Skip to content

Home

Jump to bottom
nick_tsaizer edited this page Apr 25, 2026 · 5 revisions

Tensor Planner Wiki

Welcome to the Tensor Planner wiki. This wiki explains what the library is, how it works internally, which bindings are available, and how to use it from C, C++, C#, Unity, and Jai.

What is Tensor Planner?

Tensor Planner is a compact C++20 planning library. It solves problems where you know:

  • the objects in a world,
  • the facts currently true about those objects,
  • the actions that can change those facts,
  • and the goal facts you want to make true.

The planner searches for an ordered action sequence that transforms the initial state into a goal state.

In application terms, you do not hard-code every possible sequence. You describe the rules of the world and the desired outcome. Tensor Planner searches the available action space and returns a valid sequence for the current state.

Why use it?

Use Tensor Planner when behavior is easier to describe as rules and goals than as hand-written step-by-step scripts.

Good fits:

  • NPC goal planning,
  • dependency-heavy gameplay systems,
  • simulation and design-validation tools,
  • tactical, scheduling, or logistics-style sequencing,
  • AI/model experiments that need tensor-shaped planning data.

Avoid it for one-off behaviors where a direct script is simpler.

Main features

  • Typed objects, predicates, action parameters, facts, and goals.
  • STRIPS-like actions: preconditions, add effects, delete effects.
  • Numeric functions, numeric preconditions, and numeric effects for scalar state.
  • Guided search with bounded candidate grounding.
  • Optional external candidate scorer callback.
  • Tensor exports for schema, problem state, candidates, and action graph.
  • C ABI with explicit ownership.
  • C++ fluent wrapper.
  • C# wrapper and Unity package.
  • Jai module with generated C bindings and type-first wrapper.

Wiki pages

Page What it covers
Getting Started Build, test, package, and run the included checks.
Core Concepts Domains, objects, predicates, actions, states, goals, and plans.
How the Planner Works Grounding, guided search, scoring, tensors, and limits.
C API Native ABI handles, solve flow, tensors, ownership.
C++ Fluent API Type-safe C++ wrapper with code snippets.
C# and Unity Managed wrapper, Unity package, async API, and runtime notes.
Jai Bindings Generated C bindings and Jai typed-domain wrapper.
Planning Patterns Small code snippets that show how to model tasks.
Memory Ownership Lifetime rules for C, C++, C#, Unity, and Jai.
Packaging build.sh, build.ps1, output layout, Linux/Windows artifacts.
FAQ and Limitations Common questions, constraints, and current limitations.

Binding overview

Binding Best for Main file
C API Engines, FFI, language bindings include/tensor_planner.h
C++ fluent API Native gameplay/simulation code include/tensor_planner.hpp
C# / Unity Unity and .NET code dev.nick.tensor-planner/Runtime/TensorPlanner.cs
Jai Jai projects and compile-time schemas modules/Tensor_Planner/module.jai

Minimal mental model

Tensor Planner answers this question:

Given the facts that are true now, which actions can make my goal facts true?

You provide:

  1. Domain: the rules of the world.
  2. State: the current world instance.
  3. Goals: facts that must become true.
  4. Limits: search and memory bounds.

The result is:

  1. solved flag,
  2. solve metrics,
  3. ordered plan steps,
  4. action arguments mapped back to object IDs or real wrapper objects.

Repository

Main repository: https://github.com/NickTsaizer/tensor_planner

Key folders:

include/                    C API and C++ fluent wrapper
src/                        native implementation
tests/                      native smoke tests
csharp/                     .NET wrapper project and smoke test
dev.nick.tensor-planner/    Unity package
modules/Tensor_Planner/     Jai module

Start with Getting Started if you want to build and run it.

Clone this wiki locally