BatonX / Security

Control the environment around the agent.

Model API governance alone cannot control local file access, command execution or child processes. BatonX governs the session in which agent activity happens.

Conceptual security boundary: a coding agent operates inside an authorised project and governed session. Policy controls approved access, with evidence of decisions and security events.
Authorised projectGoverned sessionAudit evidence
Defined boundaries. Attributable decisions.

Public security principles

Apply least privilege and organisation-owned controls across the session, model access and evidence.

Project and session

  • Authorised project access only.
  • Isolated, non-privileged runtime sessions.

Routes and credentials

  • Default-deny network posture with approved destinations.
  • Model credentials kept outside the agent runtime.

Decisions and evidence

  • Versioned policy and attributable decisions.
  • Structured security events and integrity-oriented audit evidence.
  • Source code, prompts and responses are not centrally retained by default.

Policy inheritance

Organisation policy flows to project and repository levels. Lower scopes may tighten permitted rules; mandatory organisation controls cannot be weakened.

Policy inheritance

Organisation policy flows to project and repository levels, producing the effective policy applied to each session.

Conceptual architecture. Central standards with controlled project and repository refinement.

Evidence without overcollection

BatonX can record identities, timestamps, versions, policy decisions, operation types and integrity hashes while avoiding central storage of source content by default.

Evidence according to policy

  • Identities and timestamps
  • Versions and policy decisions
  • Operation types and integrity hashes
  • Session and security-event metadata

Not centrally retained by default

  • Source code
  • Prompts and responses
  • Command output
  • Secrets

Customer-managed deployment keeps core services and retained evidence in the customer-controlled environment. Context required by an approved external model may still be sent to that endpoint.

Explore customer-managed deployment

Governance, not surveillance

BatonX records what is necessary for security, operations and audit. It is not designed to rank developers, score employee performance or monitor activity outside authorised BatonX sessions.

  • No developer rankings.
  • No employee performance scores.
  • No monitoring outside authorised BatonX sessions.

Security review

Detailed control design, deployment hardening and threat-model material are shared during a qualified security review under appropriate confidentiality.

Discuss a security review

Let’s discuss security

Contact us to discuss BatonX’s security approach, access controls and audit boundaries. We can explore your organisation’s security requirements and how they relate to your development environment.

Contact us about security