Logic Fabric Agents
Overview
Logic Fabric Agents are a suite of AI agents that work directly on the logic you build with Logic Fabric on Microsoft Dataverse. One agent authors that logic from natural language; the rest read it and turn it into something you can act on: enterprise documentation, governance reports, process intelligence, and test suites. Each agent produces one specific, high-value deliverable.
Every agent connects to your environment through the Logic Fabric MCP server and acts on behalf of the signed-in user, combining deterministic work with large-language-model reasoning, and each ships as a self-contained container with its own web application. Logic Authoring is the one agent that writes, and only ever with your approval; the four reading agents are strictly read-only and never change your project, which makes them safe to point at production.
The five agents
| Agent | What it produces | Use it to |
|---|---|---|
| Logic Authoring | Grounded, validated Logic Fabric artifacts, created live from natural language | Build artifacts, one or a whole suite, by describing them |
| Logic Insights | Process intelligence: health scores, metrics, simulation, standards audits, and AI redesign | Understand how your processes really perform, and improve them with evidence |
| Logic Context | Enterprise Business Requirements Documents in plain business language | Document a project for handover, audit, or onboarding, on demand |
| Logic Guard | A governance report: per-artifact issues with severity and suggested fixes | Enforce a consistent quality bar before you publish |
| Logic Assurance | Schema-accurate test suites grounded in live Dataverse metadata | Protect your logic from regressions when it changes |
Each agent is licensed and shipped independently. Install only the ones you need. They do not depend on each other.
A shared foundation
All five agents are built on the same platform and follow the same shape, so once you know one, you know the pattern.
- Connect. Each agent authenticates the incoming user (Entra ID or ADFS) and calls the Logic Fabric MCP server on behalf of that user, so it sees exactly the projects and data the user is allowed to see.
- Read (and, for Logic Authoring, write). Each agent reads the project's artifacts through the MCP tools. The four reading agents never write back; Logic Authoring additionally creates and updates artifacts through the MCP write tools, always on your approval.
- Analyze. A fleet of specialist workers does the job. Deterministic work such as metrics, rule checks, and metadata validation runs in code. Language work such as prose, findings, recommendations, and test cases runs through a configurable model.
- Stream. Results stream back to the browser over Server-Sent Events as each part completes, so long-running jobs render progressively instead of blocking.
- Persist. Results and per-project configuration are stored in Redis in production, or in memory for local evaluation.
What they have in common
- One image, one endpoint per agent. Each agent is a single container that serves both its REST API and its web application on port 8080, with a health probe at
/health. - Per-project configuration. Every agent has tunable, per-project settings stored as sparse overrides. You change only what you need, and everything else keeps tracking product defaults across releases.
- Provider-flexible reasoning. Claude on Azure AI Foundry, Azure OpenAI, OpenAI, or Anthropic, chosen by configuration.
- Secure by default. Incoming requests require authentication, and downstream calls to the MCP server use the on-behalf-of flow, so the agent acts as the user, not as a shared service identity.
- Read-only reading agents; approved writes for authoring. The four reading agents call only the MCP read tools and never change your project. Logic Authoring writes, but only the artifacts you have approved, through Logic Assist, with managed-solution protection still enforced.
Getting started
- Read the agent pages above to decide what you need.
- Follow the Installation guide to run the container images, connect them to your Logic Fabric MCP server and a model, and provision Redis and an Entra ID app registration for production.