TramAI - governed AI workflows for Java and Kotlin

Sovereign AI, Approval Gates, and Policy Control

Some organizations do not just want AI features. They need controlled AI systems.

That usually means one or more of these constraints:

  • data cannot flow freely to any provider
  • tool execution must be gated or approved
  • policies must shape what the system is allowed to do
  • deployment needs to remain compatible with sovereignty or regulated infrastructure requirements

What Sovereign AI Means In Practice

For a software team, sovereignty is not a slogan. It is a set of operational and architectural constraints:

  • control over provider choice
  • control over deployment mode
  • control over approval and escalation points
  • control over policy enforcement
  • control over evidence and auditability

Where TramAI Fits

TramAI is aimed at teams that want these controls to be part of the AI integration layer itself, not an unrelated governance system sitting beside it. All of the following ship in 0.6.0:

  • sovereign mode — tramai.profile: sovereign with explicit model, provider, fallback, tool, and permission allowlists, provider trust zones, and fail-fast startup validation. The profile compiles to the deny-by-default policy configuration and rejects wildcards.
  • offline deployment — the OFFLINE deployment mode requires every route to target a local trust zone; offline deployment covers what that does and does not enforce at the network layer.
  • approval workflows — high-risk tool calls suspend with an ApprovalSuspendedException, a challenge bound to the workflow run, tool, argument digest, policy version, and workflow digest, and resume only when an authorizer presents the token with the expected approval and continuation versions. Side effects execute at most once across a resume.
  • DLP — regex-based redaction of sensitive data from model outputs and tool results, with redaction audit events.
  • artifact verification — digest receipts for local model artifacts.
  • evidence — hash-chained audit events, policy.decision / tool.permission / approval.decision runtime evidence, and evidence packs describing registry, trust zones, deployment mode, and verification receipts.
tramai:
  profile: sovereign
  sovereign:
    allowed-models: [local-invoice-model]
    allowed-providers: [deterministic-local-provider]
    allowed-tools: [schedule-payment]
    allowed-permissions: [payment.schedule]
    provider-zones:
      deterministic-local-provider: LOCAL
    models:
      local-invoice-model: deterministic-local-provider

Durable approval and audit state comes from tramai-spring-boot-starter-sovereign-persistence-file or -jdbc, so a suspended approval survives a restart. The boundary to be aware of: tramai.sovereign.enabled: false is rejected at startup — tramai.profile is the only runtime selector.

0.6.0 is a published release candidate, not a certified product or a compliance claim. See API stability for which surfaces are RC+ Stable and which are Preview.

When This Becomes Important

  • enterprises adopting AI in governed environments
  • corporations with internal approval requirements
  • teams operating under strict legal or security review
  • organizations that need a credible sovereign AI story rather than a generic wrapper around third-party APIs

Continue

Previous
Next