# Nodeblue

> Nodeblue is an independent applied research lab studying how machines can understand, remember, and act inside the operations the real world runs on. Its first system, Nexus, is operational intelligence for industrial environments, running live in production today.

Key facts about Nexus:

- Nexus reads a plant's PLC and control logic, SCADA and supervisory data, engineering documentation, and systems of record (MES, ERP, CMMS), and reasons over them as one connected picture.
- Questions are asked in plain language, but the answer comes from a deterministic engine underneath, not from a model reading a raw export: answers are exact, cited to the specific rung of control logic, and reproducible on every run. The model on top is replaceable, and Nexus improves as those models improve.
- It deploys on premise and is air gap capable by default, the shape defense, utilities, and pharmaceutical operations require.
- It is cross vendor by design: Studio 5000 and RSLogix 5000, Siemens TIA Portal and STEP 7, Ignition, and the CODESYS family directly, plus a PLCopen parser that reaches 500+ OEM brands.
- It is validated on real production data: 4,386 real PLC project files parsed with zero parser errors, and 137 of 137 AI-written program documents graded rung-accurate by three independent Controls Engineers.
- The early foundational connectors are open source (MIT) on GitHub (studio5000-mcp-server, ignition-mcp-server, bridge-mcp-server): public and auditable, though no longer actively maintained now that the tooling built today stays in-house.

Nodeblue is based in St. Petersburg, Florida and works with industrial operations across the United States. Contact: contact@nodeblue.ai or https://www.nodeblue.ai/contact.

## Systems

- [Nexus](https://www.nodeblue.ai/nexus): Operational intelligence for industrial environments. Live in production today; hosted demo at nexus.nodeblue.ai is gated while onboarding is done by hand.
- [Systems overview](https://www.nodeblue.ai/systems): Nexus today, with Forge (safe execution) and Atlas (a living model of an operation) in development.

## Research

- [Research directions](https://www.nodeblue.ai/research): The four directions the lab pursues: Operational Intelligence, Knowledge and Memory, Human and AI Collaboration, and Autonomy and Execution.

## Company

- [About](https://www.nodeblue.ai/about): What the lab is and how it operates.
- [Careers](https://www.nodeblue.ai/careers): Open roles.
- [Partners](https://www.nodeblue.ai/partners): Working with Nodeblue.
- [Security](https://www.nodeblue.ai/security): Security posture, including on premise and air gapped deployment.
- [Contact](https://www.nodeblue.ai/contact): Request Nexus access or book an intro walkthrough.
- [FAQ](https://www.nodeblue.ai/faq): Common questions about the lab and Nexus.

## Customer Stories

De-identified records of Nexus running against real production control systems. Customer, site, controller, network, and tag names are synthetic and hardware and third-party platform names are generalized; the reasoning, citations, and measured outcomes are unchanged.

- [First Pass Yield Up 6.76 Points After Jams Stopped Being Found on Foot](https://www.nodeblue.ai/case-studies/chute-jam-board-first-pass-yield): First pass yield up 6.76 points across the four weeks after go-live, defect rates down on both categories, and a jam record the site had never had before. (Read-only continuous collection, in production 24/7)
- [A Merge Releasing in Slugs: Eliminating the Code Before Touching the Line](https://www.nodeblue.ai/case-studies/eliminating-the-code-on-a-merge-releasing-in-slugs): Deployed code eliminated as a cause in one session, the fault space reduced to three testable mechanisms with a fix path for each, two of them later confirmed and corrected on the line. Inbound starvation fell 9.75 points across the rollout. (Offline export byte-diff, plus live read-only connection)
- [Two Stacked Root Causes Behind a Five-Hour Palletizer Stop](https://www.nodeblue.ai/case-studies/two-root-causes-behind-a-palletizer-stop): Cell placing to pallets again, verified live: assignments accepted, real destinations assigned, pallet counts climbing. (Live read-only connection, plus offline export)
- [Reproducing a Cloud Starvation Metric From 18 Controller Tags](https://www.nodeblue.ai/case-studies/starvation-monitoring-without-the-cloud): The backend's own starvation math reproduced locally from live controller tags, with a windowed alert that reports insufficient coverage rather than firing on a data gap. Shipped the same day and running continuously since. (Live read path in production 24/7, plus offline export)
- [Scoping a Ghost Fault Down From a Conveyor to One Zone](https://www.nodeblue.ai/case-studies/clearing-a-ghost-fault-across-three-layers): The dead reference scoped down from a whole conveyor to a single zone, verified against the control logic before anything was deleted. (Offline export, plus visualization project files)
- [Decommissioned in the Host, Still Running on the Floor](https://www.nodeblue.ai/case-studies/decommissioned-but-still-running): The whole lane reduced to two enable bits, verified live, blast radius vetted across the area, and every listed unit confirmed not-enabled after the change. (Live read-only connection, plus offline export)
- [OEE Was Counting Four Lanes That Were Switched Off](https://www.nodeblue.ai/case-studies/disabling-oee-on-lines-that-were-already-off): Four lanes cleared from the metric with no effect on product motion, engineer-verified, and the follow-on question answered live when the goal changed to physically stopping the lines. (Offline export, then live read-only connection)
- [Four Faults on the Screen, One in the Controller](https://www.nodeblue.ai/case-studies/four-faults-on-the-screen-one-in-the-controller): One genuine motor connection fault to chase physically, and a supervisory-layer mismatch on the other three, each backed by cited live reads. (Live read-only connection, plus offline export)
- [A Cross-Controller Feed Permit: 8 Minutes by Hand, 20 Seconds Traced](https://www.nodeblue.ai/case-studies/tracing-a-cross-controller-handshake): The permit output, its four enabling conditions, and the zone handshake behind it, each cited to the rung, against a search the engineer had not finished by hand. (Offline controller export)
- [Reading an OEE State Machine Out of a Vendor Instruction Body](https://www.nodeblue.ai/case-studies/decoding-an-oee-state-machine): The state machine and its alarm-code threshold read out of the instruction bodies, three unrelated code defects surfaced, and a one-page operator reference produced. (Offline controller export)

## Writing

- [Intelligence Inside the Loop, Not Beside It](https://www.nodeblue.ai/blog/intelligence-inside-the-loop): Software has gotten very good at observing the world and very slow at understanding it. The result is operations that are heavily instrumented and barely comprehended. We think the next useful step is not another dashboard.
- [What We Learned Pointing Nexus at a Real Site](https://www.nodeblue.ai/blog/what-we-learned-putting-nexus-on-a-floor): Notes from running our operational intelligence system against real production controllers instead of a simulator. The surprises were not about the model. They were about which artifact you read, what a correct answer is worth when nobody can use it, and where a system has to hand back to a person.
- [Building Agents That Act: What the Demo Does Not Show](https://www.nodeblue.ai/blog/building-reliable-agents-that-act): The gap between a working prototype and a reliable agent running in a real operation is wider than most teams expect. Here is what breaks, and how we design around it when a wrong action has a cost.
- [Evaluating Systems That Act in the Real World](https://www.nodeblue.ai/blog/evaluating-systems-that-act): A benchmark score tells you how a model does on a fixed test. It tells you almost nothing about how a system will behave inside a live operation. Here is how we think about evaluation when the output is an action, not an answer.
- [Why Dashboards Stopped Being Enough](https://www.nodeblue.ai/blog/why-dashboards-stopped-being-enough): The dashboard was a genuine advance. It is also where a lot of operations got stuck. Observation is not understanding, and adding another chart does not close the gap between them.
- [How Operational Knowledge Decays](https://www.nodeblue.ai/blog/how-operational-knowledge-decays): Most of what an organization understands about itself lives in people, and it leaves when they do. Documentation goes stale the day it is written. We study what it would take for a system to hold that knowledge and keep it current.
- [Judgment and Recall: Dividing the Work Between People and AI](https://www.nodeblue.ai/blog/judgment-and-recall): The useful question is rarely whether a system can do a task. It is which parts of skilled work should be handed to a system, which must stay with a person, and how the seam between them earns trust.
- [The Gap Between Advising and Acting](https://www.nodeblue.ai/blog/the-gap-between-advising-and-acting): Describing a situation is one thing. Acting on it safely is another. The distance between the two is where most applied AI quietly stops, because in the physical world a wrong action has a real cost.
- [Forge: Turning Engineering Intent Into Working Systems](https://www.nodeblue.ai/blog/forge-engineering-intent-into-systems): A look at the research behind Forge, our effort to let teams describe what a system should do and get a trustworthy first version of the logic and configuration built with them, rather than from scratch.
- [AI for PLC Troubleshooting: What Works on a Live Floor](https://www.nodeblue.ai/blog/ai-for-plc-troubleshooting): Most AI troubleshooting tools were designed for a demo, not a down line. What fault diagnosis actually requires, where AI genuinely helps, and the failure modes that only show up against a real controller.
- [Generating PLC Documentation with AI, Graded by Controls Engineers](https://www.nodeblue.ai/blog/generating-plc-documentation-with-ai): Machine-written PLC documentation is only useful if every statement traces to the rung that makes it true. How we generate program docs, how 137 of them survived grading by three independent Controls Engineers, and what generation still cannot capture.
- [Can AI Read PLC Code? An Honest Assessment](https://www.nodeblue.ai/blog/can-ai-read-plc-code): Large language models can discuss ladder logic convincingly and still get the rung wrong. What it actually takes for AI to read Studio 5000, TIA Portal, CODESYS, and the rest of a plant's control logic correctly.

## Optional

- [Privacy](https://www.nodeblue.ai/privacy)
- [Terms](https://www.nodeblue.ai/terms)