Current status
Early-stage open-source project with a simulator-backed Process Sentinel vertical slice.
Public Roadmap and Milestones
The roadmap describes how OFI is turning an early-stage open-source product into a safer, more sustainable manufacturing intelligence ecosystem.
Near-term focus
Event model, drift detection, evidence timeline, governed recommendation workflow, public documentation, contributor onboarding, and the foundation for local-first site-specific onboarding.
Roadmap governance
Priorities should be shaped by manufacturing needs, contributor capacity, security review, validation-readiness concerns, and public-benefit impact.
Local-first AI and site-specific onboarding
Site-specific onboarding will help each participating facility map its assets, lines, process signals, quality events, procedures, terminology, and governance requirements into the Factory Intelligence Platform.
Future roadmap work includes a Site AI Package, RAG foundation, SLM/LLM training and evaluation workflows, model governance, validation-aware onboarding, and agent orchestration that can help adapt factory intelligence assistance to a facility's terminology, procedures, asset structure, and operational context while preserving human review, traceable evidence, and governance controls.
Any site-specific model training or adaptation work should be governed by explicit data-use agreements, privacy and security controls, human review, evaluation criteria, and adoption responsibility by the participating organization.
Planned onboarding scope
- Site, area, line, and asset mapping
- Manufacturing terminology and process context
- Data source inventory
- Historian, MES, QMS, CMMS, and ERP context mapping
- Quality and deviation workflow mapping
- Human review and approval path definition
- Risk, security, and validation-readiness considerations
- Contributor or operator training materials
Data and model governance guardrails
Public demos should use synthetic or permissioned data. Proprietary facility data should not be used for public model training without explicit authorization. Model outputs should be treated as decision support, not automatic decisions.
Foundation and MVP phases
Phase 0: Repository Foundation
Professional README, Codex instructions, architecture docs, testing docs, contribution docs, issue and PR templates, and CI baseline.
Phase 1: Simulator-First Foundation
Synthetic site, area, line, and asset model with seeded normal, drift, and excursion scenarios for local development.
Phase 2: Event Contracts + Ingestion
Unified event envelope, event validation, ingestion worker, event storage, and contract tests.
Phase 3: Process Sentinel MVP
Drift detection rules, evidence timeline, detection records, recommendation generation, and human-reviewed decision support language.
Phase 4: Governed Workflow
Recommendation review states, approval/rejection/defer workflow, audit events, RCA/CAPA draft output, and governed recommendations only.
Phase 5: Web Workbench
Dashboard, detection list, detection detail, evidence timeline, approval panel, and draft report view.
Phase 6: E2E Hardening
Playwright workflow, CI checks, documentation polish, demo scenario, contributor onboarding, and validation-aware documentation patterns.
Planned AI and site intelligence phases
These future roadmap workstreams are planned research and development areas for human-reviewed decision support, governed recommendations, evaluation workflows, and validation-aware documentation.
Future roadmap work
Local AI + Site Intelligence Foundation
- Plan a model-agnostic Local Model Gateway for local-first and explicitly configured remote-compatible providers
- Define SLM-first and LLM-fallback routing criteria around task type, context size, risk, latency, and confidence
- Set read-only and recommendation-only tool-call boundaries with usage logging for audit-friendly review
Future roadmap work
Site AI Package + RAG Foundation
- Create a Site AI Package outline for site profile, area/line/asset hierarchy, process context, and equipment metadata
- Map tag context, event contracts, prompt templates, model routing config, and tool registry config
- Use RAG before fine-tuning with document ingestion, retrieval indexes, cited evidence, and Factory Memory integration
Future roadmap work
SLM/LLM Training + Evaluation Pipeline
- Plan curated, approved, traceable, redacted, and sanitized site-specific datasets
- Prefer LoRA/PEFT adapters over full fine-tuning for site adaptation where training is appropriate
- Build evaluation workflows for grounded answers, retrieval quality, refusal behavior, hallucination checks, cited evidence, task accuracy, latency, and resource use
Future roadmap work
Model Governance + Validation-Aware Onboarding
- Define model, prompt, tool, and dataset registries with model cards, evaluation records, approval history, audit events, and rollback support
- Document intended use, GxP impact assessment, risk classification, data flows, inventories, and acceptance criteria
- Publish validation-ready documentation patterns for future site validation work, not automatic regulatory compliance
Future roadmap work
Agent Orchestration
- Plan agents that call the Local Model Gateway and Tool Registry instead of direct model providers
- Explore evidence summarization, RCA/CAPA draft language, investigation path suggestions, similar-event lookup, and drift signal explanations
- Keep agent outputs limited to human-reviewed decision support and governed recommendations
AI safety boundary
Model outputs should remain advisory, evidence-cited, and human-reviewed. Future agent and model workflows are not intended for autonomous factory control, automatic quality disposition, or regulatory approval.
Success measures
Roadmap progress should be measured by public artifacts, real community engagement, development maturity, security practices, documentation quality, and external technical input.
View repository roadmapReview Security Approach- Discovery interviews completed
- Public artifacts published
- Contributors engaged
- Issues opened and closed
- Pull requests reviewed
- Workshops or community events
- Security practices implemented
- Documentation completeness
- Test coverage and testing maturity
- External reviewers or advisors engaged