Go back to Market Research
AnyLogic supplies software for building and running simulation models through discrete-event, agent-based, and system-dynamics methods. Its desktop and online environments support model development, experiments, and shared execution. The related anyLogistix product provides supply chain network design, inventory-policy analysis, transportation modeling, and risk simulation. This assessment covers AnyLogic 8.9.10, the documented Private Cloud 2.9 deployment options, and anyLogistix 3.4.5. (1, 2, 13, 16)
Overall supply chain score: 5.8/10
| Dimension | Score |
|---|---|
| Supply chain depth | 4.8/10 |
| Decision and optimization substance | 5.6/10 |
| Product and architecture integrity | 5.6/10 |
| Technical transparency | 7.2/10 |
| Vendor seriousness | 6.0/10 |
AnyLogic provides substantial modeling machinery and unusually inspectable documentation. Its simulation engine can represent congestion, shared resources, uncertain transport times, and feedback between operational processes. Public APIs, exportable applications, experiment versions, and detailed deployment instructions give customers meaningful control over their models. The central limitation is the work needed to turn these capabilities into a maintained economic decision system.
anyLogistix makes that limitation concrete. Its network optimizer solves a static formulation with CPLEX; dynamics and uncertainty are investigated through a separate simulation workflow. Its safety-stock experiment targets a selected service level. These mechanisms can help investigate a network or policy, but they do not establish joint economic optimization of recurring decisions under uncertain demand and supply. (3, 4, 26)
The principal engineering concern is execution security. Practitioners can extend models with Java and local Python, while the documented script-based Private Cloud configuration leaves the Java sandbox switch off. Containers, authentication, and network boundaries are described, but the reviewed design does not establish strict confinement of model-authored code and its credentials. This matters increasingly as the new AI Assistant makes model generation easier. (11, 15, 21)
AnyLogic versus Lokad
AnyLogic equips a modeling team to construct and experiment with a representation of its operations. Lokad’s stated approach centers on a recurring calculation that ranks executable purchases, allocations, production orders, and other actions by expected economic return. Its inventory optimization documentation includes uncertainty in demand and lead times, with customer-specific logic maintained by Supply Chain Scientists.
The distinction becomes practical when considering a delayed component. An AnyLogic model can reproduce production queues, transport delays, and competing uses of shared capacity. A team can investigate alternative expediting or allocation rules and measure their simulated effects. AnyLogic supplies tools for that work; the team must construct the relevant probabilities, costs, dependencies, and policy. The documented anyLogistix workflow supplies predefined network and inventory formulations, whose assumptions also need scrutiny. (5, 6, 7)
Lokad places the economic choice itself at the center of the implementation. Its broader perspective, developed in Introduction to Supply Chain and Holimization, emphasizes revising executable logic when reality exposes a faulty formulation. Running a model faster or searching its parameters more thoroughly cannot repair an omitted commercial consequence.
AnyLogic has useful foundations for such revision: accessible model logic, programmatic execution, and versioned experiments. Its flexibility is valuable for process engineering and custom research. The buyer’s economic comparison must include the modeling, integration, validation, and operations work required to turn that flexibility into dependable decisions. Lokad’s specialist maintenance work belongs in the same comparison.
Static optimization and stochastic simulation
The anyLogistix network-optimization documentation identifies a profit-maximizing objective, facilities, flows, costs, periods, and constraints. It explicitly excludes dynamic characteristics. The product page describes optimizing with IBM ILOG CPLEX, changing the formulation through successive iterations, and then converting the result into a simulation scenario for deeper risk analysis. (3, 26)
This separation is consequential. A network that is attractive under fixed demand and flows may perform poorly when supplier delays, variable workloads, congestion, and lost sales interact. Testing the selected network afterward can reveal weaknesses. It does not automatically make the original search value those risks or select the best contingent policy. The workflow needs a mechanism for feeding the economic consequences back into the formulation.
The solver’s constraints also contain policy choices. anyLogistix lets users soften constraints through penalties, including demand, product flows, storage, and production. Those numbers influence which shortage or operational violation becomes acceptable. A feasible or mathematically optimal result can therefore be an artifact of how the penalties were chosen. The material issue is whether these terms represent actual consequences, and how mistakes in them are detected and corrected. (8)
The transport documentation exposes a useful example of model boundaries: certain trip-based cost formulas cannot be evaluated with less-than-truckload flows when vehicle loads are unspecified. A flow formulation has lost information needed to price the physical transport activity. This is precisely the kind of mismatch that a good solver cannot fix by searching harder. (6)
General AnyLogic simulation optimization offers a different mechanism. Its Genetic and OptQuest engines repeatedly run a model with alternative parameters and evaluate a user-defined objective. For stochastic models, repeated runs estimate the objective; OptQuest can vary the number of replications. The objective can therefore include nonlinear behavior produced by the simulation. (7)
The burden moves to model quality and search cost. Every candidate policy may require many simulated trajectories, and increasing the number of interacting decisions expands the search. More replications reduce sampling noise around the model’s assumptions. They do not validate those assumptions or discover a missing stockout consequence, supplier dependency, or commercial opportunity.
Uncertainty, safety stock, and digital-twin validity
anyLogistix explicitly permits probability distributions for transportation time. Its inventory and simulation tools support stochastic behavior across a network, rather than limiting uncertainty to a demand-forecast percentage. This is a meaningful capability. The public descriptions do not supply an automatic process for estimating and maintaining the joint distribution of demand, supply delays, and shared disruptions. That work belongs to the model implementation. (6)
The standard safety-stock experiment seeks a user-selected service level. Supported inventory policies include min-max, reorder quantities, and MRP-style replenishment. In MRP, the model compares projected stock against a buffer and schedules an order using expected lead time. These are inspectable policies, but the service target and buffer remain prescriptions supplied to the calculation. They do not express the marginal economic value of every competing use of inventory or cash. (4, 5)
A Monte Carlo experiment can test these policies across random trajectories. Its value depends on the modeled dependencies and the choice being tested. Independently sampled uncertainties can miss a disruption that simultaneously slows several suppliers and changes demand. Optimizing each buffer separately can miss a scarce component shared by several profitable products. The platform’s programmability allows richer models; the required economic logic is not furnished merely by selecting a simulation experiment.
The digital-twin proposition goes further than retrospective analysis. anyLogistix describes forecasting financial and SKU flows, refreshing forecasts from live information, and testing the coming days and weeks. These are high-dimensional conditional forecasts. They require validation at the horizons and resolutions relevant to the proposed actions. (28)
There is some public validation evidence. In the pharmaceutical case, Sterling Simulation rebuilt an inflexible model using a hybrid approach and tested its logic on another supply chain with available data, since the proposed expansion had not yet occurred. That is a concrete check. The case publishes no forecast-error distribution or measurement of how well the model predicted the economic effects of subsequent interventions. (24)
A visual match, correct event sequence, or tight simulation confidence interval addresses only part of this problem. The remaining question is whether the model’s predictions remain useful when the business changes a supplier, allocates scarce stock differently, or expands into a new market. The reviewed digital-twin material does not provide a general evaluation protocol for that question.
Results, productivity, and responsibility
The ICP Group case combines anyLogistix with SimWell consulting, Alteryx data workflows, and Power BI dashboards. The project covered a network with more than twenty North American factories, nearly thirty distribution centers, and over ten thousand SKUs. It identified a 7% cost-saving opportunity against a 2022 baseline. The published result is a projected opportunity; realized savings and the complete cost of maintaining the system are not reported. (9)
The case also describes user feedback and repeated movement between candidate-site analysis and network optimization. This is model-development work with a concrete purpose. Its economics depend on how frequently the network changes, how many assumptions require revision, and whether the proposed changes are executed successfully. Data refresh alone does not eliminate that work.
Cardinal Health provides stronger operational evidence. AnyLogic’s case reports that modeling worker interactions exposed uneven workload and congestion, and that rebalancing work reduced average shifts from 10.5 to 7.25 hours. It attributes more than $3 million in annual savings to the modeling program. This is a specific mechanism tied to a reported operating result, although implementation and recurring costs are not disclosed. (29)
These cases measure different forms of productivity. Warehouse labor saved through a layout or workload change is an economic benefit. It does not measure the productivity of the analysts who build and maintain the model. The pharmaceutical example demonstrates the latter burden: an existing simulation had to be redesigned because adapting it to new conditions was difficult. (24)
The September 2026 AI Assistant addresses part of that burden directly. It drafts simple models, explains existing structures, and assists troubleshooting and modification. The vendor describes the generated model as an initial draft requiring further work. This is a relevant application of LLMs to improving the executable model, with a more concrete scope than a broad promise of autonomous supply chain management. (15)
The missing productivity evidence is the sustained cost per maintained decision portfolio: model engineering, data repair, validation, customer review, and operational support. APIs can automate experiment execution, but automating experiments is only one step toward reliably automating economically useful actions. anyLogistix’s API tutorial covers importing scenarios, changing periods, running experiments, and exporting results; access requires the Professional Server edition. (27)
Responsibility should follow these boundaries. AnyLogic supplies the engine, while customers and implementation partners commonly supply the business model. The reviewed cases do not establish a shared failure criterion, net-return account, or obligation to repair an economically poor formulation. The buyer needs these responsibilities defined before simulation outputs become operational policy.
Technology history and delivery
The release history shows a sustained simulation product rather than a succession of unrelated enterprise applications. GIS integration, an embedded model database, domain libraries, and cloud experiments were introduced through identifiable releases. Cloud model sharing arrived with AnyLogic 8 in 2017; Private Cloud Lite followed in 2019. These developments have corresponding current documentation and execution interfaces. (25)
The 2020 Microsoft Bonsai promotion made a broad accessibility claim about teaching AI without an AI background. Its demonstrated mechanism was narrower: use a simulator to train a reinforcement-learning policy and then test the policy in simulation. The 2021 engineering presentation documented the released reinforcement-learning experiment and platform-specific exports to Bonsai and Pathmind. These were real integration features; the cited demonstrations did not establish dependable economic control of a production supply chain. (17, 18)
The partner integrations did not retain that continuity. Microsoft’s Bonsai connector repository was archived on February 7, 2024. Pathmind’s November 17, 2021 final web-application release disabled new sign-ups and upgrades, and its investor Y Combinator now lists the company as inactive. A historical demonstration built on these services cannot establish a currently supported deployment path. The short life of the Pathmind service also illustrates the maintenance risk of building an industrial decision system around an external AI platform. (32, 33, 34)
Current AnyLogic integration documentation instead describes Pypeline, ONNX Helper, and Alpyne. The vendor’s Alpyne 1.2.1 release in March 2025 included dependency and example fixes, providing a maintained alternative for controlling exported reinforcement-learning experiments from Python. This preserves an integration route, with implementation and validation work still required to replace an earlier partner workflow. The customer supplies a credible environment, objective, training design, and evaluation against operational reality. (21, 31, 35)
November 2023’s anyLogistix 3.1 release added multi-user server access and Java/Python APIs. The announcement explained the previous workflow of exporting and emailing project data, and the new ability to share a project and script repeated experiments. This is an observable reduction in handling work, supported by today’s API documentation. (30)
The 2026 releases continue both infrastructure and modeling work: Private Cloud gained Kubernetes deployment, while AnyLogic added an LLM assistant for drafting and editing models. The historical record supports a coherent engineering roadmap. Its weaker part is outcome accounting: delivery of a connector, simulator feature, or assistant has rarely been accompanied by a comparable record of sustained decision quality and total economic return. (13, 15, 16)
Software materials and configuration maintenance
The materials are substantially more visible than at most planning vendors. They also show why a sophisticated simulation service has an operating cost.
| Layer | Documented materials and scope |
|---|---|
| Exported models | Java 17 or later, compiled model JARs, AnyLogic runtime libraries, and optional model dependencies. |
| Optimization | CPLEX for anyLogistix network optimization; Genetic and OptQuest for AnyLogic simulation optimization. |
| Private Cloud services | PostgreSQL, RabbitMQ, an S3-compatible file store, NGINX, and separate execution and coordination services. |
| October 2026 deployment | Docker installation or Kubernetes/Helm; the 2.9 release replaces MinIO with SeaweedFS. |
| Vendor-operated infrastructure | Current employer vacancies specify AWS/EKS, Docker/Kubernetes, GitLab CI, infrastructure-as-code tools, and monitoring. |
The architecture records some simplification: Cassandra stopped being used in Cloud 2.5.0, with migration to PostgreSQL. The October 2026 storage change replaces MinIO with SeaweedFS. Both migrations retire an infrastructure dependency while introducing migration and compatibility work. (10, 13)
The current infrastructure still carries integration and maintenance costs. The Kubernetes guide specifies PostgreSQL 17, RabbitMQ 4.0, object storage, and executor pods that each require one CPU core and 1.5 GB of memory for a concurrent run by default. Replications and parameter searches multiply that workload. The documented Lite and Pro concurrency limits also constrain how much additional hardware a customer can use. (12)
The larger cost can sit above the runtime. A supply chain model contains policies, event rules, distributions, cost formulas, mappings, and integration code. The ICP implementation’s Alteryx and Power BI layers add customer-specific dependencies alongside anyLogistix. Every change must preserve agreement between the operational data and the model’s interpretation of it. (9)
AnyLogic offers useful maintenance primitives. Model versions retain separate experiments; source files can be downloadable, and execution APIs associate runs and results with a version. Standalone Java export allows execution outside the authoring application, although the exported application still depends on AnyLogic’s runtime. (19, 20, 22)
These capabilities support reproducibility. The implementation must still connect a model version to its input data, dependencies, validation results, and executed business actions. The published workflow does not provide a complete economic regression-testing and repair process. The question is how cheaply a customer can correct a wrong formulation and verify the correction, not just rerun the same model.
Security of practitioner-authored models
An AnyLogic model is executable software. Its Java code can use database connections; the documented database API includes operations that modify the data source. Pypeline additionally invokes a local Python installation and external Python libraries. The vendor presents that bridge as optional open-source tooling without an obligation of official support or compatibility. (21, 23)
This creates a concrete boundary to secure whenever modelers receive sensitive data or production credentials. A mistaken script, compromised dependency, or generated code can act with the permissions supplied to the process. Calling it a simulation does not remove those permissions.
Private Cloud documents Docker containers and a private-network assumption, with normal user traffic routed through a frontend proxy. These are useful deployment boundaries. The separate executor configuration for script-installed Private Cloud 2.9 shows securityManagerEnabled: false; the accompanying explanation says enabling it launches models in a sandbox at a performance cost. The documented configuration therefore makes this additional confinement optional. (10, 11)
The concern is specific to the demonstrated boundary: strong restrictions on model code, network destinations, credentials, and dependencies are not established by that configuration. The Kubernetes installation has a separate deployment path. Its pod resource allocation is documented, but the reviewed material does not establish equivalent code-level and data-access confinement there. (12)
The AI Assistant increases the importance of this issue by lowering the effort needed to create and modify executable models. Sound operation requires privileges and containment to be enforced outside the generated code, together with recoverable data access. Private hosting and authentication alone do not supply those controls.
Supply chain score
Each dimension averages five criteria. The overall score averages all twenty-five sub-scores and rounds to one decimal place.
Supply chain depth
| Criterion | Score |
|---|---|
| Economic framing | 6/10 |
| Decision end-state | 4/10 |
| Conceptual sharpness on supply chain | 6/10 |
| Freedom from obsolete doctrinal centerpieces | 3/10 |
| Robustness against KPI theater | 5/10 |
Average: 4.8/10
Economic framing — 6/10. Network optimization includes profit, transport and facility costs; models can represent additional economic consequences. Their adequacy depends on the customer’s formulation, while safety-stock calculation remains tied to a chosen service target. (3, 4, 26)
Decision end-state — 4/10. The software produces network designs, policies, and experiment results. Recurring operational actions require further modeling and integration. The APIs make that engineering feasible, but the documented standard workflow ends at experiment execution and result export. (19, 27)
Conceptual sharpness on supply chain — 6/10. The simulation tools represent interactions, queues, transport variability, and shared resources. The distinction between a realistic process representation and a reliable economic decision policy receives less explicit treatment. (6, 24, 29)
Freedom from obsolete doctrinal centerpieces — 3/10. anyLogistix supplies MRP, safety-stock, and min-max policies. Their transparency makes the doctrine inspectable; it does not resolve the economic weaknesses of fixed service targets and buffer-based replenishment. (4, 5)
Robustness against KPI theater — 5/10. The ICP case labels savings as an opportunity, and Cardinal Health supplies an operational mechanism behind reported savings. Full ownership costs, forecast validity, and comparable economic baselines remain incomplete. (9, 28, 29)
Decision and optimization substance
| Criterion | Score |
|---|---|
| Probabilistic modeling depth | 6/10 |
| Distinctive optimization or ML substance | 6/10 |
| Real-world constraint handling | 7/10 |
| Decision automation and labor productivity | 4/10 |
| Resilience under real operational complexity | 5/10 |
Average: 5.6/10
Probabilistic modeling depth — 6/10. Transport distributions, stochastic models, and replicated objective evaluation are explicit. Joint probability estimation and calibration remain implementation work, with no packaged economic treatment demonstrated across demand and supply uncertainty. (6, 7)
Distinctive optimization or ML substance — 6/10. Multimethod simulation and configurable stochastic search constitute meaningful computational machinery. CPLEX and OptQuest are identified external engines; the vendor’s contribution lies in modeling and integration. Static network formulations remain a substantial limitation. (1, 3, 7)
Real-world constraint handling — 7/10. Models can represent congestion, resource contention, production processes, and transport behavior. The documented inability to price some transport policies within a flow formulation illustrates why physical detail must remain connected to the economic objective. (6, 24, 29)
Decision automation and labor productivity — 4/10. Scripted experiments and model-generation assistance reduce parts of the analytical workload. Public evidence does not measure the total labor needed to maintain and execute a recurring decision portfolio. (15, 27, 30)
Resilience under real operational complexity — 5/10. Rich models permit testing disruptions and operational interactions. Maintaining their assumptions and validating the effects of interventions remain demanding, as illustrated by the pharmaceutical model redesign and the separation of network optimization from simulation. (3, 24)
Product and architecture integrity
| Criterion | Score |
|---|---|
| Architectural coherence | 7/10 |
| System-boundary clarity | 7/10 |
| Security seriousness | 3/10 |
| Software parsimony and ownership cost | 4/10 |
| Compatibility with programmatic and agent-assisted operations | 7/10 |
Average: 5.6/10
Architectural coherence — 7/10. Model authoring, Java runtime artifacts, experiment execution, and cloud services form a documented product chain. Current migrations retire some older components. The runtime and customer-built integration layers still require substantial coordination. (10, 13, 20)
System-boundary clarity — 7/10. The vendor distinguishes the simulation engine, cloud execution, anyLogistix policies, and third-party integrations. Edition restrictions and the unsupported status of the optional Python bridge are visible. (2, 21, 27)
Security seriousness — 3/10. Deployment boundaries and resource settings are documented, but practitioner code can access external data and the illustrated script-install configuration disables Java sandboxing. Strict restrictions on credentials, dependencies, and outbound access remain insufficiently established. (11, 12, 23)
Software parsimony and ownership cost — 4/10. The execution service has several substantial infrastructure dependencies, and repeated stochastic experiments consume additional compute. Customer modeling, data pipelines, validation, and policy maintenance add costs beyond the runtime and license. (9, 12, 14)
Compatibility with programmatic and agent-assisted operations — 7/10. Public APIs, Java export, model versions, and the new editing assistant provide useful building blocks for automated model improvement and execution. Their usefulness depends on the surrounding validation and operational controls. (15, 19, 20, 22)
Technical transparency
| Criterion | Score |
|---|---|
| Public technical documentation | 8/10 |
| Inspectability without vendor mediation | 8/10 |
| Portability and lock-in visibility | 6/10 |
| Implementation-method transparency | 7/10 |
| Evidence density behind technical claims | 7/10 |
Average: 7.2/10
Public technical documentation — 8/10. APIs, optimization behavior, inventory rules, infrastructure requirements, and executor settings are publicly described. The documentation exposes limitations that buyers would otherwise discover during implementation. (5, 6, 11, 19)
Inspectability without vendor mediation — 8/10. Public examples, learning editions, source-download options, and engineering documentation allow substantial investigation before a sales engagement. Complete evaluation of production behavior still requires a configured model and licensed capabilities. (1, 2, 22)
Portability and lock-in visibility — 6/10. Standalone Java applications and result APIs offer meaningful export paths. The applications retain proprietary engine dependencies, while modeling libraries and anyLogistix formulations would require work to replace. (19, 20)
Implementation-method transparency — 7/10. The sources describe data imports, model construction, experiments, software deployment, and customer integration patterns. The method for sustained economic validation and correction is much less standardized. (9, 12, 24, 27)
Evidence density behind technical claims — 7/10. Release notes, runnable integration mechanisms, and detailed examples substantiate the machinery. Evidence that these components consistently deliver better decisions after all operating costs is narrower. (7, 13, 18, 29)
Vendor seriousness
| Criterion | Score |
|---|---|
| Technical seriousness of public communication | 7/10 |
| Technology roadmap integrity | 7/10 |
| Conceptual sharpness | 6/10 |
| Economic accountability | 4/10 |
| Defensibility in an agentic-software world | 6/10 |
Average: 6.0/10
Technical seriousness of public communication — 7/10. Detailed engineering descriptions accompany the broad simulation claims. The 2026 assistant is presented as a draft-modeling tool with a limited role, which gives buyers a concrete capability to assess. (1, 15, 16)
Technology roadmap integrity — 7/10. The sequence from simulation libraries to cloud execution, APIs, and assisted model editing has identifiable releases and continuing documentation. The earlier Bonsai and Pathmind routes lost continuity; the current Python integration has its own release history. Dependency migration and its customer cost remain part of evaluating the roadmap. (25, 30, 32, 33, 35)
Conceptual sharpness — 6/10. The product distinguishes modeling methods, optimization formulations, and experiment types. It is less explicit about the gap between a credible simulated system and a validated policy for allocating scarce economic resources. (1, 7, 26, 28)
Economic accountability — 4/10. Named cases provide useful outcome evidence, including reported warehouse savings. The economic responsibilities of the engine vendor, implementation partner, and customer remain dispersed, without a published full-cost or remediation framework. (9, 29)
Defensibility in an agentic-software world — 6/10. A maintained simulation runtime, operational libraries, APIs, and editable models provide more substance than a conversational layer alone. Their future value depends on reducing the cost of validated model improvement and enforcing secure execution. (1, 15, 19, 21)
Sources
[1] AnyLogic modeling features
AnyLogic modeling features. Accessed November 4, 2026. [2] anyLogistix editions and version
anyLogistix editions and version. 2026-09-10; Accessed November 4, 2026. [3] Network optimization workflow
Network optimization workflow. Accessed November 4, 2026. [4] Safety stock calculation
Safety stock calculation. Accessed November 4, 2026. [5] Inventory policies
Inventory policies. Accessed November 4, 2026. [6] Transport paths, uncertainty and costs
Transport paths, uncertainty and costs. Accessed November 4, 2026. [7] Simulation optimization and replications
Simulation optimization and replications. Accessed November 4, 2026. [8] Constraints and penalties
Constraints and penalties. Accessed November 4, 2026. [9] ICP Group network design case
ICP Group network design case. 2023; Accessed November 4, 2026. [10] Private Cloud architecture
Private Cloud architecture. 2026-10-06; Accessed November 4, 2026. [11] Private Cloud service configuration
Private Cloud service configuration. Archived June 8, 2026; accessed November 4, 2026. [12] Private Cloud system requirements
Private Cloud system requirements. 2026-10-29; Accessed November 4, 2026. [13] Private Cloud release notes
Private Cloud release notes. 2026-10-29; Accessed November 4, 2026. [14] Employer technical vacancies
Employer technical vacancies. Accessed November 4, 2026. [15] AnyLogic 8.9.10 AI assistant
AnyLogic 8.9.10 AI assistant. 2026-10-12; Accessed November 4, 2026. [16] AnyLogic release history
AnyLogic release history. 2026-09-24; Accessed November 4, 2026. [17] Microsoft Bonsai and AnyLogic integration
Microsoft Bonsai and AnyLogic integration. 2020-09-02; Accessed November 4, 2026. [18] 2021 product research and development presentation
2021 product research and development presentation. 2021; Accessed November 4, 2026. [19] Cloud REST API
Cloud REST API. 2026-10-23; Accessed November 4, 2026. [20] Standalone Java model export
Standalone Java model export. Accessed November 4, 2026. [21] Pypeline integration
Pypeline integration. Accessed November 4, 2026. [22] Cloud model version management
Cloud model version management. Accessed November 4, 2026. [23] Database API and write operations
Database API and write operations. Accessed November 4, 2026. [24] Pharmaceutical supply chain simulation
Pharmaceutical supply chain simulation. Accessed November 4, 2026. [25] Product timeline
Product timeline. Accessed November 4, 2026. [26] Network optimization experiment
Network optimization experiment. Accessed November 4, 2026. [27] anyLogistix API tutorial
anyLogistix API tutorial. Accessed November 4, 2026. [28] Supply chain digital twins
Supply chain digital twins. Accessed November 4, 2026. [29] Cardinal Health warehouse modeling results
Cardinal Health warehouse modeling results. Accessed November 4, 2026. [30] anyLogistix 3.1 API and multi-user release
anyLogistix 3.1 API and multi-user release. 2023-11-10; Accessed November 4, 2026. [31] Current AI simulation integrations
Current AI simulation integrations. Accessed November 4, 2026. [32] Microsoft Bonsai connector archived
Microsoft Bonsai connector archived. 2024-02-07; Accessed November 4, 2026. [33] Pathmind final webapp release
Pathmind final webapp release. 2021-11-17; Accessed November 4, 2026. [34] Y Combinator: Pathmind inactive
Y Combinator: Pathmind inactive. Accessed November 4, 2026. [35] AnyLogic Alpyne 1.2.1 release
AnyLogic Alpyne 1.2.1 release. 2025-03-06; Accessed November 4, 2026.