Changelog
Full AI Platform — Intelligence, Automation & Control
v2.0 transforms Nexus One AI from a chat and document Q&A tool into a complete on-premises AI operating platform. This release adds agents, workflow automation, meeting intelligence, secure collaboration, and a full administrative control layer — all running entirely on your hardware.
🧠 Intelligence Tools
- Multimodal Chat Prompt Studio — Full rewrite with persistent history, sidebar search, image upload, per-conversation model switching, and markdown rendering with code blocks
- Document Intelligence Workbench — Side-by-side document comparison, table extraction, batch processing, and structured output from uploaded files
- Meeting Assistant — On-premises meeting transcription (Whisper), AI summaries, action item extraction, and follow-up email drafting from audio recordings
- Secure Chat Rooms — Multi-user AI-assisted collaboration spaces with AI moderation, private to invited members, no external relay
⚙️ Automation Tools
- Agent Builder — Build autonomous AI agents with tool access, multi-step planning, and governance/approval layer for sensitive actions
- Workflow Automation — Visual pipeline builder for chaining AI tasks: trigger → process → output, with support for scheduled and webhook triggers
- Scheduled Jobs — Cron-based AI task scheduling: daily reports, nightly document ingestion, automated monitoring — runs without user interaction
- Connectors — Integration layer for databases, file shares, REST APIs, email, and SharePoint — usable by agents and workflows
🛡️ Control & Quality Tools
- Guardrails — Keyword blocks, PII redaction, regex rules, and prompt injection detection with block/redact/warn actions
- Model Router — Automatic query routing to the optimal model based on task type, token count, or custom rules
- RAG Quality Dashboard — Chunk quality scoring, retrieval accuracy monitoring, and query testing for knowledge base collections
- AI Eval Suite — Standardised test suites for model and prompt quality, with pass rate tracking and regression detection
- Model Benchmarking — Speed and quality benchmarks across installed models with tokens/second and TTFT metrics
- API Key Manager — Generate and manage API keys for programmatic access with per-key rate limits and model restrictions
📖 Portal & Knowledge Center
- Portal expanded from 20 to 43 pages covering all new features
- Quick Start updated with tabbed layout: First Steps, Platform Features, Admin Setup
- FAQ expanded with AI Tools, Automation, and new Technical Q&As
- Use Cases updated with agent, meeting assistant, and automation scenarios
- Glossary updated with 15 new terms covering all new features
- Light lavender theme applied consistently across all 43 pages
- Navigation fixed and standardised across all pages with complete AI Tools section
🤖 New AI Models
- llama3:70b — Added for high-accuracy complex reasoning tasks
- codellama:34b — Added for code generation and technical review
- llava:13b — Added for vision/image understanding queries
- nomic-embed-text — Dedicated embedding model for RAG (required)
Basic Tier — Initial Deployment
First deployment of the Nexus One AI Basic Tier system. Full AI stack installed and configured on-premises with the following components and capabilities.
🛠️ Tools Installed
- Ollama — Local AI model engine running on NVIDIA RTX Pro 6000 GPU
- Open WebUI — Browser-based chat interface accessible at http://ai.local:3001
- ChromaDB — Vector database for document storage and RAG
- LangChain — Python framework for building AI pipelines
- Jupyter Notebook — Interactive Python environment at http://ai.local:8888
- FastAPI — REST API layer for application integration
- DCGM — NVIDIA GPU health monitoring and diagnostics
- nvtop — Live GPU dashboard for terminal monitoring
- QLoRA / Unsloth — Efficient model fine-tuning toolkit
🤖 AI Models Available
- llama3.1:8b — Primary recommended model for general tasks
- llama3.2:3b — Fast lightweight model for simple tasks
- mistral:7b — Strong multilingual and structured output model
- llama3.1:70b — High-accuracy model for complex reasoning
- gemma2:9b — Google's efficient model with strong instruction following
- phi3:mini — Ultra-fast 3.8B model for quick responses
- nomic-embed-text — Embedding model for document search
- mxbai-embed-large — High-accuracy embedding model
📖 Knowledge Portal Launched
- Home page with live service status monitoring (auto-refreshes every 60 seconds)
- Quick Start guide — 5 steps to first AI conversation
- Tool pages for all 9 installed components with direct access buttons
- Prompt Library — 32 ready-to-use prompts across 8 government departments
- Use Case Gallery — 18 real-world scenarios with time savings benchmarks
- Model Library — comparison table and "which model should I use?" guide
- Troubleshooting — 9 common issues with searchable accordion
- FAQ — 14 plain-English answers to common questions
- Glossary — 18 AI terms explained
- Security & Privacy — air-gapped architecture and data handling guide
- Admin Guide — service management, model management, log access, and procedures
- What's New — this page
Pro Tier Upgrade
The Pro Tier upgrade adds enterprise-scale AI infrastructure for higher concurrency, larger models, and production-grade deployment capabilities.
🔜 Planned additions
- vLLM — High-throughput inference engine for multi-user production workloads
- Ray — Distributed computing framework for parallel AI jobs
- DeepSpeed — Memory-efficient training for larger model fine-tuning
- Milvus — Scalable vector database replacing ChromaDB for large document collections
- MLflow — Experiment tracking and model registry
- Additional large models (70B+ class) optimised for the upgraded hardware
- Portal tier switcher — toggle between Entry, Mid, and Advanced documentation
Max Tier Upgrade
The Max Tier adds Kubernetes-based orchestration and enterprise model serving for high-availability, large-scale AI deployment.
🔜 Planned additions
- Kubeflow — ML pipeline orchestration on Kubernetes
- Triton Inference Server — NVIDIA's production model serving platform
- Megatron-LM — Large-scale model training across multiple GPU nodes
- High-availability configuration with automatic failover
- Role-based portal access — admin console with embedded terminal
- Usage analytics dashboard for administrators
📝 For administrators: Update this page whenever a new model is installed, a service is upgraded, or a configuration change is made. Edit
whats-new.html and add a new entry above v1.0 following the same format. Include the date, a brief description, and what changed.