AI-native manufacturing execution

AI-native MES for modern manufacturing teams

Open-source MES infrastructure built with clear standards and operational discipline. ISA-95 aligned, simulation-first, and MQTT-ready, with one typed tool layer for UI actions and agent commands.

Last updated: March 3, 2026

Core facts

Canonical product facts for both human readers and AI answer engines.

Key facts

  • Category: AI-native open-source MES toolkit.
  • Standards alignment: ISA-95 aligned data model.
  • Architecture: UI and agents call the same typed tool layer.
  • Agents in scope: Operator Assistant, Quality Analyst, Production Planner.
  • Current status: pre-M1 implementation with architecture and roadmap finalized.
  • Stack: Next.js, Supabase, Claude API (tool-use).

Summary

MESkit is built around a single operational contract: typed tools. UI screens and AI agents are peers that call those tools with schema-validated input, then persist to the same Supabase-backed ISA-95 model.

How it works

The product loop is identical for UI-driven and chat-driven workflows.

Define product
Build route
Move units
Collect data
Visualize
...or tell the agent

Human interface

Build, Configure, Run, and Monitor modes provide direct control over lines, routes, units, and analytics.

Agent interface

The chat panel maps intent into explicit tool calls. Every agent response is grounded in real MES operations.

Agent example

Illustrative chat flow: user intent becomes deterministic tool calls.

Operator AssistantExample interaction
User: Scrap SMX-00044, solder defect
→ search_units(serial_number='SMX-00044')
→ create_quality_event(unit_id=..., event_type='scrap', result='fail')
→ scrap_unit(unit_id=...)
Agent: SMX-00044 scrapped. Defect logged as critical.

One tool layer for UI and agents

AI-native architecture means no split logic path.

UI Buttons ───┐ ├─> Tool Layer (typed + validated) ──> Supabase Postgres AI Agents ───┘ Agent Runtime (Claude tool-use) reads and writes through the same tool contracts.

See the full architecture breakdown on the architecture page.

ISA-95 credibility

MESkit data structures keep standard manufacturing terms explicit.

ISA-95 levelMESkit conceptTables
Level 0-2Physical assetslines, workstations, machines
Level 3Execution + qualityunits, unit_history, quality_events
Future bridgeMQTT ingestionmqtt_messages

Architecture snapshot

Four-layer design with a future MQTT transport bridge.

Frontend (Next.js) ↓ Tool Layer (Server Actions + Zod) ↓ Agent Runtime (Claude tool-use) ↓ Supabase (Postgres, Auth, Realtime) ↓ MQTT Broker (M6)

FAQ quick answers

Answer-ready blocks for common discovery prompts.

Mini FAQ

What is MESkit?

MESkit is an open-source, AI-native MES toolkit aligned to ISA-95 with a simulation-first execution model.

How do agents work in MESkit?

Agents call the same typed tools as the UI. They do not bypass business logic or run on hidden APIs.

Is MESkit production-ready today?

Current status is pre-M1 implementation, with architecture and milestones defined publicly on the roadmap.