers

Our website sets no cookies and loads no analytics, and the app uses only the cookie that keeps you signed in.

Security programme

A written information security programme with a named owner, reviewed every year.

Internal audit and review

Internal audits and a management review of our security and AI management systems every year.

Risk management

Risks to the Services are assessed and treated regularly, each with an owner.

Supplier review

Suppliers are assessed before we engage them and reviewed while we use them.

Least-privilege staff access

Staff get production access by role and only as support and operations need it, reviewed regularly and protected by multi-factor authentication.

Incident response

A documented incident response process, tested regularly, that includes our breach notices.

Business continuity

Regular backups, tested restores and redundancy for critical systems.

Confidentiality agreements

Everyone at Definable signs a confidentiality agreement.

Background checks

Background checks before people join, where the law allows.

Security and privacy training

Everyone at Definable is trained in security and privacy.