AI development incoming: Smarter JeffAI, HAVEN RAG, and Scalable AI Infrastructure

AI development incoming: Smarter JeffAI, HAVEN RAG, and Scalable AI Infrastructure

We have released some upgraded models for our AI functionality. We are excited for what this will mean for customers with three core improvements which are expanded upon below.

Agentic JeffAI: a smarter, safer JeffAI with new agent architecture

What is it?

Agentic JeffAI is the new, modular architecture for JeffAI that enables JeffAI to work with a set of focused specialist agents for different tasks (i.e for understanding prompts, creating or modifying diagrams, and generating projects), rather than a single “mega‑agent” trying to do everything at once. This major upgrade enables JeffAI to deliver consistent and more predictable results, with better attention to detail. 

How does it work?

Agentic JeffAI works by sending every request through a central triage agent that checks for safety and relevance, then routes it to specialised agents for tasks like understanding your prompt, creating or updating diagrams, or generating an IriusRisk project. These agents return structured, validated outputs behind the scenes, so Jeff still feels like a single assistant while delivering more predictable, safer, and easier‑to‑understand results across prompt, image, and prompt+image flows.

What benefits does it bring?

More reliable, focused JeffAI behaviour: 

• Customers experience more predictable outputs, with Jeff following instructions more consistently and staying on-topic rather than producing overly long or meandering answers, “going off script” or mixing concerns in a single reply
• Map components correctly, replace components correctly upon request, avoid losing components, better compliance to instructions in conversation

Higher-quality diagrams and conversations with less back‑and‑forth

• Better prompt design, structured outputs and clear rules per agent reduce errors like misinterpretation of requirements, inconsistent formats, or half‑completed actions.
• Users will now spend less time correcting Jeff or re-phrasing prompts to “steer” it, so they can move faster from idea → diagram → threat model

Safer, enterprise-ready JeffAI interactions

• Clear guardrails and input/output validation at each agent step reduce the risk of unexpected content or malformed outputs reaching users 
• Security teams and decision-makers evaluating JeffAI have increased trust and confidence in pushing for wider deployment.

HAVEN: smarter component matching for JeffAI and diagram imports 

What is it?

HAVEN (Holistic Archive for Vectorized ENgineering) is a new context and retrieval system that replaces the previous RAG layer used by JeffAI. It improves how architectures and diagrams are understood and mapped to IriusRisk components.

How does it work?

HAVEN analyses project context and diagram elements, searches across IriusRisk data and catalogs, and returns more accurate component matches with clearer confidence signals. It is backed by a high-availability Redis architecture and is used consistently across diagram imports and JeffAI interactions.

What benefits does it bring?

• More accurate diagram imports: Fewer unmapped or incorrectly mapped components.
• Clearer AI confidence: Weak matches are ignored, while deterministic mappings always take priority.
• Less manual cleanup: Users spend less time fixing AI-generated models.
• More consistent results: The same smarter context is used whether starting from a diagram or working with JeffAI chat.
• Built for real-world use: High availability, graceful failure handling, and support for both cloud and on-prem deployments. 

Securing and scaling cloud infrastructure for JeffAI and HAVEN

What is it?

This enhancement strengthens the cloud infrastructure that powers JeffAI’s agentic system and the HAVEN context engine, ensuring it is secure, scalable, and production-ready.

How does it work?

While the underlying LLMs remain unchanged, the AI platform has moved from experimental and free-tier environments to enterprise-grade cloud subscriptions. This involved selecting suitable vendors, completing due diligence, and onboarding a cloud foundation designed specifically for AI workloads like JeffAI and HAVEN.

What benefits does it bring?

• Improved reliability and availability for AI-driven features.
• Better performance at scale as usage grows.
• Stronger security and compliance controls suitable for enterprise environments.
• A future-proof foundation that allows JeffAI and HAVEN to evolve safely without compromising stability or cost efficiency.

‍