What's New

This release expands Cranium's AI discovery, transparency, and continuous monitoring capabilities, adding visibility into agentic AI architectures and system prompts, automated BOM updates on code push, and traceability between Arena models and the repositories using them.

AgentSensor

Cranium's CodeSensor now detects AI agent frameworks in your code repositories and automatically generates an Agentic Systems BOM within the existing AI BOM view. The Agentic Systems tab captures each agent's name, instructions, model, role, dependencies, and inter-agent relationships, giving security and governance teams a structured map of multi-agent architectures without manual inventory work.

Agentic Systems data is included in CycloneDX exports, and manual edits to agent relationships are supported the same way as other BOM components. The Agentic Systems tab appears in the AI BOM detail view for any scanned repository where a supported framework is detected. Supported frameworks at launch are:

Amazon Strands
CrewAI
Hugging Face smolagents
Microsoft AutoGen-AgentChat
LangGraph
OpenAI Agents

Additional frameworks will be added in subsequent releases. API access to Agentic Systems data and detection of dynamically assembled prompts are planned for future releases.

Note: Agentic Systems data will only appear for BOMs created or rescanned after this release. If you do not see any data in the Agentic Systems tab for an existing repository, rescan the repository to pick up agent framework detection.

System Prompt Detection

Static system prompts captured by CodeSensor now surface directly in the AI BOM, nested under the model they govern in a read-only panel. Security analysts can use this view to validate expected model behavior, and compliance teams can document AI system configuration as part of audit workflows.

System prompts detected in the BOM also populate in the Shield run configuration dropdown, enabling downstream testing of models with known prompts. Dynamic prompt assembly—where prompts are constructed at runtime—is outside the scope of this release. Direct editing of system prompts via the UI is planned for a future Shield integration.

Note: System prompt data will only appear for BOMs created or rescanned after this release. If you do not see a system prompt panel for an existing model, rescan the repository to surface any detected prompts.

Continuous Monitoring

Cranium now automatically rescans a repository's BOM whenever code is pushed, keeping your AI inventory current without manual intervention. Continuous Monitoring is powered by webhook integration with your VCS platform — when a push event is received, Cranium triggers an incremental rescan of the changed files and updates the BOM accordingly. Supported platforms are GitHub, GitHub Enterprise Cloud, GitLab, and GitLab Self-Managed.

When a BOM changes as a result of a rescan, stakeholders with published AI Cards for the associated AI System can receive email notifications summarizing what was added or removed. To enable these notifications, turn on the Continuous Monitoring toggle on the relevant AI System.

For full setup guidance, visit the Continuous Monitoring Overview in the Knowledge Base.

Arena Model BOM Discovery

The Arena Model Detail page now includes a BOM Discovery panel listing every AI BOM in which that model has been detected. Each entry links directly to the corresponding BOM detail page. The list is paginated to maintain performance across large datasets.

Did this answer your question?