Dexter

​

Overview

Dexter is ProDex’s AI assistant — an operational agent built into the platform with direct access to your factory data. It can ingest and transform your operational data, build simulation models, run analyses, generate reports, and make changes to your configuration, all through natural-language conversation. Dexter is not a general-purpose chatbot. It works with your actual data inside a secure, per-factory sandbox. When you ask “show me the bottleneck in Line 3,” Dexter writes a query against the simulation’s raw event data, reads the results, and reports back with numbers. When you upload a messy MES export, it profiles the data, asks you to clarify what the status codes mean before assuming, and traces every derived value back to its source. Dexter defaults to asking rather than guessing. At major decision points it presents structured questions with clear options; when interpreting your data it flags anything it’s uncertain about — ambiguous column names, unlabeled units, codes without a legend — and asks you to confirm before building anything.

​

Page Layout

The Dexter page is the first thing you see after logging in. It is organized into three panels.

​

Left Panel — Conversations & Settings

The left sidebar is your control center for managing Dexter’s context and configuration. At the top you’ll see the Dexter heading and a Collapse panel button to hide the sidebar and give the chat area more room. Below the heading are seven navigation items that open full-page views in the center panel:

• Memories
• User Files
• Reports
• Scheduled Jobs
• Custom Workflows
• Integrations
• Settings

Each of these is described in detail in its own section below. Further down the sidebar:

• + New Conversation — start a fresh conversation at any time.
• Search conversations — search across all your conversations by keyword.
• Conversation history — conversations are grouped by date (Today, Yesterday, etc.). Each entry shows a status indicator (Idle or In Progress), the conversation title, and a three-dot menu with Rename and Delete options.

​

Center Panel — Chat

This is where you interact with Dexter. When you open a new conversation, the center panel displays a welcome prompt: “How can Dexter help you today?” Below the welcome message are three quick-action buttons that route you into common workflows:

• Start with data — attach or upload data files and begin a data-driven workflow. Dexter profiles the data, asks clarifying questions, and proposes building a documented pipeline to trace every derived value.
• Make sim model — start building a simulation model from scratch. Dexter investigates your process (what flows through the line, constraints, routing logic), asks about your goals (what question the model should answer), then enters plan mode and builds a discrete-event simulation model.
• Run an experiment — set up a what-if comparison across model snapshots. Dexter helps define what varies across cases, what metrics to compare, then captures snapshots, runs them, and analyzes the differences.

​

Input Toolbar

At the bottom of the chat area is the message input with a toolbar:

• Text input — type your message or question in plain language. You can also drag and drop files directly onto the chat area to attach them.
• Attach (paperclip icon) — attach files from your computer.
• Plan mode toggle (clipboard icon) — turn plan mode on before sending a message.
• Todo (checklist icon) — view or manage the task list for the current conversation.
• View current plan (eye icon) — open the active plan document (enabled only when a plan exists).
• Follow agent navigation (compass icon) — when toggled on, the platform automatically navigates to whatever page Dexter is working on (e.g., Modeler, Results). Toggle off to stay where you are while Dexter works.
• Submit (enter icon) — send your message.

​

Right Panel — Artifacts

The Artifacts panel tracks outputs from the current conversation. It has three sections:

• Files — generated files (reports, exports, data outputs) appear here as Dexter creates them. Click to download or preview.
• Todos — when Dexter breaks complex work into steps, each step appears as a trackable task with a status (pending, in progress, completed). You can see progress against the overall plan.
• Plan — when plan mode is active, the full plan document is displayed here. Review it, then confirm or ask for changes before Dexter proceeds.

The panel shows a count badge next to each section header. Click the Collapse artifacts button at the top right to hide the panel.

---

​

Memories

Dexter’s persistent knowledge base — operational rules, corrections, naming conventions, and factory context that carry across all conversations. Dexter writes memories automatically and curates them actively, building two layers: per-factory workspace memories and read-only organization memories shared across your team. Full documentation →

---

​

User Files

Every file you attach in chat is saved so Dexter can reference it in any future conversation. Supports spreadsheets, PDFs, documents, images, and data files. Full documentation →

---

​

Reports

All Dexter-generated outputs — PDFs, Excel spreadsheets, PowerPoint decks — collected in one place. Ask Dexter to generate a report at any time, or let it suggest one after completing analysis. Full documentation →

---

​

Scheduled Jobs

Run Dexter tasks on a recurring or one-off schedule. Tell Dexter what to automate and when in plain language — it handles the scheduling syntax, creates a new conversation each time the job fires, and saves the results for your review. Full documentation →

---

​

Custom Workflows

Reusable procedures that turn your best practices into executable definitions. Walk through a process with Dexter once, save it as a workflow, and any team member can trigger it by name. Workflows can also be referenced by scheduled jobs for recurring execution. Full documentation →

---

​

Integrations

Integrations allow Dexter to connect to your ERP, MES, or database systems so it can query live factory data directly. Once an integration is configured, Dexter can access that data in any conversation. Contact the ProDex team to set up an integration for your organization.

---

​

Settings

The Settings panel lets you configure Dexter preferences:

• Desktop notifications — toggle on to receive browser notifications when Dexter finishes a task or needs your input while you’re outside the app.

---

​

What Dexter Can Do

​

Understand Your Operation From Data

Upload production logs, MES exports, inventory snapshots, or any operational data. Dexter profiles the file — columns, data types, value distributions, row counts — and asks you to clarify anything ambiguous before proceeding. When the work goes beyond simple inspection, Dexter creates a documented pipeline that traces every derived value back to its source data: what file it came from, what query produced it, what assumptions were made, and how confident the result is. When you upload fresh data next week, the same pipeline can be re-executed to produce updated outputs without rebuilding anything.

​

Model and Simulate Your Process

A simulation model maps your physical operation — machines become stations, conveyors become buffers, workers become resources — and Dexter runs discrete-event simulation to predict throughput, utilization, and bottlenecks. Dexter can:

• Create and edit model components — sources, sinks, stations, processes, resources, buffers, routers, combiners, separators, transformers, model nodes, and connections
• Define KPIs and charts for your results dashboard — each backed by a precise, auditable query against the simulation data
• Set up constants, lookup tables, and entity definitions
• Build schedules that define material releases, shift patterns, and capacity changes
• Execute simulations and explain results — throughput, utilization, cycle times, bottlenecks
• Compare runs side by side and explain what changed

​

Analyze Data with Insights

Dexter can create free-form data visualizations and exploratory analyses on the Insights page — factory-scoped dashboards you can return to at any time. This is designed for open-ended exploration beyond what pipelines and simulation results provide.

​

Plan Production

Dexter can configure and run the planning optimizer — a MILP-backed solver that takes demand orders, current inventory, your bill of materials, and resource constraints and produces a feasible production plan. Describe the planning problem in plain language and Dexter will set up the inputs, run the solver, and explain the result.

​

Run What-If Scenarios

• Configure Experiments with multiple Snapshots for side-by-side comparison
• Run Monte Carlo batches — hundreds of simulation replications to quantify variability, build confidence intervals, and identify which metrics are stable vs. sensitive to randomness
• Test parameter changes and report the impact
• Summarize experiment results and highlight key findings

​

Work With BOMs

Dexter can help you manage your bills of materials — create BOMs, explore the BOM graph structure (what goes into a product and where a component is used), and keep your material definitions up to date.

​

Configure Products From Orders

For configure-to-order or engineer-to-order operations, Dexter can analyze an order document (PDF, spreadsheet, or other format) against a configuration template and produce a configured BOM. Configuration templates define option classes that capture a product’s full possibility space; Dexter walks the order through them — selecting options, assigning materials, gating each decision for your review — and explains its choices at every step.

​

Generate Reports

See Reports for supported formats and how to create them.

​

Automate Recurring Work

See Scheduled Jobs for how to set up recurring and one-off tasks.

---

​

Starting a Conversation

Click + New Conversation in the left sidebar, or simply start typing in the input area. You can have multiple conversations open — each one maintains its own context and history. Dexter’s memories are shared across all conversations, so teaching Dexter something in one conversation carries forward into every other one. However, the thread of analysis — tool calls, intermediate results, and conversation context — stays within each conversation. You don’t need to use specific commands or syntax. Start by describing what you want in plain language:

• “What’s the throughput of my current model?”
• “Add a buffer between the assembly process and the paint station with capacity 50”
• “Run the simulation and show me which resources are over 90% utilization”
• “Compare the last two runs and tell me what improved”
• “Upload this MES export and derive cycle times by station”
• “Run 100 replications and tell me how confident we can be in the throughput number”

---

​

What to Expect in a Conversation

​

Tool Calls

Every action Dexter takes shows up in the conversation — file reads, writes, command executions, queries. You can expand each tool call to see exactly what Dexter did and why. This makes Dexter’s work fully auditable: you can trace any result back through the steps that produced it.

​

Structured Questions

When Dexter needs your input — picking between modeling approaches, confirming a column’s meaning, choosing a snapshot to compare — it presents clickable options instead of asking you to type a free-form answer. These are checkpoints that prevent wrong assumptions from entering your models.

​

Plan Mode

For multi-step changes (building a simulation model end-to-end, designing a data pipeline, restructuring a planning setup), Dexter drafts an implementation plan before producing any artifacts. This is the “measure twice, cut once” mechanism. How it works:

1. Dexter (or you, via the plan mode toggle in the input toolbar) enters plan mode.
2. Dexter writes a structured plan — what it proposes to build, the modeling decisions and their rationale, what’s in scope and what’s deferred.
3. The plan appears in the Plan section of the Artifacts panel on the right.
4. You review and either approve (Dexter begins implementing, breaking the plan into tracked tasks) or request changes (Dexter iterates on the plan).
5. No artifacts are authored until the plan is approved.

Dexter enters plan mode automatically for substantial work. For simple single-step tasks (adjusting a parameter, renaming something), it skips the plan and acts directly. You can override Dexter’s judgment in either direction using the toolbar toggle.

​

Todos

When Dexter breaks complex work into steps, each step appears as a task in the Todos section of the Artifacts panel. Tasks have a status (pending, in progress, completed) and are created after plan approval or during multi-stage work. This gives you a clear view of progress against the overall plan.

​

Thinking and Processing

You’ll see “Thought for X seconds” indicators when Dexter is reasoning about your request. Tool call summaries show what Dexter is doing in real time (reading files, scanning workflows, running scripts). Simulations run asynchronously and may take a minute or more; Monte Carlo batches with many replications take longer.

---

​

File Attachments

You can attach files to your messages — PDFs, CSVs, Excel spreadsheets, PowerPoint decks, Word documents, images, JSON, and YAML. Drag and drop files onto the chat area, or click the Attach button in the input toolbar. Common uses:

• Uploading production data for analysis or model parameterization
• Sharing reference documents — SOPs, equipment manuals, team decks
• Providing spreadsheets to import into your model
• Uploading order documents for BOM configuration
• Capturing data from dashboards or systems without a direct integration — screenshot a BI view or export a report, and Dexter can work with it

Uploaded files persist across conversations. Once a file is in your factory’s uploads, you can reference it in any future conversation without re-uploading. Manage your uploaded files through the User Files section in the left sidebar.

---

​

What Dexter Can’t Do

• No external internet access. Dexter cannot browse the web, call external APIs, or access systems outside the platform. If you ask a general knowledge question, Dexter will redirect toward operational work it can help with.
• Sandbox-scoped data. Dexter can only access data within your factory workspace and files you upload. It cannot see other factories or reach outside the per-factory sandbox.
• One factory per conversation. Each Dexter conversation is scoped to a single factory. To work on a different factory, switch factories and start a new conversation.
• No automatic undo. Changes Dexter makes to your model, entities, BOMs, or other artifacts take effect immediately. There is no single “undo” button — use Snapshots to preserve a known-good state before substantial changes.
• Conversation context has limits. Very long conversations may lose early context. The persistent knowledge base and todos compensate, but for best results start new conversations for distinct topics rather than running everything in a single long thread.

---

​

Best Practices

• Reference specific entities by name. “Show me utilization” is good. “Show me utilization for the CNC machines over the last 5 runs” is better — naming specific models, schedules, or entities helps Dexter act faster and more precisely.
• Let Dexter ask questions. When Dexter asks clarifying questions, it’s building checkpoints that prevent wrong assumptions from entering your models. Responding to these produces better results than trying to specify everything upfront.
• Answer data questions once. When uploading data, Dexter will ask about column meanings, units, codes, and edge cases. These answers go into the knowledge base, so the same questions don’t come up again on the next upload.
• Use plan mode for big changes. Toggle plan mode on before asking Dexter to make substantial modifications. Reviewing the plan first is faster than undoing mistakes.
• Work iteratively. Start with a question, review the answer, then refine. Dexter maintains full context within a conversation, so follow-up questions build on what came before.
• Audit what Dexter remembers. After teaching Dexter something new, check the Memories panel or ask “What do you know about X?” to verify the entry was saved correctly. Corrections persist across all future conversations.