Compose visually
Create stateful nodes, connect them with relationships and define equations against explicit state and parameter references, then disable or delete any entity in place.
Build dynamic systems as graphs
Physical systems are networks. Simulate them that way.
Konjugate is open-source graph simulation software for dynamic systems modeling: a visual workbench and native simulation engine. Components contain state, relationships define how they interact and results can be explored over time.
- Model
- Nodes + relationships
- Engine
- Native C++
- Project
- MPL 2.0
brew install --cask zenineasa/konjugate/konjugate01
One engine, many domains
The same engine runs robots, banks and supply chains.
Konjugate itself is domain-agnostic: it only knows components, relationships and state. A toolbox adds a field's components, models and windows.
01In development
Robotics
Draw a robot from solids and joints in a 3D window. Each solid becomes a node and each joint an edge with its own equations, so the robot runs like any other model.
02Continuously improvingFintech
Stress-test interbank contagion and trace how shocks transmit between markets, working from your own data.
03Continuously improvingLogistics
Ports, warehouses, and road and rail lanes as stocks and flows, plus a tool that turns any region's OpenStreetMap data into a runnable network.
02
The model
Graph simulation software: one system, composed from local behavior.
Instead of assembling one monolithic system equation, describe the state owned by each component and the smaller interactions between components.
State evolution at node i
$$\dot{x}_i = \sum_j f_{ij}(x_i,x_j) + s_i(x_i,u)$$
- \(x_i\)
- State values contained in component i.
- \(f_{ij}\)
- The contribution to component i from its relationship with component j.
- \(s_i\)
- Behavior local to the component: sources, sinks, internal processes or external inputs.
Source term\(s_i(x_i,u)\)
Source term\(s_j(x_j,u)\)
Component i\(x_i\)temperature · charge
Component j\(x_j\)temperature · flow
Interaction term\(f_{ij}(x_i,x_j)\)
Interaction term\(f_{ji}(x_j,x_i)\)
Practically speaking: a battery can own temperature and charge, a coolant reservoir can own temperature and fill level and their relationships can describe heat transfer or signals between them. The engine combines those local contributions at every synchronization step.
03
Available in the beta
Model, execute and investigate in one workbench.
The current application already covers the complete path from a graph model to time-dependent results.
Recover structure from data
Import multivariate time-series CSV data and let causal inference propose candidate nodes and edges (linear or nonlinear) for review before they become part of the model.
Validate, then run
Use the same C++ validator from the desktop application or command line to reject invalid models, then execute offline at maximum speed or pace an online simulation for live interaction.
Explore over time
Scrub through results, inspect node values, compare signals on plots and extend a completed simulation from its final checkpoint.
Bring geometry
Represent components with built-in forms, a searchable library of bundled STL and STEP shapes or your own imported geometry. Appearance stays decoupled from model semantics.
Extend with add-ons and plugins
Open live and completed results in permissioned visualizers contributed through a general extensions interface.
04
AI model assistant
Describe the intent. Keep control of the model.
Use a local or hosted large language model (LLM) to turn a plain-English description into structured changes, then inspect every assumption and operation before anything reaches the workbench.
Model assistant
Ollama · Local
Requested change
Add hot and cold water vessels, then connect them with a steel thermal path.
Validated proposal5 changes
- +Hot-water vesselNode · temperature state
- +Cold-water vesselNode · temperature state
- ↔Steel thermal pathRelationship · explicit parameter references
Structure and native model validation passed
Run an LLM locally or in the cloud
Run models locally through Ollama or configure OpenAI, NVIDIA, Hugging Face and Gemini providers.
Generate model operations
The assistant proposes versioned nodes, states, relationships, parameters and equations from your natural-language request, instead of driving the interface opaquely.
Review before mutation
Invalid output is repaired or rejected. Valid changes remain a preview until the user explicitly applies one undoable transaction.
05
Writing
Notes from building it in the open.
Deep dives on the model and documentation for every feature in the workbench. Prefer video? Watch the 6-part On-Ramp series on YouTube
Loading articles…
Open development
Help shape a different way to model dynamic systems.
Follow releases, videos and new articles on our WhatsApp channel, or join the conversation and ask questions on Discord.
Prefer email? Reach us at info@konjugate.com.