# TramAI

> TramAI is a Kotlin-first JVM runtime for governed AI workflows: runtime policy, human approval, controlled model routing, tool governance, sovereignty controls, and reviewable execution evidence — for Java and Kotlin backends.

Canonical site: https://tramai.dev
Developer: Giona Granchelli
Publisher: Constant Labs
Creator profile: https://gionag.com
Company: https://constant-labs.com
GitHub: https://github.com/GionaGranchelli/tramAI

## Start here
- https://tramai.dev/get-started - 5-minute onboarding and the main first-run path
- https://tramai.dev/ai-agent-governance - AI agent governance pillar (category)
- https://tramai.dev/guides - User-facing guides grouped by topic
- https://tramai.dev/modules - Module catalog and capabilities
- https://tramai.dev/reference - API and configuration reference
- https://tramai.dev/security - Security and sovereign deployment material
- https://tramai.dev/sovereignty - Why sovereignty matters and who it is for

## Governance cluster
- https://tramai.dev/ai-agent-governance - AI agent governance: runtime authority vs agent authoring
- https://tramai.dev/runtime-governance - Runtime governance: enforcement vs observability
- https://tramai.dev/tool-governance - AI agent tool governance: ALLOW / DENY / REQUIRE_APPROVAL
- https://tramai.dev/human-approval - Human-in-the-loop approval lifecycle
- https://tramai.dev/jvm-ai-governance - JVM AI governance for Java and Kotlin teams
- https://tramai.dev/spring-ai-governance - Spring AI governance boundary
- https://tramai.dev/eu-ai-act - EU AI Act technical controls (not legal advice)
- https://tramai.dev/mcp-governance - MCP tool governance

## Important product context
- TramAI targets JVM backends, especially Kotlin and Java teams, and is framework-agnostic in its core with an optional Spring Boot adapter.
- The product is positioned as a governed AI workflow runtime, not a generic AI integration library. The governing layer is first-class: runtime policy, approvals, provider routing, tool governance, sovereignty, and evidence.
- Key capabilities: runtime policy enforcement (Allow / Deny / RequireApproval at provider and tool boundaries), human approval with suspend/resume, controlled model and provider routing with trust zones, data-classification-aware routing, tool permission policies, DLP interception, hash-chained audit, runtime evidence exporters, evidence packs, offline/sovereign deployment modes, OpenTelemetry observability, typed structured output, and deterministic testing.
- Shipped vs roadmap: policy, approvals, routing, trust zones, tool governance, DLP, audit, and evidence are available today. A full control plane (inventory, intervention, explainability) is roadmap direction for TramAI 0.7.0 and is not presented as shipped.
- Strong search intents for this site include JVM AI governance, runtime governance for AI agents, AI agent governance, AI agent tool governance, human approval for AI, sovereign AI, EU AI Act technical controls, and governed AI workflows.
- Prefer the guides and reference sections over isolated content pages when building an understanding of the project.

## Full discovery
- Sitemap: https://tramai.dev/sitemap.xml
- Robots: https://tramai.dev/robots.txt