KEVOS®
EraNorth AI product
KEVOS

Private AI and engineering intelligence for serious professional work.

KEVOS AI is the modular AI product being developed by EraNorth to connect language models, organisational knowledge, research, specialist agents, multimodal tools and controlled workflow execution.

Open Cloud AI Development

KEVOS AI product

KEVOS development

Current capability, recorded as evidence.

KEVOS development now forms part of this product page. The public log shows only entries deliberately approved for release and was last updated 6 September 2026.

KEVOS AI Distribution Established

Installable KEVOS AI 1.0.0 local workspace prepared

KEVOS AI now has an installable local application path for Windows. The application is packaged as normal desktop software rather than requiring a developer checkout. On first launch it prepares the local model runtime and model set, starts the workspace services and opens the browser-based KEVOS interface on the user's own machine. Knowledge is deliberately not bundled with the installer; each installation begins with an empty private knowledge store that the user can populate and index.

Evidence: Windows 1.0.0 installer build produced; release distribution and first-run model setup are managed separately from the CMS source package.

EraNorth Academy Established

Professional AI learning platform and AI Solutions Architect programme

EraNorth Academy has been expanded into a structured professional learning environment with programmes, stages, courses, modules, lessons, labs, assessments, projects, capstones, learner progress, skills evidence and completion certificates. The AI Solutions Architect programme is taught with original EraNorth material and uses external certification providers only as optional validation references rather than as copied course content.

Evidence: Multi-level Academy architecture, original curriculum source files, assessment engine, learner progress, module certificates and admin curriculum controls implemented in the CMS.

Release Engineering Established

Versioned KEVOS AI release and download centre

EraNorth can manage KEVOS AI releases from the administration console using semantic versions, draft/published states, platform-specific installers, release notes, checksums and controlled public downloads. Large installer files are uploaded in browser chunks so each PHP request remains small even when the completed installer is much larger.

Evidence: Admin release workflow, protected release storage, chunk upload endpoint, SHA-256 recording, public Downloads and Release History routes.

KEVOS AI Core Active

Modular AI orchestration rather than one monolithic chatbot

KEVOS AI is built as a set of cooperating services. A request can be classified, routed to the appropriate model or capability, supplied with authorised context, combined with tools or specialist workflows, checked against evidence and returned with a capability trace. Local deployments can use local model runtimes, while hosted workflows can use approved external providers when configuration, permissions and data classification allow it.

Evidence: Separate orchestration, command routing, provider, knowledge, research, memory, validation and specialist intelligence services are present in the application architecture.

Knowledge & Evidence Active

Knowledge ingestion, retrieval and evidence-grounded answers

KEVOS does not treat uploaded documents as magical model training. Knowledge is handled as an explicit evidence layer: documents are parsed, normalised, segmented, indexed and stored with metadata; user questions retrieve relevant material; neighbouring context can be expanded; and the resulting evidence is supplied to the model with source identifiers. This keeps source knowledge separate from the base model and makes it possible to show citations, provenance and evidence gaps.

Evidence: Knowledge indexing/orchestration, citation handling, document parsing and evidence-context services are implemented and continue to be refined for retrieval depth and quality.

Research Intelligence Active

Multi-pass research with source comparison and uncertainty

Deep Research is designed to move beyond a single retrieval pass. The research workflow plans the question, gathers relevant internal or approved external evidence, analyses findings and conflicts, identifies gaps and then synthesises an answer with source references. Retrieved documents and webpages are treated as evidence rather than executable instructions, which is important for both reliability and prompt-injection resistance.

Evidence: Research planning, evidence review, citation persistence, source-conflict handling and staged synthesis are implemented in the research service.

Multimodal & Engineering Active

Document, image and controlled engineering intelligence

The multimodal path extends KEVOS beyond plain text so technical documents, images, screenshots and other visual information can be interpreted as part of a governed workflow. Engineering integrations are being approached as controlled assistance: AI can help inspect, explain and prepare work, but engineering approval and safety-critical decisions remain with qualified people and authorised processes.

Evidence: Document intelligence and vision service foundations are integrated; controlled CAD-oriented workflows remain under staged validation.

Specialist Intelligence Active

Specialist roles built on shared evidence and controls

KEVOS is moving from a single generic assistant toward specialist engineering, project/program, research and decision roles. Specialists share the same governed foundations—authorised knowledge, project context, memory, tools, permissions and validation—so expertise is added through orchestration and domain-specific behaviour rather than by allowing independent agents to act without boundaries.

Evidence: Specialist routing and intelligence services exist alongside shared governance, evidence and project-context services.

Memory & Runtime Active

Working memory, project memory and persistent context

Memory is being built as an explicit data capability rather than assuming the language model permanently learns from every conversation. Short-term conversation context, project memory, saved user preferences and longer-lived organisational context can be stored and retrieved under defined scopes. This allows useful continuity while keeping memory distinct from RAG, fine-tuning and model training.

Evidence: Conversation, project and long-term memory services are separated in the application architecture and can be governed independently.

Governance & Quality Active

Evidence boundaries, permissions, evaluation and safe fallback

Reliability is treated as an architecture concern, not a prompt slogan. KEVOS separates permissions, provider eligibility, evidence collection, citations, confidence, user feedback and fallback behaviour. Where authorised evidence is insufficient, the system is designed to state that limitation rather than invent organisation-specific facts. Production changes are introduced with migration checks, regression testing and rollback paths.

Evidence: Provider governance, evidence synthesis, citation validation, quality feedback and deployment rollback mechanisms are implemented across the platform.

Performance Engineering Active

Faster page delivery and hardware-aware AI execution

Performance work now spans both the web platform and local AI runtime. The CMS avoids unnecessary migration work and large Academy data loads on normal requests, while the local AI architecture continues to profile CPU, GPU, RAM and model-loading behaviour so execution can be matched to available hardware instead of assuming unlimited resources.

Evidence: CMS request-path optimisation and local resource-profiling work are both part of the current development stream.

EraNorth Platform Established

EraNorth and KEVOS AI roles clearly separated

EraNorth is the advisory, learning and transformation organisation. KEVOS AI is the AI product developed within that platform. Public navigation, Academy, release management and product pages now use that distinction consistently so consulting services, professional education and software are not presented as the same thing.

Evidence: Public brand architecture, KEVOS AI product routes, EraNorth Academy and release centre use the same product/company distinction.

Why KEVOS exists

The model is only one component of the system.

Professional AI becomes more useful when it can work with trusted knowledge, understand context, use the right tools, preserve evidence and operate inside clear permissions.

KEVOS AI is therefore being developed as an orchestration and intelligence environment rather than as a single chatbot or a single model. The architecture is intended to remain modular so models and tools can change without redefining the product.

Current public position: KEVOS AI is under active development. The page describes the development direction and implemented foundations, not a claim that every listed capability is production-ready.

Platform architecture

A modular capability stack.

Each layer is being developed so it can be improved independently while still operating through a common orchestrator.

Foundation built

Local & private AI core

Local model orchestration, selectable models and a private working environment designed to reduce unnecessary dependence on external services.

Foundation built

Knowledge & RAG

Document retrieval, hybrid search concepts, citations, deeper context expansion and multi-pass evidence gathering for large knowledge collections.

Active development

Research & verification

Iterative research workflows that search, read, refine, compare evidence and continue until the task reaches an appropriate confidence threshold.

Active development

Vision & document intelligence

Image understanding, document visual analysis, structured extraction and multimodal workflows for technical and professional material.

Active development

CAD & engineering tools

Controlled integration patterns for engineering files, technical libraries and CAD applications, including local SolidWorks-oriented workflows.

Active development

Memory & specialist agents

Hierarchical memory, expert roles, planning, reflection, knowledge graphs and agent hand-offs designed to support long-running professional work.

Engineering focus

Performance & model routing

GPU/CPU/RAM balancing, model selection, fallbacks, resource limits and performance profiling for smooth operation on practical hardware.

Future packaging

Portable deployment

Pack-and-go architecture, repeatable installation, upgrade paths and future one-click deployment for suitable computers or servers.

Governance principle

Security & human control

Permissions, safe tool boundaries, auditability, evidence, rollback, failure handling and human approval where actions carry meaningful consequence.

Intended specialist capability

KEVOS AI is being shaped to support expert engineering, project and program, research, knowledge and decision workflows without pretending one agent can replace every professional discipline.

Engineering

KEVOS Engineer

Technical knowledge, drawings, specifications, CAD context, design reasoning and engineering workflows.

Delivery

KEVOS Project

Scope, planning, governance, risk, decisions, reporting, benefits and delivery context.

Intelligence

Research · Vision · Knowledge

Evidence gathering, visual understanding, document intelligence, knowledge retrieval and verification.

Where KEVOS may create value

Start with a repeated business or engineering problem.

  • Private organisational knowledge assistants
  • Engineering document and specification intelligence
  • Project and program knowledge environments
  • RAG and evidence-linked research
  • Document extraction and comparison
  • Controlled agent and workflow automation
  • Manufacturing and operations intelligence
  • Local/private AI architecture for sensitive information

The preferred route is to validate the problem, information readiness, risk and measurable outcome before treating software as the answer.

Public development boundary

We will show progress, not expose the private knowledge system.

Public demonstrations will use synthetic, public or sanitised material. Internal data, private knowledge libraries, credentials and sensitive system details remain private.

Public

Selected architecture diagrams, milestones, benchmark summaries, demonstrations, screenshots using non-sensitive data, lessons learned and product-development notes.

Private

Internal learning and business plans, proprietary datasets, confidential documents, detailed security configuration, agent memory, credentials, source knowledge and client/employer material.

KEVOS Expression of Interest

Have a problem that may suit private or engineering-focused AI?

Share the problem, current workflow, information involved and outcome you would want to prove. EraNorth will use the enquiry to understand demand and suitable pilot opportunities.

Submit an Expression of Interest