Encrypted in transit and at rest
All traffic is served over TLS; workspace content is encrypted at rest using industry-standard ciphers.
Scoped agent keys
MCP keys are scoped to specific workspaces, projects, and capabilities, and can be revoked in one click.
Least-privilege access
Role-based controls for teammates and capability scopes for agents. Nothing has more access than it needs.
Full audit trail
Every action, human or agent, is logged with actor, timestamp, and what changed.
Resilient by design
Regular backups, routine recovery drills, and rate-limited APIs keep your workspace durable and protected from abuse.
Your data stays yours
We never train models on your content. Model providers we route through operate under zero-retention contracts.
Our approach
We practise defence in depth: narrow attack surface, short blast radius, and strong boundaries between users, workspaces, and agents. The same engineers who build the product also own its security posture, so controls evolve with the platform rather than as an afterthought.
Agent access controls
Agents interact with Stable Baseline through the Model Context Protocol using keys you generate and scope yourself.
- Keys are scoped per workspace, per project, and per capability set.
- Every tool call is attributed to its key, so revoking a compromised key is immediate and precise.
- Rate limits protect your workspace from runaway agents.
- Write capabilities can be separated from read, which is useful for reviewer-only automations.
Data handling
- Workspace content is stored with strict tenant isolation so one customer can never access another's data.
- File uploads (images, data files) are served via short-lived, signed URLs and are never publicly indexable.
- We do not use your content to train models, and we do not share it with model providers beyond the ephemeral context required to complete a request.
Privacy by default
Stable Baseline is built around the premise that your documentation and plans are confidential. Model calls made on your behalf are routed through providers that operate under zero-data-retention contracts. Telemetry we collect is limited to what we need to operate, support, and secure the service.
Incident response
Suspected incidents are triaged immediately, with affected customers notified in line with applicable law. If you think something is wrong, tell us. We would rather look at a false positive than miss a real one.
Responsible disclosure
If you've found a security issue, please report it to [email protected]. We also publish a security.txt with current contact and policy information. We commit to acknowledging reports promptly and keeping you informed as we investigate.
Enterprise assurance
If your organisation requires additional documentation (subprocessor lists, a Data Processing Addendum, architecture detail, or a security questionnaire response), please get in touch and we'll work with your team directly.