← All posts 中文版

PMClite in Full: The Complete Shape of a PMC Workbench

The first two articles dissected the scheduling algorithm and the DeepSeek integration. This one steps back: what PMClite actually is, which functions it covers, why it is different from existing options, and which scenarios it does not fit.

Stack Tauri + Rust Database SQLCipher Roles 6 Modules 9

1. Positioning: One Sentence and Three Boundaries

In one sentence: PMClite is a desktop-grade production management workbench for PMC roles in small and medium manufacturers. It is centered on scheduling and connects the full chain of "sales — production — purchasing — inventory — finance."

But this positioning has three boundaries that must be stated clearly; otherwise it is easily misunderstood as a "lightweight ERP" or a "domestic APS."

Boundary 1: Its user is the PMC, not the boss

The software has an explicit role-permission matrix (ROLE_PERMISSIONS) with six roles:

RoleVisible modules
AdministratorAll 9 modules
SalesDashboard, materials, products, sales, inventory
PurchasingDashboard, materials, products, purchasing, inventory
WarehouseDashboard, materials, products, inventory, purchasing, production
Shop floorDashboard, materials, products, inventory, production
FinanceDashboard, materials, products, inventory, sales, purchasing, finance

The production module is only visible to administrators, warehouse, and shop floor. The finance module is only visible to administrators and finance. This is not a missing feature; it is a deliberate role split — one tool serves six roles, but each role sees the screen it should see.

And the first screen of the production module is the "decision desk," not the "dashboard." The decision desk gives you the queue of events that must be handled today (shortages, delivery risks, capacity overloads, pending report confirmations), each with an owner, SLA, and escalation button. This is for the PMC, not a BI dashboard for the boss.

Boundary 2: It is a desktop application, not cloud SaaS

The tech stack is Tauri + SQLCipher:

Front end: single HTML file + native JS (no framework)
Back end: Rust (Tauri host)
Storage: SQLCipher encrypted database (local file)

This choice has several clear consequences:

The corresponding costs are equally clear: no public internet access, no automatic upgrades, multi-PC collaboration requires firewall configuration, and data safety depends on the user backing up.

Boundary 3: It is an "explainable scheduling tool," not a "black-box optimizer"

The previous two articles covered this thoroughly. Here we only emphasize the conclusion: PMClite's scheduling chain is "master data completeness check → constraint filtering → health check → priority sorting → daily line calendar consumption." Every step can be inspected, intervened in, and recomputed. AI only scores and attributes; it does not compute dates.

It does not pursue "computing the theoretical optimum." It pursues "computing a plan the PMC can understand, modify, and take responsibility for."

2. Functional Panorama: Viewed Through One Business Chain

PMClite's nine modules are not a parallel feature list; they are a directed chain:

Material master data ─┐
                      ├→ BOM / routing ─┐
Product master data ──┘                  ├→ Inventory ─┐
                                         │             ├→ Production (scheduling) ─┐
Sales orders ────────────────────────────┘             │                            ├→ Finance (reconciliation)
                                                       │                            │
Purchase orders ←── MRP ←──────────────────────────────┘                            │
   │                                                                                  │
   └→ Receiving into stock ───────────────────────────────────────────────────────────┘

2.1 Materials: the first brick of master data

Supports image URLs (online/local/relative paths). Hover to see a larger image in the table; thumbnails are lazy-loaded (_lazyShowTooltip only injects an <img> on hover, avoiding 200 images being rendered for a 100-row material screen).

Import/export goes through Excel, but does several things beyond "export then import":

2.2 Products: the entry point to the three-level precision model

Each row in the product list carries several key pieces of information:

The product edit modal has a "production mode" dropdown with six modes (fixed takt / flexible takt / single station / batch / continuous / project), each with industry examples and formula explanations. After selecting a mode, the fields switch dynamically (_onProdModeChange), e.g. flexible takt asks for "single-worker minutes/unit," single station asks for "single-unit cycle minutes."

There is also a flowType field (flow line / batch), which determines whether a multi-process product advances in parallel or serially as a whole batch — this choice directly determines which branch the scheduling algorithm takes.

2.3 BOM: tree view + routing dual view

The BOM page has inner tabs: Bill of Materials and Routing.

Bill of Materials: multi-level tree, indentation indicates hierarchy, semi-finished products can expand sub-BOMs, with circular reference detection (visited Set backtracking). Each product row has per-product export/import buttons on the right, affecting only that product and its branch.

Routing: this is where processes are configured for L2/L2.5 precision products. Each row fills sequence, process name, work center, per-unit time, setup time, post-wait. Sequence must be unique (validated on save); 10/20/30 increments are recommended for easy insertion.

Advanced tools (in a collapsible panel): batch BOM import/export with three modes — append / full replacement / range replacement. Full replacement triggers a second confirmation explicitly stating "all existing BOMs will be cleared." A JSON backup is automatically triggered before import.

AI process inference: covered in detail earlier; here just one sentence — clicking "AI infer" or "AI infer V2" recommends a process list based on BOM materials, line mode, and historical data, with confidence-tiered checkboxes and fallback to the rule engine on failure.

2.4 Inventory: more than just quantities

The inventory page has two sub-tabs: Inventory Detail and Material Requisitions.

Inventory Detail:

There is a detail in the data source for slow-moving analysis: it prefers the back-end aggregated _inventoryLastChange (last change time per material, unaffected by log downsampling), and only falls back to scanning inventoryLogs.

Material Requisitions: this is a distinctive module.

This flow matches the most real scenario in small factories: print the requisition with a QR code, the warehouse scans it with a scanner, the material list pops up on screen, actual quantities are entered, the recipient signs by hand, confirm, and inventory is automatically deducted.

2.5 Sales: sales + PMC dual perspective

The sales order list design is distinctive: the same row shows both sales-relevant and PMC-relevant information:

Sales perspectivePMC perspective
Order amount, customer, delivery statusProduction status (unscheduled / scheduled / in production / in production with shortage / production complete)
Invoicing status, invoice number, payment status, payment amountEstimated completion date (from the linked work order's scheduledEnd)
Gross profit, gross marginWhether inventory is sufficient for shipment
Whether customer credit limit is exceededWhether delivery is overdue / about to be due

The top has an urgent reminder bar (N overdue, N due soon, N shortage, N pending shipment, N unscheduled) and KPI cards (order count, total order amount, total gross profit, overdue orders, unscheduled).

Quick filter buttons: all / overdue / due soon / shortage / unscheduled / pending shipment; click to switch views.

Gross profit is calculated with getCachedBOMCost(productId); each product's BOM cost is computed only once and cached in _bomCostCache. Production status uses the pre-built index woBySalesOrder (one O(N) pass to build a Map), avoiding repeated filters in a loop.

Customer profiles: credit limit, payment terms, status (normal / warning / frozen), current debt. Limit utilization above 90% is marked red.

Statements: generated by customer + date range, previewable in a modal, printable (separate window with company header, both-party seal areas, debt summary), or exportable to Excel.

2.6 Purchasing: MRP + in-transit tracking

MRP is the core of the purchasing page. The logic of calculateMRP():

  1. Iterate all in-progress / pending production plans, recursively expand BOMs (semi-finished stock is used first; the insufficient portion is expanded to lower levels)
  2. Accumulate total demand per material + earliest demand date
  3. Iterate pending-shipment sales orders, deduct quantities already covered by work orders, yielding net demand
  4. Deduct current inventory + in-transit quantity (remainingQuantity > 0 in purchase orders)
  5. Add safety stock (total demand × safetyStockRatio)
  6. Suggested purchase quantity = total demand − inventory − in-transit + safety stock

The result table has checkboxes; after selecting, click "one-click order" to automatically generate purchase orders, with supplier and unit price pulled from material master data.

Purchase orders: the state machine is pending approval → approved → ordered → partially received → completed, cancellable. Each material row has receivedQuantity and remainingQuantity; receiving updates inventory + reduces in-transit.

Receiving notes: a separate print template with QC checkboxes (qualified / unqualified / concession / pending inspection), three-party signature areas (deliverer / receiver / warehouse supervisor), and order total / cumulative received / remaining in-transit summary. Three copies: warehouse, finance, supplier.

Supplier profiles: automatically summarized from purchase orders, showing contact, phone, payment terms, material count, in-transit order count, on-time rate (simplified calculation based on "actual receipt date vs expected delivery date"). Click details to see historical orders.

2.7 Production: decision desk + comprehensive control

The production page has two sub-tabs: Decision Desk and Comprehensive Control.

Decision Desk (renderPmcDecision):

Comprehensive Control (renderProductionPMC):

Each row in the work order table supports 8 action buttons: operation center, edit, sandbox simulation, print, dispatch, complete, requisition, kit details, delete. This is the PMC's daily main battlefield.

PMC rules panel (collapsible): shows four core rules — priority-driven, capacity constraint, lead time and buffer, multi-level kit check. Each comes with a formula explanation so the PMC understands how the system calculates.

Schedule health check: one click outputs a three-tier list of blocking / warning / suggestion items, covering line configuration, product parameters, process definitions, BOM reference integrity, and work order status. The health check itself does not block scheduling; it only gives advance notice.

2.8 Finance: reconciliation + cash flow + profit

The finance page is a pure summary view; it does not perform new business operations:

The statement print template has its own style (getFinancePrintStyle), more formal than general printing: document title, statement number, period, counterparty info, detail table, large outstanding amount box, both-party seal areas, footer. This is something finance can actually use.

2.9 Settings: a configuration center more complex than feature pages

The settings page is grouped by sub-tabs (_setupSettingsTabs):

GroupIncluded cards
🏢 Basic configurationCompany info, message notifications, AI scheduling, industry configuration
⚙️ Production parametersPMC rules, constraint policy, scheduling parameter diagnostics, scheduling conversion self-check
🔄 System syncRecycle bin, LAN sync, mobile reporting policy, mobile access entry
👥 Users and dataUser management, audit log, data backup, data archiving, data management

Company info: name, abbreviation, address, phone, fax, email, tax number, bank, account number, logo URL. After saving, all document print headers update uniformly; preview available.

Message notifications: configure DingTalk/WeCom group robot webhooks, multiple channels can be added, each channel can set a "push level" (≥ warning / urgent only), with a test send button.

Industry wizard (_openIndustryWizard): choose an industry (electronics / machining / injection molding / heat treatment / apparel), answer 3–5 questions (how many lines, stations per line, hours per day, per-unit cycle, changeover time), automatically configure line resourceParams and product default productionParams, and set default constraint policies.

Line table: configure each line's production mode, parallel stations, parallel machines, changeover time, daily hours, with a live "daily available hours" display. There is a "learn changeover time" button that samples the median of adjacent changeover intervals from historical work orders.

Constraint policy: four constraint categories (material kit / delivery / changeover / batch merging) × three levels (soft / hard / off). Default is all soft, with the note "all-hard constraints tend to deadlock the schedule."

Reporting policy: three levels — only anomalous reports require confirmation (±10% deviation generates an event), all reports require confirmation, or no confirmation and direct commit.

Mobile access entry: shows the local IP and port 8766 access address, generates a QR code, copyable and pushable to groups. Scan with a phone to open the reporting page; browser menu "add to home screen" makes it a standalone app.

LAN sync: local role (standalone / host / client), nearby device list (auto-discovery), select host, sync diagnostics (test connectivity of each interface), reset sync state, firewall self-test (when unreachable, pops up PowerShell command guidance).

User management: add/edit/change password/delete users, six roles, enabled/disabled status.

Audit log: records eight operation types: login, logout, add, modify, delete, inbound, outbound, price change, with operator, object, object ID, and field-level change detail (beforeData → afterData). Filterable, cleanable for data older than 90 days.

Data backup: back up database (.db file), restore database (automatically backs up current data to .pre_restore first), full backup (database + JSON + license file).

Data archiving: two types — archive completed documents by day (work orders / sales / purchasing / requisitions, 180 days default), archive closed events by retention period (90 days default, aggregated into statistics then detail deleted). Automatic backup before archiving.

3. Six Core Differentiators

3.1 Three-level precision model: no need to "get it perfect in one step"

Most scheduling tools require master data to be "complete" before use — products need labor hours, routings, and BOMs. PMClite tiers it:

Every level can schedule; only precision differs. A live badge on the Gantt chart tells the PMC whether it is currently "precision" or "coarse."

The value of this design: new users can start at L0, fill in data as they go, and do not have to wait for complete master data.
Existing users naturally upgrade to L2.5 as their data fills out, without changing tools.

The companion "precision upgrade suggestions" (_detectUpgradeSuggestions) proactively prompt: a product's last 3 orders have an average deviation >30% and it is still L0 — suggest configuring the routing; a product is L2 but deviation >15% and no post-wait is configured — suggest adding post-wait.

3.2 Six production modes unified: one engine eats all industries

Covered earlier; here we only emphasize industry span:

ModeOne sentenceTypical industries
Fixed taktOne unit every N secondsSMT, filling, tableting
Flexible taktSingle-worker time ÷ stationsManual assembly, apparel
Single stationPer-unit cycle ÷ machinesInjection molding, stamping, CNC
BatchBatch-based calculationHeat treatment, electroplating, painting
ContinuousQuantity ÷ flow rateChemicals, paper
ProjectQuantity × daysMolds, non-standard equipment

All industry differences converge into one function, calcResourceHours(). Changing industries does not change code, only parameters. This is the foundation for PMClite being a "general PMC tool" rather than an "industry-specific tool."

3.3 Explainable, intervenable, traceable

These three words run through all features:

Explainable:

Intervenable:

Traceable:

3.4 Full chain connected, but not a "big ERP"

PMClite's chain is: sales order → production work order → MRP → purchase order → receiving into stock → inventory → material requisition → completion into stock → sales shipment → finance reconciliation.

Every link carries data forward:

But the difference from a big ERP is: no general ledger, no cost accounting, no fixed assets, no HR, no multi-organization/multi-ledger. It only does the core chain of a manufacturing enterprise, not the periphery.

3.5 Print system: documents are the real deliverables in a factory

PMClite has 8 built-in print templates:

DocumentCharacteristics
Production work order / dispatch orderBOM kit status, material demand list, three-party signature areas
Material requisitionQR code, electronic signature, three-party signature areas
Sales orderCompany header, customer info, both-party signature areas
Delivery noteShipping address, both-party signature areas, two copies
Purchase orderSupplier info, in-transit column, order total
Receiving note / inbound noteQC checkboxes, three-party signatures, summary row
Customer statementSeparate formal style, large outstanding amount, both-party seal areas
Supplier statementSame as above, different fields

All documents share _buildCompanyHeaderHTML; company info changes update everything at once.

Why talk about printing separately? Because in small and medium factories, paper documents are the real work vouchers. No matter how fancy the software looks, what the operator holds in the end is a piece of paper. PMClite's print templates are usable formal documents, not "export and look" attachments.

3.6 Desktop + LAN collaboration: balancing data sovereignty and convenience

This architecture solves a concrete contradiction: small factories want data in their own hands (not on the cloud), but also need multiple computers to collaborate (not just one).

PMClite's approach is "host / client" mode:

The sync mechanism is incremental push + incremental pull:

There is also mobile reporting (port 8766): a phone on the same Wi-Fi scans a QR code to open the reporting page; report data is pushed to the host in real time, and the host shows a toast + plays a sound + flashes the title. After "add to home screen" in the mobile browser, it becomes a standalone icon.

This solution is not elegant (requires firewall configuration, same subnet), but it lets "data does not leave the factory" and "multi-person collaboration" coexist without choosing one.

4. Comparison with Existing Options

This section does not rank competitors; it only explains by scenario "when PMClite is more suitable and when it is not."

4.1 vs large ERP (SAP, Kingdee, Yonyou, etc.)

Big ERP strengths: general ledger, cost accounting, multi-organization, multi-currency, supply chain collaboration, compliance auditing, industry depth.

PMClite differences:

DimensionBig ERPPMClite
Implementation time3–12 months1–3 days
Implementation costHundreds of thousands to millionsOne-time purchase
Scheduling capabilityGeneric module, poor industry fit, often needs custom developmentCore capability, six modes out of the box
Operational complexity2+ weeks of trainingHalf a day to get started
ExplainabilityScheduling is a black box; hard to trace problemsEvery step inspectable and intervenable
FlexibilityRequirement changes need developmentParameter configuration, prompts adjustable
Data ownershipPrivate deployment is expensive; SaaS data is on vendor serversLocal encrypted database

Scenarios suited to big ERP: revenue above 500 million, multiple factories and organizations, consolidated reporting needed, dedicated IT and finance teams, audit and IPO compliance needed.

Scenarios suited to PMClite: revenue 50 million to 500 million, single factory or simple multi-workshop, PMC 1–2 people, the boss wants "can we deliver on time this week" rather than "this month's cost variance analysis."

4.2 vs professional APS tools

Professional APS strengths: mathematical optimization (constraint solving, genetic algorithms), complex constraint modeling, large-scale scheduling, simulation optimization.

PMClite differences:

Scenarios suited to APS: large batch, many varieties, extremely complex constraints (e.g. semiconductors, automotive OEMs), an IE team maintaining master data, an IT team doing integration.

Scenarios suited to PMClite: small and medium batches, many varieties, imperfect master data, PMC mainly relies on experience, need fast response to changes.

A concrete difference: APS tells you "the optimal schedule is X" — whether you believe it is another matter. PMClite tells you "scheduled by priority and capacity, this is the result, and here are each work order's score and reason" — you can disagree, modify, and rerun. In small factories, a discussable schedule is more useful than an optimal one.

4.3 vs Excel

This is the most common "competitor." Many small factory PMCs currently use Excel for scheduling — one Gantt sheet, manually filling colors.

Excel strengths: free, familiar, zero cost, change whatever you want.

Excel problems:

PMClite differences: unified data, deterministic calculation, traceable operations, multi-person collaboration, role-based permissions, built-in validation.

But PMClite retains Excel's "familiarity": materials, products, inventory, sales, and purchasing all support Excel import/export; the format is just a table — edit and re-import. BOM also has a "batch paste" entry, with each row code,quantity,type.

This is not a compromise; it is deliberate transitional design — when users migrate from Excel, they do not need to change their work habits, only replace "manual maintenance" with "system maintenance."

4.4 vs cloud SaaS scheduling tools

Cloud SaaS strengths: no deployment, automatic upgrades, public internet access, multi-device sync, pay-as-you-go.

PMClite differences:

Scenarios suited to cloud SaaS: multi-location offices, boss wants to check on a phone anytime, no IT maintenance capability, accepts data in the cloud.

Scenarios suited to PMClite: internal factory use, sensitive data (e.g. defense, medical devices), unstable network, do not want long-term subscription, need offline availability.

4.5 vs black-box AI scheduling tools

In recent years, a class of "AI scheduling" tools has emerged: input data, output schedule directly, no explanation.

PMClite differences (covered in the previous article, summarized here):

Core reason: the PMC signs off on delivery dates and must be able to explain "why this order is scheduled here." A schedule that cannot be explained is one the PMC does not dare to use.

5. Boundaries: What It Does Not Do

Half of a tool's value lies in what it does; half lies in what it does not do. PMClite's "does not do" list:

5.1 No general ledger or cost accounting

The finance module only does reconciliation (receivables/payables), invoices, cash flow, and a simple profit summary. No general ledger, no accounts, no vouchers, no cost variance analysis, no fixed assets.

Reason: the general ledger is accounting work, not PMC work. Cramming it in makes the tool heavy, and it cannot beat professional finance software anyway.

5.2 No HR or attendance

No employee files, no shift scheduling, no time clock. Hours only record "actual line time occupied by a work order," not "who worked how long."

Reason: this is HR system territory, and attendance in small factories usually runs independently.

5.3 No quality management or SPC

No inspection sheets, no defect classification, no SPC control charts. The receiving note has a "QC result" checkbox (qualified / unqualified / concession / pending inspection), but that is a simple marker at the receiving step, not complete quality management.

5.4 No equipment management or OEE

No equipment ledger, no inspection checklists, no maintenance work orders, no OEE calculation. Line configuration has only four parameters: parallel stations / machines / daily hours / changeover time.

Reason: equipment management is another professional domain, and small factories usually have few machines that Excel can handle.

5.5 No CRM

Customer profiles only contain name, contact, phone, credit limit, payment terms, status, notes. No opportunities, no follow-up records, no sales funnel.

5.6 No multi-organization or multi-ledger

One database corresponds to one factory. Multiple factories require multiple instances, or use LAN sync to aggregate data onto one host.

5.7 No public internet access

Collaboration is possible within the LAN; public internet access is not. If remote data viewing is needed, it requires VPN or remote desktop.

6. Who It Fits, Who It Does Not

Fits

Does not fit

7. One-Sentence Summary

PMClite is not a "simplified ERP," nor a "domestic APS." It is a workbench that the PMC opens and uses every day, that can explain after scheduling, that can trace after modification, and that still degrades gracefully when something fails.

Its core bet is: the scheduling problem in small and medium factories is not "cannot compute the optimal solution," but "imperfect master data, unreliable constraints, and people not trusting black boxes." So it chooses to build an "explainable, tiered, intervenable, degradable" tool rather than pursuing theoretical optimality.

This choice does not apply to every enterprise, but for the batch it targets, it is the right fit.

Want to see the whole workbench on your own data?

On-premise · One-time purchase · Live in 1–3 days · 30-day full-feature trial

Explore PMClite →

Algorithms and implementation details referenced in this article reflect our public demo builds; the shipped product takes precedence. PMClite is an independent product and has no commercial relationship with any third-party vendor mentioned in this article.