-
Close Search
How to validate, block, redact, and safely fail AI inputs and outputs before your helpful assistant becomes a compliance incident with a loading spinner.
A quick reference for the core ColdFusion AI concepts, when to use them, and what to watch for.
Part 1 of an 8 part series on how to integrate your ColdFusion application with Slack to create a two way chat portal.
Part 2 of an 8 part series of building a chat gateway between Slack and ColdFusion.
Nobody admires a well-designed database—until your chat system starts mixing up customer conversations. Let’s build a data model your future self won’t hate.
This article moves beyond simple, stateless ChatModel() calls and introduces Agent() as the layer that gives ColdFusion AI conversational memory. It explains how message windows, token windows, PERUSER, memory keys, and persistent stores help an assistant remember recent context without accidentally sharing one user’s conversation with another. It also separates short-term chat memory from durable user preferences, emphasizing that important preferences belong in your application data, not in a rolling memory window. The core lesson is that memory makes an AI assistant feel more coherent, but ColdFusion still controls identity, persistence, validation, privacy, and truth.
This article introduces MCP, or Model Context Protocol, as the next layer in building more capable ColdFusion AI applications. After using ChatModel() for stateless prompts, Agent() for memory, and CFC tools for local application capabilities, MCP provides a standardized way for AI workflows to connect with tools, prompts, and resources across system boundaries. The article explains the difference between MCP hosts, clients, and servers; when to use MCP instead of local CFC tools; how ColdFusion can consume external MCP servers or expose its own capabilities; and why security, authorization, observability, and careful tool design still matter. The core lesson is that MCP gives your AI assistant a passport, but ColdFusion still decides where it is allowed to go.
This article introduces RAG, or Retrieval-Augmented Generation, as the layer that lets ColdFusion AI answer questions using your own documents instead of relying only on model knowledge, memory, or guesses. It explains how ColdFusion can load documents, split them into chunks, create embeddings, store them in a vector store, retrieve relevant content, and use that context to generate grounded answers. The article also covers simpleRAG(), ask() versus chat(), chunking, minScore, maxResults, ingestion timing, persistent vector stores, permissions, stale content, and source quality. The core lesson is that RAG gives the robot an open-book test, but ColdFusion still decides which book it is allowed to read.
This article introduces CFC tools as the next layer in building useful AI features with ColdFusion. After covering stateless ChatModel() calls and memory-enabled Agent() conversations, we now give the assistant a controlled way to request real application capabilities through ColdFusion components. The article explains how to expose specific CFC methods as tools, write useful tool descriptions, inspect toolExecutionRequests, validate model-generated arguments, enforce authorization inside the tool, handle read versus write actions, log tool usage, and keep the application firmly in control. The core lesson is simple: tools let the AI ask for real data or actions, but ColdFusion validates, authorizes, executes, and decides what happens next.
This article introduces the simplest practical starting point for using AI in ColdFusion: ChatModel(). It walks through configuring a model, sending a prompt with .chat(), reading response.message, and safely displaying the result. Along the way, it explains provider settings, model names, API keys, temperature, max tokens, timeouts, prompt design, response validation, error handling, logging, and the tradeoffs between streaming and non-streaming responses. The central lesson is that ChatModel() is ideal for simple, stateless AI tasks like summarizing, rewriting, classifying, translating, and drafting, but it is only the front door. More advanced features like memory, tools, RAG, MCP, and guardrails require additional layers.
Artificial intelligence is quickly becoming part of everyday software development, but using it well requires more than adding a chatbot to an application. This article introduces AI fundamentals for ColdFusion developers, including large language models, prompts, tokens, context windows, temperature, Top-P, Top-K, statelessness, hallucinations, privacy concerns, and the role of application architecture. It also previews how ColdFusion 2025 Update 8’s native AI services can help developers move from simple chat model calls to more capable systems using memory, CFC tools, MCP, retrieval-augmented generation, and guardrails. The central theme is simple: the LLM is not your application. It is a reasoning engine your ColdFusion application supervises.
