dynos-core

Pure-Python types for describing a planning domain. Predicates, transitions, object types, and the world states they describe. dynos-core is the data layer that every other DYNOS package shares.

This contains no planner, no executor, and no network code, it's just classes that define a domain. The DYNOS backend reads these structures to plan/execute/monitor/verify, clients read them to talk about goals, both share the same in-memory representation.

Most users do not install dynos-core directly, You get it transitively when you pip install dynos-client (or any package above it). Install it on its own if you are authoring a new domain (or extending a new one, like when making custom actions) and want the lowest-level types without pulling in the HTTP client.

Install

pip install dynos-core

Single dependency: varname. This is used to infer the names of objects being created.

Concepts

DYNOS describes the world as a set of named facts (fluents) about typed objects. A goal is a state of those facts; a plan is a sequence of transitions that move from one state to another. Actions are specific code implementations of transitions.

Fluents and ground atoms

A fluent is a named, parameterised proposition. make_new_fluent creates one; calling it on specific objects creates a GroundAtom. This is a fluent applied to particular arguments.

from dynos_core import make_new_fluent, ObjectType, World

class Robot(ObjectType): ...
class Room(ObjectType): ...

at = make_new_fluent("at", robot=Robot, room=Room)
r = Robot("r1")
kitchen = Room("kitchen")

current = World({at(r, kitchen)})
print(at(r, kitchen) in current)   # True

A World is a frozenset of GroundAtom describing what is currently true. In the current implementation, anything not in the set is false (closed-world assumption), but more principled approaches are being designed.

Object types and value fields

An ObjectType is a category of thing in the domain. For example, Robot, Room, Zone, Coordinate. ValueField descriptors attach typed numeric/string fields:

from dynos_core import ObjectType, ValueField

class Zone(ObjectType):
    altitude = ValueField(float)
    vertices = ValueField(list)

The class is pure Python; instances live in a backend-managed numeric database at runtime, accessible through the same descriptor syntax (zone.altitude.get() / .set(70.0)). dynos-core only declares the schema.

Transitions

A Transition is an action schema: name, parameter type, preconditions, effects. It says nothing about how the action runs.

from dynos_core import Transition, TransitionParams

class GotoParams(TransitionParams):
    start: Room
    end: Room

goto = Transition("goto", params_type=GotoParams).define(
    preconditions=[at(GotoParams.start)],
    adds=[at(GotoParams.end)],
    removes=[at(GotoParams.start)],
)

The define() call stores the symbolic spec on the transition. The dynos backend extends Transition with cost callbacks, applicability checks, and execution hooks; this package's Transition is just data.

Extensions

extend_object_type and extend_transition_params add fields after the class is declared. This supports the publish-stub pattern: a domain package declares a small public schema; the backend extends it with private fields at import time. The public release strips the extension calls so users see only the public surface.

from dynos_core.extension import extend_object_type
extend_object_type(Zone, depth=ValueField(float))   # backend-only addition

See extension.py for the full rules (additive only, must run before any plan construction, etc.).

What dynos-core does NOT do

• No planner. The actual planning lives in the dynos backend (not on PyPI).
• No @Action / @Belief / @Script decorators. Those run code; this package is data only. They live in the backend (and you import them from dynos.databases.typed_action when building a backend; or via the conditional-import pattern documented in dynos-client's README when authoring client-only code).
• No network or robot connectivity. That's dynos-client.

Public API

Next

To talk to a running DYNOS backend, install dynos-client. To get the Sentry survey vocabulary, install dynos-sentry-domain. The walkthrough at user_guide.md ties them together end-to-end.