<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Vijay Malviya on Medium]]></title>
        <description><![CDATA[Stories by Vijay Malviya on Medium]]></description>
        <link>https://medium.com/@vijaymalviya320?source=rss-e58b864b267------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*hzejDpC4v7JvcLtHHBa51w.jpeg</url>
            <title>Stories by Vijay Malviya on Medium</title>
            <link>https://medium.com/@vijaymalviya320?source=rss-e58b864b267------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 23:34:17 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@vijaymalviya320/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Autonomous Multi-Agent Orchestration in Mendix: Definition, Patterns, and a Real Build]]></title>
            <link>https://medium.com/@vijaymalviya320/autonomous-multi-agent-orchestration-in-mendix-definition-patterns-and-a-real-build-e7f3e2229763?source=rss-e58b864b267------2</link>
            <guid isPermaLink="false">https://medium.com/p/e7f3e2229763</guid>
            <dc:creator><![CDATA[Vijay Malviya]]></dc:creator>
            <pubDate>Tue, 04 Aug 2026 04:59:52 GMT</pubDate>
            <atom:updated>2026-08-04T04:59:52.004Z</atom:updated>
            <content:encoded><![CDATA[<p><em>How low-code and agentic AI come together to coordinate specialized agents across a business process .</em></p><h3>Introduction</h3><p>For years, “automation” in Mendix meant microflows, nanoflows, and BPMN-style workflows: deterministic logic that executes exactly the steps a developer modeled. Agentic AI changes that equation. Instead of hard-coding every branch, you can hand a goal to an LLM-backed agent and let it reason, call tools, and decide the next step.</p><p>The next evolution beyond a single chatbot-style agent is <strong>multi-agent orchestration</strong>: several specialized agents, each good at one narrow job, coordinated by an orchestration layer that routes work, checks quality, and keeps a human in the loop when it matters. Mendix now ships native building blocks for exactly this — inside Studio Pro, in the GenAI/Agents Kit, and in the broader Process Orchestration layer.</p><p>This article covers the definition, the architecture, the patterns Mendix documents, a step-by-step “how we do it,” and a worked example with diagrams.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*FLMCM2W92r5gixlH" /></figure><h3>1. Definition</h3><p><strong>Autonomous multi-agent orchestration</strong> is an architecture in which multiple independent AI agents — each with its own prompt, tools, and scope of responsibility — are coordinated by an orchestration layer to jointly complete a task that would be too broad, too ambiguous, or too error-prone for a single agent to handle reliably.</p><p>In Mendix terms, three concepts combine to make this possible:</p><ul><li><strong>GenAI Agent</strong> — an autonomous computational unit that reasons over a prompt, calls tools (function calling), and optionally retrieves grounding data from a knowledge base, then decides what action to take next.</li><li><strong>Multi-Agent System</strong> — several such agents wired together with a defined interaction pattern (chaining, routing, evaluation, etc.), so that specialized task allocation and hand-offs happen deliberately rather than inside one giant prompt.</li><li><strong>Process Orchestration</strong> — the governance and coordination layer (what Mendix calls Business Orchestration and Automation, or BOAT) that anchors agents inside real business processes, adds human-in-the-loop checkpoints, and keeps every agent action traceable and auditable.</li></ul><p>Put simply: a single agent <em>answers</em>; a multi-agent system <em>collaborates</em>; process orchestration <em>governs</em> the collaboration so it’s safe to run in production.</p><h3>2. Why Multi-Agent Instead of One Big Agent?</h3><p>A single, monolithic agent tends to become a “God prompt” — trying to do intake, validation, retrieval, decision-making, and formatting all at once. That’s brittle: one bad instruction and the whole chain misbehaves, and it’s hard to test or reason about.</p><p>Splitting responsibility into specialized agents gives you:</p><ul><li><strong>Separation of concerns</strong> — one agent classifies, another retrieves, another drafts, another checks quality.</li><li><strong>Independent iteration</strong> — you can swap the model or tune the prompt for one agent without touching the others.</li><li><strong>Better traceability</strong> — each hand-off is a discrete, loggable step, which matters for audit and compliance.</li><li><strong>Graceful failure</strong> — a “gatekeeper” or evaluator agent can catch bad outputs before they propagate.</li></ul><h3>3. Core Building Blocks in Mendix</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4k-V9S1hAJE0sf9oXLGAog.jpeg" /></figure><h3>4. The Orchestration Patterns Mendix Supports</h3><p>Mendix’s GenAI documentation defines four common multi-agent patterns. Understanding these is the real “definition” of orchestration flow — they are the reusable shapes you compose your solution from.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/921/0*jymsqeXjne9W11FZ.png" /></figure><h4>4.1 Prompt Chaining</h4><p>A linear sequence of LLM calls, where each step has its own system prompt and the output of one becomes the input of the next. Different steps can use different models.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/983/1*F86ybYm7Mcgwqxy0fG6XdA.png" /></figure><h4>4.2 Prompt Chaining with Gatekeeper</h4><p>Same idea, but a gatekeeper agent sits in the flow and decides whether to continue the happy path or break out into an exception/unhappy flow.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/933/1*AK_-hk7RngwXzGxn1UJA4Q.png" /></figure><h4>4.3 Evaluator–Optimizer (LLM-as-a-Judge)</h4><p>One agent generates content; another evaluates it against criteria and sends feedback back. The loop repeats until the output passes, or a max-attempt limit is hit.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/921/1*r-ZKwn9_W5EgsXoEl_jWdA.png" /></figure><h4>4.4 Routing</h4><p>A router agent classifies the incoming request and hands off full responsibility to the best-suited specialist agent, which then completes the task with no further router involvement.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/980/1*IvLEWOwDPVRhCz2aTBBvqQ.png" /></figure><p>These four patterns are composable — a production flow often nests a router at the top, with an evaluator-optimizer loop inside one of the routed branches, all wrapped by a workflow-level human approval step.</p><h3>5. How We Do It: Building a Multi-Agent Orchestration Flow in Mendix</h3><p>Here’s the practical build sequence we follow.</p><h4>Step 1 — Start from the right foundation</h4><p>Add <strong>Agent Commons</strong> to your app (or start from the Agent Builder Starter App in the Marketplace). This gives you the domain model and activities for agent conversations, tool calls, and context/session handling.</p><h4>Step 2 — Define each specialist agent</h4><p>For every distinct responsibility (classify, retrieve, draft, validate, summarize), create a separate agent in the <strong>Agent Editor</strong>:</p><ul><li>Write a narrow, single-purpose system prompt.</li><li>Pick the model per agent (a cheap/fast model for classification, a stronger model for drafting).</li><li>Attach a knowledge base if the agent needs grounding (RAG via a vector database).</li><li>Register the tools (functions) it’s allowed to call — each tool maps to a microflow or REST action.</li></ul><h4>Step 3 — Choose and wire the orchestration pattern</h4><p>Decide, per use case, which of the four patterns fits:</p><ul><li>Simple linear enrichment → <strong>Prompt Chaining</strong></li><li>Need an exception/compliance check mid-flow → <strong>Chaining with Gatekeeper</strong></li><li>Quality matters more than speed (e.g., legal or customer-facing text) → <strong>Evaluator–Optimizer</strong></li><li>Requests vary widely in type → <strong>Routing</strong></li></ul><p>Wire the hand-offs in a microflow/nanoflow, or, for longer-running, human-inclusive processes, model them as steps in a <strong>Mendix Workflow</strong> using the <strong>AI Agent Task</strong> activity.</p><h4>Step 4 — Let agents reach outside the app safely</h4><p>Use the <strong>MCP Client</strong> module when an agent needs to call an external system (a database, another SaaS tool, a partner API) as a governed tool call. Use the <strong>MCP Server</strong> module if you want <em>other</em> agents or LLM clients (including Maia or third-party copilots) to call your Mendix logic as a tool.</p><h4>Step 5 — Put it inside a governed business process</h4><p>Anchor the agent flow inside <strong>Process Orchestration</strong>. This is what turns “a clever prompt chain” into something safe to run in production:</p><ul><li>Human-in-the-loop checkpoints for high-stakes decisions (approvals, exceptions).</li><li>Full traceability of every agent action and hand-off.</li><li>Policy control over what each agent is permitted to do.</li></ul><h4>Step 6 — Test, observe, iterate</h4><p>Use the <strong>GenAI Showcase App</strong> as a reference implementation of each pattern. Monitor agent runs the same way you’d monitor any Mendix runtime — logs, metrics, and the workflow admin pages — and refine prompts/tools per agent independently.</p><h3>7. Worked Example: Knowledge Assistant</h3><p>If you work inside an Mendix Team— a central team or Center of Excellence pumping out many Mendix apps across product teams — the recurring pain isn’t claims or approvals, it’s <em>tribal knowledge</em>. Developers constantly ask things like: “Is there already a reusable module for OTP login?”, “What’s our naming convention for microflows?”, “Which team owns the SSO connector?”, “Why did we choose Workflow over a state machine for onboarding?”</p><p>A single chatbot over one wiki usually fails here because the knowledge is scattered across very different systems with very different shapes: a component catalog, coding-standards docs, architecture decision records, and Git history. That’s a natural fit for <strong>Routing</strong> into specialist agents, each grounded in one knowledge source.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/985/1*fcQN92gauCDL6B3ctb4l7A.png" /></figure><h3>8. Closing Thoughts</h3><p>Multi-agent orchestration in Mendix isn’t about replacing your process models with free-roaming AI — it’s about giving well-scoped agents clear jobs, wiring them together with a documented pattern (chaining, gatekeeper, evaluator-optimizer, or routing), and anchoring the whole thing inside a governed, auditable business process. That combination — low-code speed plus process-orchestration discipline — is what makes “autonomous” agents safe enough to actually ship.</p><p>If you’re starting out, the fastest path is: clone the <strong>GenAI Showcase App</strong>, pick the pattern closest to your use case, and swap in your own agents one at a time.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e7f3e2229763" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Mendix 11.12]]></title>
            <link>https://medium.com/@vijaymalviya320/mendix-11-12-21b9594789c9?source=rss-e58b864b267------2</link>
            <guid isPermaLink="false">https://medium.com/p/21b9594789c9</guid>
            <dc:creator><![CDATA[Vijay Malviya]]></dc:creator>
            <pubDate>Tue, 14 Jul 2026 13:51:04 GMT</pubDate>
            <atom:updated>2026-07-14T13:51:04.216Z</atom:updated>
            <content:encoded><![CDATA[<h3>Mendix 11.12 LTS Is Here — And Maia’s “Bring Your Own LLM” Is the Feature Enterprises Have Been Waiting For</h3><p><em>Why the first Long-Term Support release since Mendix 10.24 matters, and why letting Maia run on your own model changes the AI conversation in low-code.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/542/0*kAdNw1-eC62SS8N7" /></figure><p>Mendix just shipped <strong>11.12</strong>, and it’s not just another release — it’s the first <strong>Long-Term Support (LTS)</strong> version since Mendix 10.24. If you’ve been riding the Mendix 11 wave since it launched, or waiting for the “safe” moment to jump from Mendix 10, this is that moment. LTS releases are built to be the stable, supported foundation teams can build on for years, and 11.12 bundles nearly two years of Mendix 11 innovation — agentic AI, enterprise governance, mobile improvements, and a dramatically faster Studio Pro — into one production-ready package.</p><p>But if you ask me what the single most important line item in this release is, it’s not the performance boost or the governance tooling. It’s this:</p><p><strong>Maia can now run on your own LLM.</strong></p><p>Let’s talk about why that matters.</p><h3>A Quick Refresher: What Is Maia?</h3><p>Maia is Mendix’s built-in AI assistant, woven directly into the Studio Pro development experience. It’s grown a lot since it started as a documentation chatbot:</p><ul><li><strong>Maia Plan</strong> turns rough requirements into a structured backlog.</li><li><strong>Maia Make</strong> takes that plan and scaffolds actual application artifacts — data models, pages, microflows — directly inside Studio Pro.</li><li><strong>Maia for OQL</strong>, <strong>Maia Chat</strong>, and workflow-specific assistants round out a toolkit that now touches nearly every phase of app delivery.</li></ul><p>By default, all of this runs on Mendix’s own hosted AI service. That’s convenient — zero setup, no infrastructure to manage — but for a lot of enterprises, “convenient” isn’t the only requirement on the checklist.</p><h3>The Problem “Default Maia” Didn’t Solve</h3><p>If you work in a regulated industry, or your organization has strict data residency, model governance, or vendor policies, handing your prompts and context to a platform-managed AI service can be a hard no — regardless of how good the privacy policy is. Some common blockers I’ve heard from architects and platform owners:</p><ul><li><strong>“We already have an enterprise agreement with a specific LLM vendor, and it has to stay that way.”</strong></li><li><strong>“Our compliance team needs the model endpoint to sit inside our own cloud tenant.”</strong></li><li><strong>“We want the flexibility to swap models as the landscape changes, without waiting on a platform vendor’s roadmap.”</strong></li><li><strong>“We need uninterrupted access to AI assistance even if the platform’s shared service has an outage or rate limit.”</strong></li></ul><p>Until now, these were reasonable objections to leaning on Maia at scale. With 11.12, Mendix addresses them head-on.</p><h3>Enter: Custom AI Provider for Maia (a.k.a. Bring Your Own LLM)</h3><p>Starting in Studio Pro 11.12, you can configure Maia to use <strong>your own LLM provider</strong> instead of the default Mendix platform AI service. This is currently shipping as a <strong>beta feature</strong>, and it’s enabled per-project through Studio Pro’s Preferences.</p><p>Here’s the shape of it:</p><h3>How to turn it on</h3><p>Open <strong>Studio Pro Preferences</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/843/1*hbxaGQlTqAAp5yb3NdgnBA.png" /></figure><p>Go to the <strong>New Features</strong> tab.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Hgv_zAxPo_X13eHcPcpzug.png" /></figure><p>Enable <strong>“Enable configuring a custom AI provider for Maia.”</strong></p><p>A new <strong>Settings</strong> icon appears in the top-right corner of the Maia pane.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/790/1*tmCy-DnTjybzYDT43pHK5w.png" /></figure><p>Click it, and you’re into provider configuration.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/781/1*9Yv34iI73-2KVslqGRdFag.png" /></figure><p>Configuration is <strong>per-project and local to your workstation</strong> — it isn’t committed to version control, so each developer or each app can point at a different provider without stepping on anyone else’s setup.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/762/1*IbNYxa0OW9gZ3w3xncFX8g.png" /></figure><h3>The two supported paths</h3><p><strong>Amazon Bedrock</strong></p><ul><li>Select Amazon Bedrock as the LLM Provider.</li><li>Create an API key in AWS, then supply the Base URL and API key.</li><li>Enter the Model ID (found in your AWS Bedrock console).</li><li>Mendix’s own recommendation here is worth noting: <strong>Claude Sonnet for optimal Maia performance</strong> — a signal of which models Mendix has actually validated Maia’s prompting and tool-use patterns against.</li></ul><p><strong>OpenAI-Compatible Endpoints</strong></p><ul><li>Select OpenAI Compatible as the provider.</li><li>Enter your Base URL and Model ID.</li><li>API key is optional, depending on your endpoint.</li></ul><p>Studio Pro securely stores credentials across sessions, so this isn’t a “re-enter your key every morning” experience.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/781/1*9Yv34iI73-2KVslqGRdFag.png" /></figure><h3>Why This Is Bigger Than a Settings Toggle</h3><p>It’s tempting to file this under “nice enterprise checkbox,” but I think it reframes the whole low-code + AI conversation in a few concrete ways:</p><p><strong>1. Governance without giving up productivity.</strong> Enterprises no longer have to choose between “use Maia’s agentic development capabilities” and “keep model traffic inside our approved vendor and infrastructure boundary.” You can have both.</p><p><strong>2. Model choice becomes a platform decision, not a vendor lock-in.</strong> As new frontier models ship, teams aren’t stuck waiting for Mendix to bake support in — they can point Maia at whatever Bedrock or OpenAI-compatible model fits their needs today, and change their mind next quarter.</p><p><strong>3. Resilience.</strong> Mendix explicitly frames this as a way to get <strong>uninterrupted access to Maia capabilities</strong> — useful if your organization needs guaranteed availability independent of a shared multi-tenant service.</p><p><strong>4. It complements — not replaces — BYO Agent.</strong> If you want to go the other direction and bring <em>external</em> AI agents or MCP-based tools directly into Studio Pro, that’s what the <strong>Studio Pro MCP Server’s “Bring Your Own Agent” (BYO Agent)</strong> capability is for. Between BYO LLM (Maia runs on your model) and BYO Agent (your agents drive Studio Pro), Mendix is building a genuinely two-way AI integration story rather than a single walled garden.</p><h3>A Word on Data Handling</h3><p>Regardless of which provider setting you choose, it’s worth remembering how Maia treats your data by default: it doesn’t use project, customer, company, or user-entered data to train its models, and it runs on pre-trained, off-the-shelf LLMs. If you disable Studio Pro’s usage-data setting, prompts aren’t stored for analytics either — though that setting doesn’t affect what’s sent to whichever third-party provider you’ve configured as your custom AI provider. Worth a conversation with your security team before flipping the switch in a regulated environment.</p><h3>Should You Turn It On?</h3><p>If your organization has no particular data residency or vendor constraints, the default Mendix-hosted Maia experience is still the path of least resistance — it requires zero configuration and Mendix explicitly recommends it for most teams. But if you’ve been sitting on the sidelines of Maia adoption because of governance concerns, <strong>11.12 LTS removes that excuse</strong>. Combined with the long-term support promise of this release, it’s a legitimate moment to pilot Maia at scale, on your own terms.</p><h3>The Bigger Picture</h3><p>Mendix 11.12 LTS is a statement release: it says “everything we’ve built in agentic AI over the last two years is now stable enough to bet your production roadmap on.” The custom AI provider feature for Maia is the clearest evidence of that maturity — it’s the kind of feature that only gets built once a platform vendor is confident enough in its AI layer to let enterprises point it at whatever model <em>they</em> trust.</p><p>If you’re evaluating the jump to 11.12, don’t just look at the performance numbers or the governance features. Go configure a custom AI provider for Maia in a sandbox project, and see what it feels like to have enterprise-grade AI assistance running entirely within your own model boundary. That’s the real headline of this release.</p><p><em>Currently a beta feature in Studio Pro 11.12 and above — expect the configuration experience to keep evolving as Mendix moves it toward general availability.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=21b9594789c9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Building an AI-Powered Natural Language Query Engine for Mendix Using LLM and OQL]]></title>
            <link>https://medium.com/@vijaymalviya320/building-an-ai-powered-natural-language-query-engine-for-mendix-using-llm-and-oql-a3d059091894?source=rss-e58b864b267------2</link>
            <guid isPermaLink="false">https://medium.com/p/a3d059091894</guid>
            <dc:creator><![CDATA[Vijay Malviya]]></dc:creator>
            <pubDate>Sun, 14 Jun 2026 11:27:23 GMT</pubDate>
            <atom:updated>2026-06-14T11:27:23.604Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><p>One of the most common requirements in enterprise applications is allowing users to ask questions in natural language and receive answers directly from application data.</p><p>Traditionally, users need reports, dashboards, filters, or predefined search screens to retrieve information. However, with Large Language Models (LLMs), we can build intelligent chatbots that understand user questions, generate OQL dynamically, fetch data from the Mendix database, and return human-readable answers.</p><p>In this article, I will explain how I built a generic AI-powered query engine for Mendix that enables users to ask questions such as:</p><ul><li>Show all customers from Pune</li><li>How many orders were created this month?</li><li>List employees working in the IT department</li><li>What is the total revenue generated this quarter?</li></ul><p>without requiring users to know anything about entities, associations, or OQL.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/796/1*fdf6wpucYsLGdziV62Eh5g.jpeg" /></figure><h3>Architecture Overview</h3><p>The solution follows a multi-stage AI workflow:</p><ol><li>User asks a question.</li><li>LLM extracts important keywords.</li><li>Java Action identifies matching entities in the Mendix domain model.</li><li>Entity metadata is sent back to the LLM.</li><li>LLM generates a valid OQL query.</li><li>Java Action executes the OQL query.</li><li>Query results are sent to the LLM.</li><li>LLM converts raw data into a natural language response.</li><li>Response is displayed in the chatbot.</li></ol><p>This creates a complete Natural Language → OQL → Database → Natural Language pipeline.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/738/1*LnuyqmyUI4cqhAZTsE9KWQ.jpeg" /></figure><h3>Why Not Let the LLM Generate OQL Directly?</h3><p>Many implementations simply send the user’s question to the LLM and ask it to generate SQL or OQL.</p><p>This approach often fails because:</p><ul><li>The LLM does not know your Mendix domain model.</li><li>Entity names may differ from business terminology.</li><li>Associations are unknown.</li><li>Attribute names may be inaccurate.</li></ul><p>For example:</p><p>User asks:</p><p>“Show customers from Pune”</p><p>The actual entity may be:</p><p>CRM.Person</p><p>instead of:</p><p>CRM.Customer</p><p>Without metadata, the LLM will generate invalid queries.</p><p>Therefore, domain model discovery is an important step.</p><h3>Step 1: User Enters a Question</h3><p>Example:</p><p>“Show all customers from Pune”</p><p>The chatbot sends the question to the LLM.</p><h3>Step 2: Extract Important Keywords</h3><p>Prompt:</p><p>Extract business keywords from the question and return comma-separated values.</p><p>Input:</p><p>Show all customers from Pune</p><p>Output:</p><p>customer,pune</p><p>These keywords become the input for entity discovery.</p><h3>Step 3: Discover Entities Using a Java Action</h3><p>A custom Java Action receives:</p><p>customer,pune</p><p>The action scans the Mendix domain model metadata and attempts to find matching entities.</p><p>Example output:</p><p>{<br>“entities”: [<br>{<br>“module”: “CRM”,<br>“entity”: “Customer”<br>}<br>]<br>}</p><p>The action uses Mendix metadata APIs such as:</p><ul><li>Core.getMetaObjects()</li><li>IMetaObject</li><li>IMetaPrimitive</li><li>IMetaAssociation</li></ul><p>to dynamically inspect the domain model.</p><p>This allows the solution to work with any Mendix application.</p><h3>Step 4: Send Metadata Back to the LLM</h3><p>The chatbot now sends:</p><p>User Question:</p><p>Show all customers from Pune</p><p>Available Metadata:</p><p>{<br>“entity”:”CRM.Customer”,<br>“attributes”:[<br>“Name”,<br>“Email”,<br>“City”<br>]<br>}</p><p>The LLM now has sufficient context to generate an accurate OQL query.</p><h3>Step 5: Generate OQL Dynamically</h3><p>LLM Output:</p><p>SELECT Name, Email, City<br>FROM CRM.Customer<br>WHERE City=’Pune’</p><p>Because the LLM knows the entity structure, query accuracy improves significantly.</p><h3>Step 6: Execute OQL Using a Java Action</h3><p>A second Java Action executes the generated OQL.</p><p>Input:</p><p>SELECT Name, Email, City<br>FROM CRM.Customer<br>WHERE City=’Pune’</p><p>Execution:</p><p>Core.retrieveOQLDataTable(<br>context,<br>oqlQuery<br>);</p><p>Result:</p><p>[<br>{<br>“Name”:”John”,<br>“Email”:”<a href="/service/mailto:john@test.com">john@test.com</a>”,<br>“City”:”Pune”<br>},<br>{<br>“Name”:”Alex”,<br>“Email”:”<a href="/service/mailto:alex@test.com">alex@test.com</a>”,<br>“City”:”Pune”<br>}<br>]</p><p>No predefined entity retrieval is required.</p><p>The query result is returned as JSON.</p><h3>Step 7: Convert Raw Data into Natural Language</h3><p>Raw database output is difficult for business users to understand.</p><p>The result JSON and original question are sent back to the LLM.</p><p>Prompt:</p><p>Question:<br>Show all customers from Pune</p><p>Data:<br>[<br>{<br>“Name”:”John”,<br>“Email”:”<a href="/service/mailto:john@test.com">john@test.com</a>”<br>},<br>{<br>“Name”:”Alex”,<br>“Email”:”<a href="/service/mailto:alex@test.com">alex@test.com</a>”<br>}<br>]</p><p>The LLM responds:</p><p>“There are 2 customers from Pune. John (<a href="/service/mailto:john@test.com">john@test.com</a>) and Alex (<a href="/service/mailto:alex@test.com">alex@test.com</a>).”</p><p>This provides a conversational experience instead of a technical data dump.</p><h3>Complete Request Flow</h3><p>The complete workflow becomes:</p><p>Natural Language Question<br>↓<br>Keyword Extraction<br>↓<br>Entity Discovery<br>↓<br>Metadata Generation<br>↓<br>OQL Generation<br>↓<br>OQL Execution<br>↓<br>Data Retrieval<br>↓<br>Natural Language Response</p><h3>Security Considerations</h3><p>When allowing AI to generate queries dynamically, security is extremely important.</p><h3>Validate Entities</h3><p>Ensure generated entities exist in the domain model.</p><h3>Validate Attributes</h3><p>Ensure all attributes are valid.</p><h3>Validate Associations</h3><p>Verify association paths before execution.</p><h3>Limit Result Size</h3><p>Add TOP clauses or pagination.</p><p>Example:</p><p>SELECT TOP 100<br>Name<br>FROM CRM.Customer</p><p>This prevents excessive database load.</p><h3>Use MCP</h3><p>The architecture can be further enhanced using Model Context Protocol (MCP).</p><p>MCP can expose Mendix metadata and query capabilities as tools directly to the LLM, reducing custom prompt engineering and improving maintainability.</p><h3>Benefits of This Approach</h3><p>Advantages include:</p><ul><li>No predefined reports required</li><li>Works across any Mendix domain model</li><li>Supports conversational analytics</li><li>Reduces development effort</li><li>Enables self-service reporting</li><li>Provides natural language answers</li><li>Scales to enterprise applications</li></ul><h3>Future Enhancements</h3><p>Potential improvements:</p><ul><li>Multi-entity joins</li><li>Association path discovery</li><li>Automatic aggregation queries</li><li>Chart generation</li><li>Dashboard generatio</li><li>MCP integration</li><li>Retrieval-Augmented Generation (RAG)</li><li>Semantic search using vector databases</li></ul><h3>Conclusion</h3><p>By combining Mendix metadata APIs, OQL, Java Actions, and Large Language Models, we can build a powerful Natural Language Query Engine that allows users to interact with application data conversationally.</p><p>The architecture converts user questions into executable OQL queries, retrieves data from the Mendix database, and transforms the results into meaningful natural language responses.</p><p>This approach brings AI-powered analytics directly into Mendix applications and provides a foundation for next-generation enterprise chatbots.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a3d059091894" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Model Context Protocol (MCP) in Mendix: The Future of AI-Powered Low-Code Applications]]></title>
            <link>https://medium.com/@vijaymalviya320/model-context-protocol-mcp-in-mendix-the-future-of-ai-powered-low-code-applications-46b48e1dbdb6?source=rss-e58b864b267------2</link>
            <guid isPermaLink="false">https://medium.com/p/46b48e1dbdb6</guid>
            <dc:creator><![CDATA[Vijay Malviya]]></dc:creator>
            <pubDate>Tue, 02 Jun 2026 06:02:04 GMT</pubDate>
            <atom:updated>2026-06-02T06:02:04.840Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><p>Artificial Intelligence is rapidly changing how applications are built and interact with external systems. As AI assistants become more capable, developers need a standardized way to connect AI models with enterprise applications, databases, APIs, and business processes.</p><p>This is where <strong>Model Context Protocol (MCP)</strong> comes into the picture.</p><p>In this article, we’ll explore what MCP is, why it matters, and how Mendix developers can leverage MCP to build smarter applications.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*uSi6luJaTXKxd46r" /></figure><h3>What is MCP?</h3><p><strong>Model Context Protocol (MCP)</strong> is an open protocol that allows AI models to securely interact with external tools, applications, APIs, and data sources.</p><p>Think of MCP as a universal connector between AI models and business systems.</p><p>Instead of building custom integrations for every AI use case, MCP provides a standardized approach for exposing:</p><ul><li>Business APIs</li><li>Databases</li><li>Documents</li><li>Enterprise systems</li><li>Custom tools and services</li></ul><p>to AI assistants and Large Language Models (LLMs).</p><h3>Without MCP</h3><pre>AI → Custom Integration → Application<br>AI → Custom Integration → Database<br>AI → Custom Integration → API</pre><p>Every integration requires separate development and maintenance.</p><h3>With MCP</h3><pre>AI → MCP Server → Business Systems</pre><p>A single standardized interface can expose multiple tools and data sources.</p><h3>Why MCP Matters</h3><p>Organizations often struggle with AI integration because:</p><ul><li>Data is distributed across multiple systems.</li><li>APIs have different formats.</li><li>Security requirements vary.</li><li>Custom integrations are expensive to maintain.</li><li>MCP solves these challenges by providing:</li></ul><h3>Standardization</h3><p>AI models interact with systems using a common protocol.</p><h3>Reusability</h3><p>One MCP server can expose multiple capabilities.</p><h3>Security</h3><p>Access control can be managed centrally.</p><h3>Faster Development</h3><p>Developers spend less time building custom connectors.</p><h3>How MCP Works</h3><p>MCP consists of three main components:</p><h3>1. MCP Client</h3><p>The AI application or assistant.</p><p>Examples:</p><ul><li>Claude Desktop</li><li>AI Agents</li><li>Chat Applications</li><li>Custom AI Solutions</li></ul><h3>2. MCP Server</h3><p>A server that exposes tools and resources.</p><p>Examples:</p><ul><li>CRM APIs</li><li>ERP Systems</li><li>Mendix Applications</li><li>Databases</li></ul><h3>3. Tools &amp; Resources</h3><p>Actual business functionality exposed to AI.</p><p>Examples:</p><ul><li>Create Customer</li><li>Fetch Order Details</li><li>Generate Reports</li><li>Query Databases</li></ul><p>The flow looks like:</p><pre>User<br> ↓<br>AI Assistant<br> ↓<br>MCP Client<br> ↓<br>MCP Server<br> ↓<br>Mendix APIs / Microflows / Data</pre><h3>MCP and Mendix</h3><p>Mendix applications already expose functionality through:</p><ul><li>REST APIs</li><li>OData Services</li><li>Published Microflows</li><li>Custom Java Actions</li><li>This makes Mendix an excellent candidate for MCP integration.</li></ul><h3>Possible Architecture</h3><pre>AI Assistant<br>      ↓<br>   MCP Server<br>      ↓<br> Mendix REST APIs<br>      ↓<br> Microflows &amp; Domain Model</pre><p>The MCP Server acts as a bridge between AI and Mendix.</p><h3>Use Cases for MCP in Mendix</h3><h3>1. AI-Powered Customer Support</h3><p>Imagine a support agent asking:</p><blockquote><em>“Show all open tickets for customer ABC.”</em></blockquote><p>The AI can call a Mendix API through MCP and retrieve real-time data.</p><h3>2. Natural Language Business Operations</h3><p>Users can interact with Mendix applications using plain English.</p><p>Example:</p><blockquote><em>“Create a high-priority incident for SAP integration failure.”</em></blockquote><p>The AI translates this request into a Mendix operation.</p><h3>3. AI-Driven Reporting</h3><p>Users can ask:</p><blockquote><em>“Generate a quality report for the last 30 days.”</em></blockquote><p>The AI fetches data from Mendix and creates insights automatically.</p><h3>4. Mendix Quality Analyzer Assistant</h3><p>For teams building Mendix analysis tools, MCP can expose:</p><ul><li>Model metrics</li><li>Quality rules</li><li>Best-practice checks</li><li>Security findings</li></ul><p>An AI assistant could answer:</p><blockquote><em>“Why is this microflow flagged as complex?”</em></blockquote><p>and retrieve explanations directly from your analyzer.</p><h3>How to Implement MCP with Mendix</h3><h3>Step 1: Expose Mendix Functionality</h3><p>Publish REST APIs that perform business operations.</p><p>Examples:</p><ul><li>Get Customer</li><li>Create Incident</li><li>Fetch Reports</li><li>Run Quality Checks</li></ul><h3>Step 2: Create an MCP Server</h3><p>Build an MCP server using:</p><ul><li>Node.j</li><li>Python</li><li>.NET</li></ul><p>The server wraps Mendix APIs as MCP tools.</p><h3>Step 3: Define Tools</h3><p>Example:</p><pre>{<br>  &quot;name&quot;: &quot;createIncident&quot;,<br>  &quot;description&quot;: &quot;Create a new incident in Mendix&quot;<br>}</pre><h3>Step 4: Connect AI Client</h3><p>Connect Claude, AI agents, or custom applications to the MCP server.</p><h3>Step 5: Execute Mendix Operations</h3><p>When users submit requests, the AI invokes MCP tools that call Mendix APIs and return results.</p><h3>Example: Mendix Incident Management</h3><p>User Prompt:</p><blockquote><em>Create a critical incident for API downtime.</em></blockquote><p>AI Flow:</p><ol><li>AI identifies the required tool.</li><li>MCP calls the Mendix Incident API.</li><li>Mendix creates the incident.</li><li>Response is returned to the user.</li></ol><p>Result:</p><pre>Incident #INC-10234 created successfully.<br>Priority: Critical<br>Status: Open</pre><h3>Benefits for Mendix Developers</h3><h3>Faster AI Integration</h3><p>No need to build AI-specific integrations repeatedly.</p><h3>Reusable Architecture</h3><p>One MCP server can support multiple AI use cases.</p><h3>Better User Experience</h3><p>Natural language interaction becomes possible.</p><h3>Future-Proof Design</h3><p>As AI platforms evolve, MCP provides a standard integration layer.</p><h3>Challenges and Considerations</h3><p>Before implementing MCP:</p><h3>Security</h3><p>Ensure APIs are authenticated and authorized.</p><h3>Data Privacy</h3><p>Limit access to sensitive business information.</p><h3>Performance</h3><p>Optimize API response times.</p><h3>Governance</h3><p>Monitor AI actions and maintain audit logs.</p><h3>Conclusion</h3><p>Model Context Protocol (MCP) is emerging as a standard way for AI systems to interact with enterprise applications. For Mendix developers, MCP opens the door to building intelligent, AI-powered low-code solutions without creating countless custom integrations.</p><p>By exposing Mendix business capabilities through APIs and wrapping them with an MCP server, organizations can enable AI assistants to securely access real business data and processes.</p><p>As AI adoption grows, understanding MCP today can help Mendix developers stay ahead of the next wave of enterprise application development.</p><h3>Key Takeaways</h3><ul><li>MCP standardizes communication between AI and business systems.</li><li>Mendix applications can easily participate through REST APIs.</li><li>MCP enables AI-powered workflows, reporting, and automation.</li><li>AI assistants can execute Mendix business processes securely.</li><li>MCP is likely to become a foundational component of enterprise AI architectures.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=46b48e1dbdb6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Agentic AI in Mendix: Building Autonomous Enterprise Apps with Azure OpenAI]]></title>
            <link>https://medium.com/@vijaymalviya320/agentic-ai-in-mendix-building-autonomous-enterprise-apps-with-azure-openai-e714026f5aac?source=rss-e58b864b267------2</link>
            <guid isPermaLink="false">https://medium.com/p/e714026f5aac</guid>
            <dc:creator><![CDATA[Vijay Malviya]]></dc:creator>
            <pubDate>Fri, 13 Mar 2026 05:05:33 GMT</pubDate>
            <atom:updated>2026-03-13T05:05:33.853Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><p>Artificial Intelligence in business applications is rapidly evolving from simple chatbots to autonomous systems that can plan, decide, and act. This new paradigm is known as <strong>Agentic AI</strong> — AI systems that behave like goal‑driven agents rather than passive responders.</p><p>For low‑code platforms like Mendix, Agentic AI unlocks a powerful future: applications that can automate complex workflows, interact with enterprise systems, and continuously improve outcomes with minimal human intervention.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*drVE_7ATFB2_8IA9.jpeg" /></figure><p>In this article, we will explore:</p><ul><li>What Agentic AI is</li><li>Why it matters for Mendix developers</li><li>How to implement Agentic AI using Azure OpenAI</li><li>A practical enterprise example</li><li>Architecture, implementation steps, and best practices</li></ul><h3>What is Agentic AI?</h3><p>Agentic AI refers to AI systems that can:</p><ul><li>Understand goals</li><li>Break them into tasks</li><li>Make decisions</li><li>Execute actions using tools/APIs</li><li>Observe results</li><li>Iterate until completion</li></ul><p>Unlike traditional AI (which only responds to prompts), agentic systems are proactive and autonomous.</p><h3>Why Agentic AI is Powerful for Mendix Apps</h3><p>Mendix excels at orchestrating enterprise workflows. Agentic AI enhances this by adding intelligent decision‑making.</p><h3>Key Benefits</h3><p><strong>1. Hyper‑Automation</strong><br>Automate complex processes like incident resolution, procurement, or compliance checks.</p><p><strong>2. Intelligent Assistants Inside Apps</strong><br>Move beyond chatbots to systems that actually perform tasks.</p><p><strong>3. Reduced Manual Work</strong><br>AI can call APIs, update records, and trigger workflows automatically.</p><p><strong>4. Context‑Aware Decisions</strong><br>Agents can analyze business data before acting.</p><h3>Using Mendix Prebuilt Modules (Marketplace)</h3><p>One of the biggest advantages of Mendix is its rich ecosystem of prebuilt modules available in the Mendix Marketplace. These modules significantly reduce development time and allow you to focus on business logic rather than infrastructure.</p><h3>AI‑Focused Modules (if available in your environment)</h3><ul><li>GenAI Connector modules</li><li>Prompt management components</li><li>Vector database integrations</li><li>Document processing modules</li><li>Agent Common Modules</li></ul><p>These components can help you build Retrieval‑Augmented Generation (RAG) capabilities on top of your agent.</p><h3>Architecture: Agentic AI with Mendix + Azure OpenAI</h3><p>A typical architecture looks like this:</p><p>User → Mendix UI → Microflow → Agent Service → Azure OpenAI → External Tools/APIs → Mendix DB</p><h3>Components</h3><p><strong>Mendix Application</strong></p><ul><li>UI for users</li><li>Microflows for orchestration</li><li>Business data and workflows</li></ul><h3>Practical Example: Autonomous IT Support Agent</h3><h3>Use Case</h3><p>An employee reports:</p><blockquote><em>“My laptop VPN is not working and I cannot access internal systems.”</em></blockquote><p>Instead of creating a ticket for manual handling, an agentic system can:</p><ol><li>Diagnose the issue</li><li>Check user device status</li><li>Verify VPN service health</li><li>Reset credentials if needed</li><li>Create or update ticket</li><li>Notify the user</li></ol><h3>Recommended Controls</h3><ul><li>Human approval for sensitive actions</li><li>Audit logging of agent decisions</li><li>Role‑based access control</li><li>Rate limiting</li><li>Guardrails for prompts</li><li>Secure secret management</li></ul><h3>Performance Tips for Production</h3><ul><li>Cache frequent responses</li><li>Use smaller models for simple tasks</li><li>Maintain conversation state efficiently</li><li>Implement retry strategies</li><li>Monitor token usage and cost</li></ul><h3>Real Enterprise Use Cases</h3><p>Agentic AI in Mendix can power:</p><ul><li>Autonomous IT support</li><li>Procurement automation</li><li>Compliance monitoring</li><li>Customer service resolution</li><li>Data quality management</li><li>DevOps incident handling</li></ul><h3>Future of Agentic AI in Low‑Code Platforms</h3><p>Low‑code + Agentic AI will redefine enterprise development. Instead of building rigid workflows, developers will design goal‑oriented systems where AI decides the execution path.</p><p>Mendix developers who adopt this early will be able to build next‑generation applications that are adaptive, intelligent, and highly automated.</p><h3>Conclusion</h3><p>Agentic AI is not just an enhancement — it represents a shift from software that executes instructions to software that pursues outcomes.</p><p>By combining Mendix’s orchestration capabilities with Azure OpenAI’s reasoning power, you can build enterprise applications that diagnose problems, take action, and continuously improve — all with minimal human intervention.</p><p>The future of enterprise apps is autonomous, and Mendix is well‑positioned to lead this transformation.</p><h3>About the Author</h3><p>Mendix developer passionate about building intelligent enterprise tools, performance optimization, and modern cloud‑native architectures.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e714026f5aac" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Maia AI Assistant in Mendix: Your AI Co‑Developer in the Low‑Code Journey]]></title>
            <link>https://medium.com/@vijaymalviya320/maia-ai-assistant-in-mendix-your-ai-co-developer-in-the-low-code-journey-941ad1d8c637?source=rss-e58b864b267------2</link>
            <guid isPermaLink="false">https://medium.com/p/941ad1d8c637</guid>
            <dc:creator><![CDATA[Vijay Malviya]]></dc:creator>
            <pubDate>Mon, 05 Jan 2026 05:22:18 GMT</pubDate>
            <atom:updated>2026-01-05T05:22:18.897Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>How Mendix Maia is changing the way businesses and developers build enterprise applications — faster, smarter, and with confidence</strong></p><h3>The Moment Every Mendix Developer Recognizes</h3><p>You open Mendix Studio Pro.</p><p>A requirement just landed in your inbox. The deadline is tight. The app needs to be scalable, secure, and enterprise‑ready. You start modeling — domain entities, microflows, pages — and suddenly questions pile up:</p><ul><li><em>Is this the right design pattern?</em></li><li><em>Am I following best practices?</em></li><li><em>Is there a faster way to build this?</em></li></ul><p>This is exactly where <strong>Maia AI Assistant</strong> steps in — not as a replacement, but as a <strong>co‑developer sitting beside you</strong>, guiding every step of your low‑code journey.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*H8qUQJYcfq1pUQThBVngPQ.png" /></figure><h3>What Is Maia AI Assistant?</h3><p><strong>Maia (Mendix Artificial Intelligence Assistance)</strong> is an AI-powered assistant embedded directly inside the Mendix ecosystem — especially within <strong>Mendix Studio Pro</strong>.</p><p>Unlike generic AI tools, Maia understands Mendix <em>deeply</em>. It speaks the language of:</p><ul><li>Domain models</li><li>Microflows and nanoflows</li><li>Pages and widgets</li><li>Workflows and security</li><li>Mendix best practices and performance rules</li></ul><p>Trained on anonymized Mendix models and platform knowledge, Maia provides <strong>context-aware guidance exactly when you need it</strong> — without breaking your flow.</p><p>Think of Maia as a mix of:</p><ul><li>A senior Mendix architect</li><li>A documentation expert</li><li>A productivity booster</li></ul><p>All wrapped into one AI assistant.</p><h3>The Three Pillars of Maia: Guidance, Assistance, and Generation</h3><p>Maia’s power comes from how seamlessly it integrates into the development experience.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yAa7l4YRPZcTpAbSs4eILA.png" /></figure><h3>1. Guidance: Answers Without Context Switching</h3><p><strong>Maia Chat</strong> lives inside Studio Pro. Instead of opening documentation or forums, you simply ask:</p><blockquote>“What is the best way to handle validations here?”</blockquote><p>Maia responds instantly — with explanations, examples, and links.</p><p>For beginners, Maia explains complex concepts in simple terms. For experienced developers, it becomes a fast reference and decision support tool.</p><h3>2. Assistance: Real‑Time Intelligence While You Build</h3><p>As you model your application, Maia quietly observes and assists.</p><ul><li><strong>Best Practice Recommender</strong> detects architectural issues early</li><li><strong>Logic Recommender</strong> suggests the next logical step in microflows</li><li><strong>Workflow Recommender</strong> improves approval and business process design</li><li><strong>UI Recommender</strong> suggests the right widgets for the job</li></ul><p>This is not passive feedback — it’s <strong>real-time, contextual intelligence</strong> that improves quality as you build.</p><h3>3. Generation: From Idea to App — Faster</h3><p>Maia doesn’t just advise — it builds with you.</p><p>With simple natural-language prompts, Maia can:</p><ul><li>Generate a starter application</li><li>Create domain models from descriptions</li><li>Build pages and layouts</li><li>Generate workflows and OQL queries</li><li>Translate UI text into multiple languages</li></ul><p>What once took hours can now take minutes.</p><h3>How Maia Helps Businesses (Beyond the Code)</h3><p>Maia’s real value is not just technical — it’s <strong>business impact</strong>.</p><h3>Faster Time to Market</h3><p>By automating repetitive tasks and accelerating decision-making, Maia enables teams to deliver applications faster — turning ideas into production-ready apps in record time.</p><h3>Lower Cost, Higher Quality</h3><p>Early detection of best-practice violations reduces rework, bugs, and technical debt. This directly translates to <strong>lower development and maintenance costs</strong>.</p><h3>Better Collaboration Between Business and IT</h3><p>Because Maia understands natural language, business users and citizen developers can actively participate — bridging the gap between requirements and implementation.</p><h3>How Maia Empowers Developers</h3><p>For developers, Maia feels like having a mentor built into the IDE.</p><h3>Faster Learning Curve</h3><p>New to Mendix? Maia explains concepts instantly. No long onboarding cycles. No guesswork.</p><h3>Focus on What Matters</h3><p>By handling boilerplate and repetitive modeling, Maia frees developers to focus on architecture, innovation, and user experience.</p><h3>Confidence in Every Decision</h3><p>With continuous best-practice validation, developers build with confidence — knowing the app is scalable, secure, and maintainable.</p><h3>Real‑World Prompts Developers Actually Use</h3><p>Here’s how developers interact with Maia in day‑to‑day work:</p><p><strong>Kick‑Start an App</strong></p><blockquote><em>Create a Mendix app for employee leave management with Employee and Manager roles.</em></blockquote><p><strong>Design a Domain Model</strong></p><blockquote><em>Create a domain model for an e‑commerce app with orders and customers.</em></blockquote><p><strong>Understand Logic</strong></p><blockquote><em>Explain what this microflow does in simple terms.</em></blockquote><p><strong>Improve Quality</strong></p><blockquote><em>Optimize this microflow using Mendix best practices.</em></blockquote><p><strong>Design Workflows</strong></p><blockquote><em>Create a two‑level approval workflow for expenses.</em></blockquote><p>These prompts feel natural — because that’s the point.</p><h3>Why Maia Is a Glimpse Into the Future of Low‑Code</h3><p>Maia represents a shift in how applications are built.</p><p>Low-code is no longer just about visual modeling — it’s about <strong>AI‑augmented development</strong>.</p><p>For businesses, Maia means:</p><ul><li>Faster delivery</li><li>Reduced cost</li><li>Enterprise-grade quality</li></ul><p>For developers, Maia becomes:</p><ul><li>A co‑developer</li><li>A mentor</li><li>A quality guardian</li></ul><h3>Final Thoughts</h3><p>Maia AI Assistant doesn’t replace developers.</p><p>It <strong>amplifies them</strong>.</p><p>By combining human creativity with AI intelligence, Mendix teams can build applications that are not only faster — but better designed, more maintainable, and ready for the enterprise.</p><p>If Mendix is the engine of low-code innovation, <strong>Maia is the intelligence that makes it smarter</strong>.</p><p>#Mendix #MaiaAI #LowCode #ArtificialIntelligence #EnterpriseApps #SoftwareDevelopment #NoCode #DigitalTransformation</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=941ad1d8c637" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Customizing Widgets in Mendix: Turning Use-Cases into Powerful UI Components]]></title>
            <link>https://medium.com/@vijaymalviya320/customizing-widgets-in-mendix-turning-use-cases-into-powerful-ui-components-f795d7a38f20?source=rss-e58b864b267------2</link>
            <guid isPermaLink="false">https://medium.com/p/f795d7a38f20</guid>
            <dc:creator><![CDATA[Vijay Malviya]]></dc:creator>
            <pubDate>Sat, 29 Nov 2025 06:21:19 GMT</pubDate>
            <atom:updated>2025-11-29T06:21:19.725Z</atom:updated>
            <content:encoded><![CDATA[<p>A practical guide for Mendix developers using Pluggable Widgets, JavaScript Actions, and Atlas styling</p><p>Mendix gives developers a fast, visual way to build applications — but when project requirements go beyond what standard widgets can offer, customization becomes essential.</p><p>From creating entirely new pluggable widgets to extending existing ones with JavaScript and UI styling, customization is one of the skills that separates intermediate developers from advanced Mendix engineers.</p><p>In this article, we’ll walk through:</p><ul><li>Why widget customization is needed</li><li>Types of customization in Mendix</li><li>Real-world use case examples</li><li>How to extend existing widgets</li><li>How to build pluggable widgets</li></ul><p>Styling with Atlas, Sass, and Design Properties</p><p>Passing microflow/nanoflow events to custom widgets</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*hOSyjAle1ZtCofyC.jpg" /></figure><p><strong>Best practices &amp; patterns</strong></p><p>This guide applies to Mendix 9+ and 10+, especially versions 10.12+, where pluggable widgets have become more stable and feature-rich.</p><p>— -</p><p>🌟<strong> Why Customization Matters in Enterprise Apps</strong></p><p>Standard Mendix widgets cover 80% needs, but enterprise projects often require:</p><p>✔ Custom interactions</p><p>✔ Complex data visualizations</p><p>✔ Dynamic UI behavior</p><p>✔ Integrations with third-party JS libraries</p><p>✔ Performance-optimized components</p><p>✔ A branded UI design system</p><p>Examples include:</p><ul><li>A custom drag-and-drop flow editor (React Flow)</li><li>Interactive timeline chart</li><li>Editable spreadsheet-like grid</li><li>Visual network graph</li><li>Dynamic forms based on user input</li><li>Custom map with layers</li><li>Advanced search and filter controls</li></ul><p>These aren’t possible with default widgets — hence customization.</p><p>— — —</p><p>1. <strong>Levels of Widget Customization in Mendix</strong></p><p>Customization can be broken into four levels, depending on your need.</p><p># <strong>Level 1 — Styling Customization (Atlas UI + Sass)</strong></p><p>Modify the appearance of existing widgets.</p><p>Techniques:</p><ul><li>CSS/Sass overrides</li><li>Atlas theme variables</li><li>Custom classes</li><li>Layout adjustments</li><li>Conditional visibility formatting</li></ul><p>Example:</p><pre>.mx-button-custom {<br>background-color: #2b5876;<br>border-radius: 12px;<br>font-weight: bold;<br>}</pre><p>When to choose:</p><p>→ UI branding</p><p>→ Replacing standard Atlas look</p><p>→ Improving UX with layout tuning</p><p>— — —</p><p># <strong>Level 2 — Behavior Customization Using JavaScript Actions</strong></p><p>Add custom behavior without building a full widget.</p><p>Use:</p><ul><li>DOM manipulation</li><li>Read/modify widget states</li><li>Post-render logic</li><li>Custom event listeners</li></ul><p><strong>Example</strong>: Automatically scroll to top after a microflow completes.</p><pre>export async function ScrollToTop() {<br>window.scrollTo({ top: 0, behavior: &quot;smooth&quot; });<br>}</pre><p>When to choose:</p><p>→ Quick enhancements</p><p>→ Post-render tweaks</p><p>→ Adding shortcuts/interactions</p><p>— — —</p><p><strong>Level 3 — Extending an Existing Mendix Widget</strong></p><p>Wrap an existing widget and add custom logic on top.</p><p>Example:</p><ul><li>Extend Data Grid 2 to support “Bookmark Filters”</li><li>Add auto-submit behavior to Dropdown</li><li>Add animations to a container widget</li></ul><p>You use:</p><ul><li>Widget DOM</li><li>Widget props</li><li>Mendix client APIs</li><li>JS event binding</li></ul><p>This is great for adding missing features.</p><p>— — —</p><p><strong>Level 4 —</strong> <strong>Building a Full Pluggable Widget (React-based)</strong></p><p>For fully custom UI logic or embedding third-party libraries.</p><p>Tools:</p><ul><li>Pluggable widgets (React)</li><li>Webpack</li><li>TypeScript</li><li>Mendix Widget Generator (yo mendix)</li><li>External JS frameworks/libraries</li></ul><p>When to choose:</p><p>→ New UI component not available in Mendix</p><p>→ Complex UX interactions</p><p>→ Third-party UX library integration</p><p>— — —</p><h4>2. Understanding When to Customize (Use Case Examples)</h4><p>Here are real-world scenarios where customization is necessary:</p><p>—</p><p><strong>Use Case 1</strong>: <strong>Data Grid Custom Filters (Advanced Search Panel)</strong></p><p><strong>Problem</strong>: <strong>User wants multi-column, multi-condition advanced filtering</strong></p><p>Example:</p><p>Q. “Amount &gt; 1000 AND Status = Pending AND CreatedDate between X–Y”</p><p><strong>Solution:</strong></p><p>Create a custom “Filter Panel” pluggable widget</p><p>Bind fields to Data Grid 2 filter API</p><p>Save user presets (Bookmarks)</p><p>— — —</p><p><strong>Use Case 2:</strong> <strong>Interactive Flow Editor (React Flow Integration)</strong></p><p><strong>Problem</strong>: Mendix cannot draw dynamic graphs natively.</p><p><strong>Solution:</strong></p><p>Build a pluggable widget using React Flow</p><p>Expose Mendix events to JS</p><p>Allow node creation, drag-drop, connections</p><p>Sync back to Mendix microflows</p><p>This becomes a modelling interface inside Mendix.</p><p>— — —</p><p><strong>Use Case 3:</strong> <strong>Editable Excel-like Table</strong></p><p><strong>Problem:</strong> Customer needs an inline editable table with:</p><p>Ctrl+C/Ctrl+V</p><p>Multi-row select</p><p>Keyboard navigation</p><p>Real-time validation</p><p><strong>Solution:</strong></p><p>Implement Handsontable or AG Grid inside a pluggable widget</p><p>Expose “onCellEdit” events</p><p>Use local state + commit on save</p><p>— -</p><p><strong>Use Case 4: Customized File Upload with Progress Bar</strong></p><p><strong>Problem</strong>: Standard upload widget is too limited.</p><p><strong>Solution:</strong></p><p>Create a custom upload component</p><p>Show sequential upload</p><p>Show progress indicator</p><p>Provide error-handling UI</p><p>— -</p><p><strong>Use Case 5: Dynamic Sidebar Menu Based on Roles</strong></p><p><strong>Problem</strong>: Standard navigation is static.</p><p><strong>Solution:</strong></p><p>Build a menu widget</p><p>Fetch allowed menu items via microflow</p><p>Animate open/close transitions</p><p>— — —</p><h4>3. Anatomy of a Pluggable Widget</h4><p>A Mendix widget consists of:</p><p>1. XML definition (MyWidget.xml)</p><p>Defines:</p><p>Properties</p><p>Events</p><p>Data sources</p><p>Design-time preview</p><p>2. React Component (MyWidget.tsx)</p><p>Core logic:</p><pre>export function MyWidget(props: MyWidgetContainerProps) {<br>const { text, onClick } = props;<br>return (<br>&lt;button className=&quot;custom-button&quot; onClick={onClick}&gt;<br>{text}<br>&lt;/button&gt;<br>);<br>}</pre><p>3. Styling (ui/MyWidget.css)</p><p>4. Package.json &amp; build configuration</p><p>5. Widget Preview (Editor)</p><p>— — —</p><p><strong>4. Adding Design Properties (Making Your Widget Configurable)</strong></p><p>Design properties let the user customize widget behavior in Studio Pro.</p><p><strong>Examples:</strong></p><p>Number of items</p><p>Color</p><p>Alignment</p><p>Boolean flags</p><p>DataSources</p><p>Microflow/Nanoflow events</p><p>Example design property:</p><pre>&lt;property key=&quot;showBorder&quot; type=&quot;Boolean&quot; defaultValue=&quot;true&quot;&gt;<br>&lt;caption&gt;Show border&lt;/caption&gt;<br>&lt;/property&gt;</pre><p>— — —</p><p><strong>5. Integrating Third-Party Libraries</strong></p><p>This is the biggest use case of pluggable widgets.</p><p><strong>Examples:</strong></p><p>React Flow (diagramming)</p><p>Chart.js or Recharts</p><p>Leaflet / Mapbox</p><p>AG Grid / Handsontable</p><p>D3.js</p><p>FullCalendar</p><p>PDF.js</p><p><strong>Example installation:</strong></p><p>npm install reactflow</p><p>Inside widget:</p><p>import ReactFlow from “reactflow”;</p><p>— -</p><h4>6. Passing Data Between Mendix and Widgets</h4><p>You can pass:</p><p>✔ Attributes</p><p>✔ Objects</p><p>✔ Associations</p><p>✔ Actions (microflows/nanoflows)</p><p>✔ Editable values</p><p>Example: Calling a nanoflow on click:</p><p>props.onNodeClick?.execute();</p><p>— -</p><h4>7. Best Practices for Widget Customization</h4><p>1. Hide complexity from the Mendix developer</p><p>Widget should look simple:</p><ul><li>Clear properties</li><li>Minimal configuration</li><li>Good defaults</li></ul><p>2. Keep styling separate</p><p>Use:</p><ul><li>CSS Modules</li><li>Tailwind (optional)</li><li>Scoped class names</li></ul><p>3. Avoid storing large data inside Mendix attributes</p><ul><li>Use widget internal state to improve performance.</li></ul><p>4. Use memoization</p><ul><li>For expensive renders:</li></ul><p>useMemo(…)</p><p>useCallback(…)</p><p>5. Add error boundaries</p><ul><li>To avoid widget crash in runtime.</li></ul><p>— — —</p><p><strong>8. Final Example — A Real Custom Widget Workflow</strong></p><p>Let’s say the use case is:</p><p><strong>❝ Show a card layout where clicking a card opens a popup and refreshes a list ❞</strong></p><p>Steps:</p><p>1. Create a pluggable widget → CardViewer</p><p>2. Add properties</p><ul><li>Title (string)</li><li>Image (binary attribute)</li><li>OnClick event (nanoflow)</li></ul><p>3. Build a React component that displays styled cards</p><p>4. Call the nanoflow when card is clicked</p><p>5. Add custom CSS for branded styling</p><p>6. Publish widget inside Mendix App Store module</p><p>7. Use inside pages</p><p>This workflow generalises to almost any use case.</p><p>— — —</p><h4>Conclusion</h4><p>Widget customization is one of the strongest tools in Mendix development. Whether you’re enhancing existing widgets, styling them for specific use cases, or building entirely new pluggable components, you’re effectively extending the Mendix platform to deliver true enterprise-level experiences.</p><p>You now know:</p><p>✔ When customization is needed</p><p>✔ Different types of customization</p><p>✔ How to build/extend widgets</p><p>✔ How to integrate JS libraries</p><p>✔ Best practices for scalable widget development</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f795d7a38f20" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ Implementing OAuth 2.0 On-Behalf-Of (OBO) Flow in Azure and Connecting It with Mendix]]></title>
            <link>https://medium.com/@vijaymalviya320/implementing-oauth-2-0-on-behalf-of-obo-flow-in-azure-and-connecting-it-with-mendix-e6b3b155aaad?source=rss-e58b864b267------2</link>
            <guid isPermaLink="false">https://medium.com/p/e6b3b155aaad</guid>
            <dc:creator><![CDATA[Vijay Malviya]]></dc:creator>
            <pubDate>Thu, 13 Nov 2025 13:17:05 GMT</pubDate>
            <atom:updated>2025-11-13T13:17:05.398Z</atom:updated>
            <content:encoded><![CDATA[<p>In modern enterprise applications, secure data access is essential — especially when your Mendix app needs to call downstream APIs (like Microsoft Graph, Microsoft Fabric, or custom Azure-hosted APIs) on behalf of the currently logged-in user.</p><p>That’s where OAuth 2.0 On-Behalf-Of (OBO) Flow comes in.</p><p>In this article, we’ll understand what OBO is, why it’s needed, and how to integrate it step-by-step in a Mendix application using Azure AD.</p><p>— — —</p><p>🧩 <strong>What Is the OBO Flow?</strong></p><p>The On-Behalf-Of (OBO) flow is an OAuth 2.0 grant type that allows a service (like Mendix) to exchange an incoming user token for another token that can access downstream APIs.</p><p>Think of it like a chain of trust:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*X_oZ7pFVQDpMMmWFwnSnSg.png" /></figure><p>When a user logs into Mendix via Azure AD (OIDC or SSO), Mendix receives an access token. But that token usually has an audience (aud) for Mendix itself — not for Microsoft Graph or another API.</p><p>If Mendix directly calls Microsoft Graph with that token, it will fail because the Graph API expects a token whose audience is <a href="https://graph.microsoft.com./">https://graph.microsoft.com.</a></p><p>So, Mendix performs the OBO flow:</p><p>1. Takes the user’s token (for Mendix)</p><p>2. Exchanges it with Azure AD</p><p>3. Gets a new access token (for the downstream API)</p><p>4. Calls the downstream API on behalf of the user</p><p>— — —</p><p><strong>What is Downstream API ?</strong></p><ul><li><strong>Definition</strong>: Refers to the direction of data or requests <strong>away from the source</strong> or <strong>toward the consumer</strong> of the data.</li><li><strong>Example</strong>: If Service A calls Service B, Service A is considered <strong>downstream</strong> relative to Service B.</li><li><strong>Use Case</strong>: A downstream service consumes data or functionality from another service.</li></ul><pre>Service B (Upstream) → Service A (Downstream)</pre><p>🧠 <strong>When Do You Need OBO?</strong></p><p>You should use OBO when:</p><p>Your Mendix app authenticates users using Azure AD.</p><p>You need to access another protected resource (like Microsoft Graph, Fabric Lakehouse SQL, or a custom Azure API).</p><p>You want the access to respect the user’s context (permissions, roles, etc.).</p><p>— — —</p><p>⚙️ <strong>Step-by-Step Implementation</strong></p><p>Let’s walk through how to implement OBO in your Mendix environment.</p><p>— —</p><p>🏗 Step 1: <strong>Set Up Your Azure AD Applications</strong></p><p>You’ll need two app registrations in Azure AD:</p><p>1️⃣ The Client App (Mendix App)</p><p>Represents your Mendix app (frontend + backend).</p><p>Handles user authentication.</p><p>Calls downstream APIs.</p><p>Key configurations:</p><p>Redirect URI → Mendix SSO callback URL (e.g., <a href="https://yourapp.mendixcloud.com/oauth/callback)">https://yourapp.mendixcloud.com/oauth/callback)</a></p><p>Enable “ID tokens” and “Access tokens” in the authentication section.</p><p>Grant API permissions to downstream APIs (e.g., Microsoft Graph, your Azure API).</p><p>Note down:</p><p>Client ID — xxxxxx</p><p>Client Secret — xxxxxx</p><p>Tenant ID — xxxxxx</p><p>2️⃣ <strong>The Resource App (Downstream API)</strong></p><p>Represents the API Mendix will call on behalf of the user.</p><p>Exposes permissions/scopes for access.</p><p>Key configurations:</p><p>In “Expose an API” → Add a Scope (e.g., api://&lt;api-client-id&gt;/.access_as_user)</p><p>— —</p><p>🔁 Step 2: <strong>Configure the OBO Permission</strong></p><p>In the Client App (Mendix):</p><p>Under API permissions, click Add a permission → My APIs.</p><p>Select your Resource App.</p><p>Choose the scope you created earlier.</p><p>Grant Admin consent for the tenant.</p><p>This ensures Mendix can request tokens for the downstream API.</p><p>— —</p><p>🔑 Step 3: <strong>Mendix OIDC Configuration</strong></p><p>If you’re using Mendix SSO (OIDC), you’ll already receive a valid access token for the Mendix audience.</p><p>To perform OBO:</p><p>1. Extract the incoming access token from the user session.</p><p>In Mendix, this can be done via the OIDC module.</p><p>Example: use the “GetTokenFromSession” microflow or a Java action.</p><p>2. Pass that token to a Java Action that performs the OBO exchange.</p><p>— —</p><p>💻 Step 4: <strong>Java Action — Exchange Token Using OBO Flow</strong></p><p>Here’s a sample Java implementation for the OBO token exchange:</p><pre>import java.io.OutputStream;<br>import java.net.HttpURLConnection;<br>import java.net.URL;<br>import java.util.Scanner;<br>public class OboTokenHelper {<br>public static String getOboAccessToken(String tenantId, String clientId, String clientSecret, String userToken, String apiScope) throws Exception {<br>String url = &quot;https://login.microsoftonline.com/&quot; + tenantId + &quot;/oauth2/v2.0/token&quot;;<br>String body = &quot;client_id=&quot; + clientId<br>+ &quot;&amp;client_secret=&quot; + clientSecret<br>+ &quot;&amp;grant_type=urn:ietf:params:oauth:grant-type:jwt-bearer&quot;<br>+ &quot;&amp;requested_token_use=on_behalf_of&quot;<br>+ &quot;&amp;assertion=&quot; + userToken<br>+ &quot;&amp;scope=&quot; + apiScope;<br>HttpURLConnection conn = (HttpURLConnection) new URL(url).openConnection();<br>conn.setRequestMethod(&quot;POST&quot;);<br>conn.setDoOutput(true);<br>conn.setRequestProperty(&quot;Content-Type&quot;, &quot;application/x-www-form-urlencoded&quot;);<br>try (OutputStream os = conn.getOutputStream()) {<br>os.write(body.getBytes());<br>}<br>Scanner sc = new Scanner(conn.getInputStream());<br>String response = sc.useDelimiter(&quot;\\Z&quot;).next();<br>sc.close();<br>return response;<br>}<br>}<br><br></pre><pre>You&#39;ll get a JSON response like:<br>{<br>&quot;token_type&quot;: &quot;Bearer&quot;,<br>&quot;expires_in&quot;: 3599,<br>&quot;access_token&quot;: &quot;&lt;NEW_ACCESS_TOKEN&gt;&quot;<br>}</pre><p>Use this token to call your downstream API or Microsoft Graph.</p><p>— -</p><p>🧩 Step 5: <strong>Use the Token in Mendix</strong></p><p>Once you have the OBO token, you can call APIs directly from Mendix using:</p><p>Call REST Activity (Microflow)</p><p>Custom Java Action with HttpClient</p><p>Example in Mendix Microflow:</p><p>HTTP Header: Authorization: Bearer &lt;obo_access_token&gt;</p><p>URL: <a href="https://graph.microsoft.com/v1.0/me">https://graph.microsoft.com/v1.0/me</a></p><p>This will now return data for the logged-in user — securely and in context.</p><p>— — —</p><p>🧰<strong> Common Issues and Fixes</strong></p><p>Issue Cause Fix</p><p>invalid_grant Wrong token type (use user access token, not ID token) Pass the access_token from OIDC</p><p>invalid_audience API permission mismatch Ensure the API’s Application ID URI matches the requested scope</p><p>unauthorized_client App not granted consent Grant admin consent in Azure AD</p><p>AADSTS50013 Missing requested_token_use=on_behalf_of Add this parameter to token request</p><p>— — —</p><p>🧠<strong> Pro Tips</strong></p><p>Always cache tokens for performance (Mendix can store them in non-persistent entities or runtime memory).</p><p>Use Postman to test your OBO flow before Mendix integration.</p><p>Use jwt.ms to decode and verify your tokens.</p><p>Use App roles or groups claims in the token to apply role-based access in Mendix.</p><p>— — —</p><p>✅ <strong>Summary</strong></p><p>The OAuth 2.0 On-Behalf-Of flow bridges the gap between user authentication in Mendix and secure access to downstream Microsoft or Azure APIs.</p><p>By setting up your Azure AD apps correctly and implementing the OBO exchange logic in Mendix, you can:</p><ul><li>Authenticate once (SSO)</li><li>Access multiple Azure services securely</li><li>Maintain user context end-to-end</li><li>This is particularly powerful for scenarios like:</li><li>Accessing Microsoft Graph (user calendars, emails, etc.)</li><li>Querying Microsoft Fabric Lakehouse SQL endpoints</li><li>Calling custom APIs hosted on Azure App Service</li></ul><p>— -</p><p>🧾 References</p><p>Microsoft Docs: <a href="https://learn.microsoft.com/en-us/entra/identity-platform/v2-oauth2-on-behalf-of-flow">OAuth 2.0 On-Behalf-Of Flow</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e6b3b155aaad" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Unlocking the Power of Generative AI in Mendix: Agent Commons & MCP Server]]></title>
            <link>https://medium.com/@vijaymalviya320/unlocking-the-power-of-generative-ai-in-mendix-agent-commons-mcp-server-a53e11f075ad?source=rss-e58b864b267------2</link>
            <guid isPermaLink="false">https://medium.com/p/a53e11f075ad</guid>
            <dc:creator><![CDATA[Vijay Malviya]]></dc:creator>
            <pubDate>Thu, 16 Oct 2025 14:30:33 GMT</pubDate>
            <atom:updated>2025-10-16T14:30:33.438Z</atom:updated>
            <content:encoded><![CDATA[<p>Generative AI (GenAI) is rapidly transforming how applications interact with users, generate content, and automate workflows. Mendix, a low-code platform, provides specialized modules like Agent Commons and MCP Server to help developers build intelligent, AI-driven applications. In this article, we’ll explore how to integrate GenAI into Mendix, leverage agents, and manage AI workflows using the MCP Server Module.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*Vr3QKoFmBbf1AfI9.jpg" /></figure><p>— — —</p><p><strong>What is Generative AI?</strong></p><p>Generative AI refers to AI systems capable of creating new content based on input prompts. Examples include:</p><p><strong>Text generation</strong>: Chatbots, summaries, automated content</p><p><strong>Image generation</strong>: AI-generated graphics, product visuals</p><p><strong>Code generation</strong>: Automated code suggestions</p><p><strong>Audio/Video generation</strong>: Voice cloning, video scripts</p><p>Mendix allows you to integrate these capabilities through REST APIs or specialized AI modules.</p><p>— — —</p><p><strong>Mendix Agent Commons: AI Agents Made Simple</strong></p><p>Agent Commons is a Mendix module that allows you to create autonomous agents. Agents are “digital workers” that can:</p><ul><li>Execute tasks independently</li><li>Interact with users or other agents</li><li>Integrate with AI models or external APIs</li><li>Automate repetitive business processes</li></ul><p><strong>Key Features:</strong></p><p>1. <strong>Role-based Agents</strong>: Each agent has a specific role, e.g., TranslatorBot, SummarizerBot.</p><p>2. <strong>Task Management</strong>: Agents manage pending, in-progress, and completed tasks.</p><p>3. <strong>Extensible</strong>: Agents can call external services, including GenAI APIs.</p><p>— — —</p><p><strong>MCP Server</strong></p><p>The <a href="https://modelcontextprotocol.io/introduction">Model Context Protocol (MCP)</a> is an open protocol that standardizes how Large Language Models (LLMs) can autonomously connect to apps. Many AI platforms and third-party systems have already adopted MCP for easier integration and empowerment of LLMs.</p><p>Mendix provides an <a href="https://docs.mendix.com/appstore/modules/genai/genai-for-mx/mcp-server/">MCP Server</a> module to facilitate an MCP server from a Mendix app, enabling developers to expose tools and prompts to external MCP clients as well as an <a href="https://marketplace.mendix.com/link/component/244893">MCP Client</a> module. The MCP Client module enables your app to connect to MCP servers, allowing it to discover and use tools and prompts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*7-TcIoy45xIYCSEY.png" /></figure><p>MCP ensures that multiple agents can process tasks concurrently, allowing scalable AI operations.</p><p>— — —</p><p><strong>Advanced Use Cases</strong></p><p>1. Multi-Agent Collaboration: Different agents perform translation, summarization, and content enrichment.</p><p>2. Scheduled Bulk Processing: Process thousands of tasks asynchronously using MCP.</p><p>3. Real-time Chatbots: Agents respond to user input instantly, coordinated via MCP.</p><p>4. AI-Driven Decision Making: Combine GenAI with business rules for intelligent workflows.</p><p>— — —</p><p><strong>Best Practices</strong></p><p>Error Handling: Ensure agents can retry failed API calls.</p><p>Task Logging: Maintain task history for auditing and debugging.</p><p>Prompt Optimization: Carefully design AI prompts for reliable output.</p><p>Scalability: MCP server allows multiple agents without overloading Mendix.</p><p>— — —</p><p><strong>Conclusion</strong></p><p>By combining Generative AI, Agent Commons, and the MCP Server Module, Mendix developers can create powerful AI-driven applications. Agents can autonomously process tasks, generate content, and collaborate, while MCP orchestrates their workflow efficiently.</p><p>This approach unlocks new possibilities, from automated content creation to intelligent digital assistants, without writing complex infrastructure code.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a53e11f075ad" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Mendix 11: Ushering in the Agentic Era in Low-Code]]></title>
            <link>https://medium.com/@vijaymalviya320/mendix-11-ushering-in-the-agentic-era-in-low-code-889f00dd61d5?source=rss-e58b864b267------2</link>
            <guid isPermaLink="false">https://medium.com/p/889f00dd61d5</guid>
            <dc:creator><![CDATA[Vijay Malviya]]></dc:creator>
            <pubDate>Mon, 06 Oct 2025 17:00:10 GMT</pubDate>
            <atom:updated>2025-10-06T17:00:10.811Z</atom:updated>
            <content:encoded><![CDATA[<p>Mendix version 11 is more than an incremental update — it represents a shift in how low-code platforms handle AI, agents, workflows, and the developer experience. With this release, Mendix pushes further into what it calls the agentic era, bringing more autonomy, smarter tooling, generative AI, and better platform ergonomics. Below, I break down the key features, enhancements, implications, and what to keep in mind.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*F4_LJ5BThi25EmQPrgkUKA.jpeg" /></figure><p>— — —</p><h4><strong>What’s New in Mendix 11</strong></h4><p>Here are the major new capabilities of Mendix 11 (and its subsequent minor versions 11.1, 11.2, 11.3) as of mid-2025:</p><p>1. <strong>Agentic / AI Enhancements</strong></p><ul><li>Agents Kit &amp; Agent Commons Module</li></ul><p>Mendix 11 introduces an “Agents Kit” that brings together existing smart-app components with new “agentic” AI features. The Agent Commons module (evolution of Prompt Management) helps you build agents inside Mendix apps, link microflows/tools, connect to knowledge bases, and iterate on prompts at runtime.</p><ul><li>Model Context Protocol (MCP)</li></ul><p>You can expose tools and prompts from Mendix via MCP, allowing external clients or other LLM-based systems to consume them. This makes the agents more interoperable.</p><ul><li>GenAI Resource Packs (GA)</li></ul><p>Access to curated AI models (e.g. Claude Sonnet, Cohere Embed) and vector databases is now generally available. These are natively integrated in Mendix Cloud and useful for tasks like semantic search, embeddings, RAG (Retrieval Augmented Generation), etc.</p><ul><li>“Start with Maia”</li></ul><p>A major boost to the developer onboarding / scaffolding experience. You provide a textual description (and optionally image or PDF), and Maia helps generate a domain model, basic pages, test data, and a homepage. It helps you bootstrap apps more rapidly.</p><ul><li>Maia for Workflows / Maia for OQL</li></ul><p>Generative-AI assistance in designing workflows from text/image prompts; also assistance in writing &amp; previewing OQL queries (which are used for view entities, etc.).</p><p>2. <strong>Platform &amp; UX / Studio Pro Improvements</strong></p><ul><li>Atlas UI theme refresh (Atlas 4)</li></ul><p>New look &amp; feel: updated templates, new sidebar, more flexible design properties (e.g. full flex support for containers), more CSS variables for easier theming and runtime style changes. Enhancements to layout grids, label widths, form orientation, etc.</p><ul><li>Offline-First Enhancements</li></ul><p>Many-to-many (reference set) associations now supported in offline-first apps. Synchronization improved: “online synchronisation enabled by default” for new entities in offline-first pages.</p><ul><li>View Entities Enhancements &amp; OQL DELETE / UPDATE / INSERT</li></ul><p>Better user experience for view entities with live preview under a specific user. OQL interface improved. Also bulk operations via OQL (delete, update) are being introduced to handle large-scale data operations more efficiently.</p><ul><li>Workflow Improvements</li></ul><p>Workflow interrupting boundary events now GA.</p><p>Workflow Groups: define sets (teams, departments) so tasks can be routed to groups rather than individuals. Dynamic group membership.</p><p>Conflict resolution for workflows: when running instances of workflows become incompatible due to model changes, you can now group conflicting instances and resolve them (continue, restart, abort, jump) en masse.</p><p>3. <strong>Governance, Deployment, Extensibility &amp; Developer Productivity</strong></p><ul><li>Studio Pro Extensibility &amp; UI Improvements</li></ul><p>Enhancements like “Call Hierarchy” view, improved navigation, better search ordering, preferences and theme API for web extensions, access keys in context menus, autofocus configuration per page, etc.</p><ul><li>Version Control / File Storage Changes</li></ul><p>Apps open in 11.0 automatically convert from MPRv1 to MPRv2 format. MPRv2 improves how files are stored (individual document storage) to help with performance, Git integration, and repository size.</p><ul><li>Native Mobile Updates</li></ul><p>Updated React Native version (0.77.3 in 11.2) to improve stability (e.g. fix screen reload issues). Also, mobile build support improvements for PMP (private cloud) customers.</p><ul><li>Widgets &amp; UI Component Enhancements</li></ul><p>Multi-page selection in Data Grid 2 / Gallery (retain selected rows across pages).</p><p>New Check-box / Radio selector widget with flexibility to use for single or multiple selection, static or dynamic contents.</p><p>Combo Box enhanced: filter input event, debounce settings, more control over filtering behavior.</p><ul><li>Other Developer Tools</li></ul><p>MacOS command-line tools for Studio Pro introduced in 11.2.</p><p>More APIs for showing dialogs, accessing user preferences, notifications in web extensions.</p><p>4. <strong>Governance, Portal, and Admin Facilities</strong></p><p>User Classification</p><p>New module to distinguish internal vs external users (for licensing, roles, metrics), based on roles or microflows; existing users can be bulk classified. Helps with licensing accuracy.</p><p>Improved Portal Experience &amp; Search</p><p>Better search across Mendix Portal (Docs, Community, Academy, Marketplace) with smarter categorization, highlights, more compact layout, “Recently Visited” etc. Also improved portal homepage experience.</p><p>Governance over AI usage</p><p>Settings to enable/disable AI help (Maia assistance) in parts of the app (epics, app insights, etc.).</p><p>— — —</p><p><strong>What This Means for Developers &amp; Organizations</strong></p><p>Here are some of the implications — both what’s promising and what to watch out for.</p><p><strong>Pros</strong></p><p>Faster app bootstrapping — “Start with Maia” and AI-assisted tools will reduce setup time and help non-experts or smaller teams get productive quickly.</p><p>More powerful offline &amp; mobile apps — with offline-first improvements and React Native stability, mobile/remote scenarios become more robust.</p><p>Better scalability for data operations — bulk OQL operations, view entity enhancements reduce overhead for large data sets.</p><p>Improved UI/UX and theming — refreshed design system, CSS variables, more layout flexibility help with modernizing look and feel, accessibility, theming, runtime styles, and so on.</p><p>Tighter integration with AI and agents — ability to build agentic apps, integrate with knowledge bases, use external tools (MCP), etc., opens new possibilities (smart assistants, document chat, intelligent workflows).</p><p>Stronger governance — better classification of users, control over AI usage, improvements in version control and storage format, improved tooling for managing conflicts, etc.</p><p><strong>Challenges / Considerations</strong></p><ul><li>Migration &amp; Compatibility</li></ul><p>Projects using older clients (Dojo) need to move to React client, because Dojo is deprecated and will be removed in version 12.</p><p>Modules/widgets/themes might need updates to fully support new Atlas theme enhancements or CSS variable usage.</p><ul><li>Learning curve</li></ul><p>Generative AI tools like Maia and agent features are powerful, but teams will need to adapt to best practices for prompt engineering, knowledge base structuring, AI governance (bias, auditing, cost).</p><ul><li>Performance &amp; Runtime Costs</li></ul><p>Bulk OQL operations, agent tools, knowledge base querying, etc., may have performance implications or cost (especially with cloud AI model usage, training, embeddings). Need to monitor usage and scale accordingly.</p><ul><li>Dependency on Features in Beta / Public Beta</li></ul><p>Some features (Workflow Groups, parts of bulk OQL, etc.) are still in beta or require opt-in. Production readiness may vary.</p><ul><li>Licensing, Governance, and Security</li></ul><p>With more AI and external tools usage, data privacy, licensing (for AI models, knowledge bases), audit trails, security of agents will become more important.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7SKwP1U01VM9RW9Krd0AHg.jpeg" /></figure><p>— — —</p><p><strong>What to Watch &amp; Strategy for Adoption</strong></p><p>Here are a few strategic recommendations if you’re considering Mendix 11 for your team or organization:</p><p>1. Assess whether your apps need to migrate to React client now, especially if you depend on UI client behavior. If you’re still using Dojo, plan migration.</p><p>2. Run pilot projects using “Start with Maia” + Agent Commons to understand how AI/agent integration will affect your work, both technically and from a process/governance standpoint.</p><p>3. Evaluate your offline-first use cases: if your users commonly work offline or with spotty connectivity, test the new offline features (e.g. reference set associations, sync behavior, etc.).</p><p>4. Update UI themes and design systems to leverage Atlas 4 features, including using CSS variables rather than old SASS variables.</p><p>5. Prepare for version control &amp; storage format updates: ensure your source control is ready, that your team understands the MPRv2 format, and test any effects.</p><p>6. Governance &amp; security planning: define who gets access to which AI features, how knowledge bases are maintained, how to monitor AI/agent behavior, cost controls, etc.</p><p>7. Performance testing: especially for large datasets, many concurrent workflow instances, or agent interactions — benchmark to ensure the new features don’t introduce bottlenecks.</p><p>— — —</p><p><strong>Conclusion</strong></p><p>Mendix 11 positions itself as a platform ready for the “agentic era” — combining low-code + generative AI + tooling to build smarter, more autonomous apps without sacrificing control or governance. For developers and enterprise organizations, it offers powerful new tools: better AI integrations, more flexible design systems, stronger offline &amp; mobile support, improved workflows, and enhanced governance.</p><p>If you start planning now — migrating client code, updating themes, piloting agentic use-cases — you can get the benefits early while minimizing friction. Mendix 11 is less about incremental change, more about enabling a new paradigm of what low-code apps can do.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=889f00dd61d5" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>