Coming from NetLogo

Naga has agents on a grid with patches, in continuous space or on a network, with links, ticks, births and deaths. Six classic models ship with it. What it does not have is NetLogo's language: an agent's behaviour is a statechart and a list of declarative actions, not procedures.

What maps to what

NetLogo	Here
a breed	an agent type (agent_types) and a population of it (populations)
turtles-own	attributes of the agent type; parameters for what all agents share
to go ... tick	a tick on the type: on_tick actions and tick transitions, every dt
ask turtles [ ... ] in random order	"activation": "random"; also sequential and synchronous
a turtle variable that says what it is doing	a state of the statechart; transitions by timeout, rate, condition, message or tick
patches, patches-own, diffuse, evaporation	a grid space with patches: init, diffuse, evaporate; read with PATCH("name")
neighbors, turtles-here, in-radius	NEIGHBOURS(...), CELL_COUNT(...), NEAR(...), and their sums and means
forward, right, face	movement actions: random walk, turn, move to
links, link-neighbors	a population's network (random, small-world, scale-free); LINKED(...), LINK_MEAN(...)
hatch, die	the spawn and remove actions
random-float, random-normal	RAND(), UNIF, NORM, EXPO; every population has its own random streams
BehaviorSpace	naga doe (factorial, Latin hypercube, one at a time), naga scenarios
reporters and plots	the agent_summary and agent_attributes frames; timeseries
random-seed	--seed; replication i uses seed + i

Each row is a section of Agent-based modelling.

The classics

abm_wolf_sheep.json, abm_schelling_segregation.json, abm_boids_flock.json, abm_sugarscape_lite.json, abm_nagel_schreckenberg.json and abm_sir_small_world.json, with their numbers in the gallery. Each model's description says which published model it follows.

naga run abm_schelling_segregation

What is missing here

• A language. No procedures, no ask with a block of code, no lists and no anonymous reporters. An action is one of a fixed set (set, send, move, spawn, remove, ...) with expressions as arguments. If the behaviour does not fit, an agent type can carry a Python function instead; it runs only with your consent (Python behaviour).
• The interface. No world view, sliders, buttons or live plots. A run is headless; a recorded run replays agents as tokens, and patches are not drawn.
• Patches as agents. A patch holds named numbers that diffuse and evaporate. It runs no code and has no state of its own beyond them.
• Links across breeds. A link joins two agents of the same population.
• A reader for .nlogo or .nlogox files.
• Extensions (GIS, R, Python, sound, ...).

What you gain

• Event-driven agents as well as ticked ones: a transition with a rate or a timeout fires at its own time, with no step to choose.
• Agents beside queues, resources and stocks in one model: an agent can send an entity into a block, and a block can message agents.
• Replications, confidence intervals and experiments from the command line.