Agents.jl

Agents.jl is a library for writing agent-based models in Julia, a programming language.^1,2 An agent-based model is a simulation of individual agents that each react to their surroundings, including other agents, by a set of rules.¹ The library supplies the common parts: agents, several kinds of space, schedulers that decide which agent acts when, automatic collection of results into a table (a DataFrame), batch runs and distributed computing.¹ It belongs to the JuliaDynamics organisation of scientific software projects.^1,2

Space can be a network, a regular grid, continuous space or a real road network from OpenStreetMap, with road speed limits, searches for nearby agents and roads, and route finding.¹ Time can advance in fixed steps or, for continuous-time models, by an event queue.¹ Version 7 added an experimental layout for faster runs and a model type that links agent-based models with reinforcement learning, in which agents learn by trial and reward.¹ Plots, videos and interactive apps come through Julia's Makie graphics packages.¹ The project says its models take much less code than in other open-source frameworks and that it ran faster than every framework it compared itself with.¹

It installs from Julia's package manager and is free under the MIT licence.^1,2 The paper to cite is Datseris, Vahdati and DuBois, "Agents.jl: a performant and feature-full agent-based modeling software of minimal code complexity", in the journal SIMULATION, 2022.² Version 7.1.0 was released in October 2026.⁴

For someone modelling a place, the OpenStreetMap space lets agents travel real streets. The documentation's example spreads a zombie outbreak through a city's street network, with agents planning random routes and moving along roads, and the project points to a simulation of the whole city of Chicago.^1,3 The same pattern fits commuters, deliveries or evacuations. Mesa does the same job in Python, and NetLogo is a visual, lower-code start.