Dexter

​

Overview

Dexter is ProDex’s embedded AI operations partner. It has direct, sandboxed access to your factory’s data and platform artifacts, and works alongside you in natural language to ingest and transform data, build and run simulation models, configure planning optimizations, analyze results, generate shareable reports, and edit configuration across the platform. Dexter defaults to asking rather than guessing. When something material is ambiguous — a column’s meaning, a modeling assumption, a tradeoff — it surfaces the question rather than picking silently, and it exposes its reasoning as it works so you can audit every step. Each Factory gets its own Dexter with its own memory and its own access boundary. Dexter cannot see across factories and has no open access to the public internet — it works inside your factory’s sandbox. It can, however, reach the external systems your organization has connected: configured integrations let it read from and call your ERP, MES, and databases.

​

The Dexter Page

Dexter is the Home page — the first thing you land on after login. The left sidebar has four top-level sections:

• Chat — your conversations with Dexter, with search and history grouped by date. See Chat, Tasks & Reviews.
• Knowledge Base — what Dexter knows about your factory: its memories and your standing rules. (Uploaded files are not here — they live under Data → My Data; see User Files.)
• Workflows — reusable, runnable procedures you hand to Dexter. See Custom Workflows.
• Projects — Dexter’s longer-running, phased work. See Projects.

While you’re on other pages of the platform, an AI Assistant dock keeps the active conversation available.

​

Starting a conversation

Start with New Conversation or just type in the current one — plain language, no slash commands or special syntax. A fresh conversation opens with quick-start cards that seed common starting points. The input toolbar takes typed prompts and drag-and-drop attachments, and lets you:

• Attach a file (paperclip) — from your machine or reused from User Files.
• Set the conversation type — Chat by default; switch to Project or Workflow to open a Project or run a workflow up front.
• Watch context usage — a live indicator of how full the conversation’s context window is. As it fills, older turns get compacted; a high reading is a good cue to start fresh.

Conversations are independent — each has its own working context and history — but the knowledge base and every platform artifact are shared across all of them.

​

Message actions

Hover a message to act on it:

• Your own messages show when they were sent and — when Dexter saved a checkpoint at that turn — an hourglass that opens a read-only, point-in-time view of your factory as it stood then.
• Dexter’s replies carry copy, thumbs up, and thumbs down. Thumbs are product feedback — your ratings go to the ProDex team to help us improve Dexter; they don’t retrain it or change how it behaves inside your factory. To change how Dexter works for you, set a rule or just tell it directly.

​

Session Context

Dexter always knows what page you’re on and which objects you have selected when you send a message. You don’t have to restate “the Line A model” or “the run I have open” — Dexter reads that from the page.

​

What You’ll See While Dexter Works

• Every tool call is visible and expandable — file reads, scripts, queries, edits. Nothing happens behind the curtain.
• Thinking vs. sub-agents — a “Thought for…” indicator is Dexter reasoning before it answers (expand it to read that reasoning); a sub-agent appears as its own pill when Dexter hands a subtask to a specialized helper (click it to watch that helper’s transcript).
• Long jobs run asynchronously — simulations, planning runs, and other long tasks kick off in the background and report back when they finish.

Alongside the reply stream, Dexter keeps a live task list, shares review briefs before involved work, drops action cards for steps that live on another page, and asks structured questions as clickable cards — with generated files and the task list collecting in the right-hand Artifacts panel. That’s the full interaction surface; see Chat, Tasks & Reviews. Higher-stakes approvals run through Gates.

​

Attachments

Drag files into the input or use Attach. Dexter reads tabular data, documents, images, and CAD files. It profiles the data it inspects — sampling large files and disclosing the scope of what it read rather than implying it read every row — and asks before assuming what a column or code means. Uploads persist and are managed under Data → My Data; see User Files for formats, limits, and lifecycle.

​

Capabilities at a Glance

• Data — profiles uploads and builds documented pipelines that trace every derived value back to its source. See User Files.
• Simulate — creates and edits every simulation object, runs models, and explains throughput, utilization, and bottlenecks. See Simulation.
• Plan — configures and runs the planning optimizer end-to-end. See Planning.
• Insights — builds free-form, factory-scoped dashboards. See Insights.
• What-if — compares scenarios with Experiments and Snapshots, and characterizes variability with Monte Carlo.
• BOMs & CAD — builds BOMs, explores the graph, drives the configurator, and analyzes CAD.
• Reports — authors polished, shareable PDF, Word, PowerPoint, and Excel deliverables. See Reports.

​

Building Blocks

Each concept has its own page — the short version:

• Memories — Dexter’s structured, per-factory knowledge base of your operation, your data sources, and the artifacts it has built. A read-only organization layer sits alongside it.
• Rules & Preferences — your standing instructions for how Dexter works in this factory, applied in every conversation.
• User Files — everything you’ve uploaded, managed under Data → My Data and readable in any conversation.
• Reports — polished deliverables Dexter generates on request, delivered in chat and in the relevant project’s phase folder.
• Scheduled Jobs — recurring or one-off tasks Dexter runs on a schedule; each run opens as its own conversation.
• Custom Workflows — reusable procedures you and Dexter define once and trigger by name later.
• Projects — durable, phased containers for multi-session work.

​

Limits

• No open web access. Dexter doesn’t browse the public internet. Through configured integrations it can read from and query the external systems your organization has connected (ERP, MES, databases), but nothing beyond those.
• Sandboxed to a single factory — one factory per conversation, no cross-factory visibility.
• Changes land immediately. There’s no global undo button — the quickest way to roll back something Dexter just did is to ask it to revert that change.

​

Best Practices

• Name specific entities (“the CNC station on Line B”, “the July demand plan”) rather than “the thing we looked at earlier.”
• Let Dexter ask. Answers to its questions are recorded once and reused — a five-minute Q&A up front saves rework later.
• Approve or push back on review briefs. They’re the point where Dexter is asking you to catch anything wrong before it becomes commitment.
• Iterate. Short, focused turns beat one giant prompt.
• Audit Memories and Rules periodically. They shape every future response — keep them accurate and prune what’s outdated.