TL;DR. DeepCube 1.12 turns Tessa from one assistant into 46 specialists, grouped into 6 skills. They come from the open-source Jerry framework. Tessa can now read a screenshot you paste and tell you what is wrong in it.

Immersive Fusion has released DeepCube 1.12.0. It is the biggest update to the AI Assistant since it launched in September 2025. Tessa used to be one assistant you talked to. She is now a team: 6 skills, 46 purpose-built agents, and the ability to read screenshots as well as text.

Skills: From One Prompt to Forty-Six Specialists

Earlier versions of Tessa worked as one agent you talked to. Powerful, but general. Version 1.12 adds the Skills system. It sorts what Tessa can do into domains, and gives each domain its own agents, built for that kind of work.

The six skills shipping in 1.12.0:

Problem Solving (9 agents). Works a problem through, step by step. It has specialists for research, root cause, design decisions, quality scoring, checking, synthesis, code review, failure digging, and progress reports. Each one follows a known method. The investigator uses 5 Whys and Ishikawa diagrams. The architect writes Nygard-format Architecture Decision Records. The critic scores the work against six quality measures.

Red Team (11 agents). Covers the full MITRE ATT&CK kill chain for security testing. Nothing runs unscoped. An Engagement Lead agent must set Rules of Engagement first. Three agents, for persistence, exfiltration and social engineering, have to be named in those rules before they will run. It puts real penetration testing in reach of teams who own both uptime and security.

Engineering Team (10 agents). Security-first build work, from design through incident response after a deploy. It covers threat modeling with STRIDE, DREAD and PASTA. It covers backend and frontend security (OWASP Top 10, ASVS 5.0), hardening (CIS Benchmarks, SLSA), pipeline scanning (SAST/DAST), security QA, and incident runbooks.

NASA Systems Engineering (10 agents). Runs the NPR 7123.1D processes. Agents cover requirements, technical design, verification and validation, system integration, risk (NPR 8000.4C), configuration control, formal review gates (SRR, PDR, CDR, FRR), trade studies, quality, and status reports.

Adversary (3 agents). How Tessa argues with herself. It maps the stakes of a job (C1 advisory up to C4 mission-critical) to a set of attacks, runs those attacks on the draft, and scores it across six measures with a PASS, REVISE or ESCALATE verdict. The assistant reviews its own answer before you see it.

Prompt Engineering (3 agents). A skill for writing better prompts. It walks you through the five parts of a prompt, writes the constraints, and scores the result against seven tests.

Built on the Jerry Framework

All skills come from the Jerry framework, an open-source multi-agent system created by Adam Nowak. Jerry defines the skill designs, the agent specs, the routing rules, and how quality gets scored. Immersive Fusion ported those definitions into the DeepCube assistant library. The domain knowledge, the constraints and the workflows are still Jerry's. What changed is that they now run natively inside DeepCube. Every ported skill carries a header citing the Jerry source path, version, and license.

Automatic Routing

You do not need to know that 46 agents exist. Routing happens in two passes. Common questions match on keywords and route at once. Anything unclear falls back to the model, which reads intent. Ask "why is checkout slow?" and Problem Solving picks it up. Ask "review this for security vulnerabilities" and the Engineering Team answers. Ask "what's the attack surface?" and the Red Team engages, once scope is checked. You just ask the question.

Multimodal Vision

DeepCube 1.12 adds multimodal vision support. Paste a screenshot into the AI Assistant with Ctrl+V and ask Tessa what she sees. A dashboard, an error dialog, a log, a topology view: paste it and ask "what's wrong here?" Tessa reads the image, checks it against the telemetry she can already see, and gives you a diagnosis.

That cuts out the describe-copy-paste-wait cycle that slows an incident down. Instead of typing out what is on the other screen, just show it.

Try It

DeepCube 1.12 is available today for all subscription tiers, including Start (free).

To try the Skills system and vision, start free and fill your grid with real-looking telemetry from our open-source trace generator:

go install github.com/ImmersiveFusion/snowglobe/cmd/snowglobe@latest
snowglobe -endpoint otlp.deepcube.ai:443 -headers "api-key=YOUR_KEY" -complexity light

Then ask Tessa anything. She'll route to the right specialist.

Watch it in action: From Error to Full RCA in One Conversation | Introduction to DeepCube

About Immersive Fusion

Immersive Fusion LLC builds DeepCube, an application performance monitoring platform that draws distributed systems in real-time 3D. We think engineers deserve to see their applications the way those applications really run. DeepCube joins spatial views, AI-guided diagnosis, and full OpenTelemetry support to give observability that goes past dashboards.

Download: immersivefusion.com/download | Steam | Documentation

Media Contact: Dan Kowalski, dkowalski@immersivefusion.com

Start Free. Immersive. AI-guided. Full-stack observability. Enter the World of Your Application®.