Grab a chair before proceeding.
You’re about to access the full Coreflux platform. Some people report weakness in their knees upon first contact.
You’ve been warned.
Features
%20(1).png)
An episode of "Everything Reminds Me of Coreflux"
A good tour guide in Porto doesn't read the city from a leaflet. They know it the way you know your own kitchen. The history, the shortcuts, what's busy on a Tuesday morning, what's quiet on a Sunday evening, which café still has the original tile work, which bridge has the best view at golden hour. They give you the right answer, in the right context, in that moment.
Most AI tools work the other way around. Loud about everything, useless about your specific situation. They've read every textbook ever published and not a single line of your broker's actual data.
The Coreflux AI Assistant is the guide. Built for your project, not the world.
If you've ever asked a generic AI tool about your industrial system, you've seen the moment. The model answers with confidence. It cites a topic name that doesn't exist. It references a protocol your broker doesn't use. It invents a route you never built. The answer sounds correct. It's wrong.
This isn't a bug. It's a feature of how generic AI works. The model is trained on the internet. The internet contains a lot of MQTT documentation, a lot of Python tutorials, a lot of Slack threads about industrial protocols. It does not contain your factory. It does not know that machine 3 lost calibration last Thursday. It cannot tell you which topic stopped publishing at 4am. It has never seen your dashboards.
The traditional answer is to keep AI out of the loop. Use it for documentation lookup, use it for code generation, but never ask it about your data. Most teams have an unwritten rule about this.
We built the AI Assistant because that rule was costing more than it saved.
.jpeg)
The AI Assistant lives in the Coreflux Hub. You open it from the dock, like any other app. It's a chat panel. You type in plain language. It reads your actual broker as context: real topics, real Routes, real payloads, real history.
It has two modes by design:
Insight mode.
Read-only. The Assistant can list topics, read payloads, explore Routes, render tables and charts. It can answer the question "which topics haven't received a message in the last hour" by actually checking. It can't change anything. Safe for operators, safe for the curious, safe for the engineer tuning a prompt before promoting it.
.png)
Full access, subject to your Hub permissions. The Assistant can create and update Routes, Actions, Models, and dashboards. You say "build a dashboard with temperature, pressure, and uptime for line 1" and it generates the panel and deploys it. Everything it deploys goes through the same validation as a human-deployed change.
Two doors. Same Assistant. Different posture depending on what you trust it to do.
%20(1).png)
Some teams run on OpenAI. Some on Anthropic. Some on Mistral. Some have signed agreements that data cannot leave the building under any circumstances. The Assistant is designed for all of them.
Cloud. OpenAI, Anthropic, Mistral. API key, stored encrypted on the broker. The Assistant calls out, gets a response, brings it back into the Hub. Standard.
Local. Ollama on the same host or your LAN. No API key, no outbound traffic, no contract with a foreign company. The Assistant uses your local model. Same modes, same UI, same workflow. The data never leaves.
Same panel, different LLM. The decision is yours, per project, per site, per regulation.
.png)
The AI Assistant is the visible face of a deeper integration.
AI Routes. The same engine is available as a building block inside LoT. Define an AI agent as a Route, call it from any Action. Wire it to MCP Routes so the agent can take action on external systems. The AI isn't a separate product. It's a directive in the same language as the rest of your broker logic.
MCP for development. Coreflux exposes its documentation and project rules through the Model Context Protocol. AI assistants in Cursor, Claude Desktop, and VS Code can look up real LoT syntax instead of guessing. Combined with an AGENTS.md project file, the assistant learns your team's conventions. The developer in the editor and the operator in the Hub are using the same intelligence, calibrated for the same project.
It's still in beta. We're not pretending otherwise. It gets better every week.
(2)%20(1).png)
In the episode, Tiago Abreu, our Product Manager, walks through the Assistant from his desk. Except he isn't alone. Popping up next to him is an unexpected someone: João Barroso's mother.
It's a joke. It's also the point.
She's been answering Barroso's questions a lot longer than any AI has. She knew his nicknames, his test schedules, his cartoon preferences, his bad ideas about which bands were cool in 2004. She had context. She still does. The AI Assistant is the same kind of relationship, applied to a broker.
A mother doesn't need general knowledge of children. She needs specific knowledge of one. The AI Assistant doesn't need general knowledge of industrial systems. It needs specific knowledge of yours.
.png)
General knowledge is cheap now. Every model has it. Every search bar returns it. The differentiator isn't what an AI knows about the world. It's what an AI knows about your world.
That's why the Assistant lives inside the broker, reads your topics, holds your project as context, deploys into your environment, and refuses to guess when it doesn't know.
The internet is the manual. The Assistant is the guide.
.png)
Tiago Abreu, our Product Manager walks through the AI Assistant in the fifth episode of Everything Reminds Me of Coreflux. João Barroso starts the conversation, walking through the city with André, asking too many questions, until André gives up and freezes him mid-sentence. A mother appears. You will understand why.
You should watch it and share it with your own mother.
Try Coreflux for free 👉 https://www.coreflux.org/
%20(1).png)
