Powder RPG

Powder RPG is a survival game mod for The Powder Toy — real falling-sand physics, not a fake meter on top. You mine, craft, breathe real oxygen, build machines, and survive in a side-scrolling world made entirely of simulated particles. This repo is not the upstream Powder Toy project; it is an RPG fork (phoenixfire808/Powder-RPG on GitHub).

Download the latest Windows build — unzip, then double-click PowderRPG.exe. No batch file required.

Repo layout

Path	What
src/	Engine fork (C++ elements, bridge HTTP, game view)
scripts/lua/rpg.lua	Powder RPG gameplay (survival, world gen, HUD)
scripts/lua/rpg_plugins/	RPG plugins (machines, companion, save, UI, …)
bridge_src/	Lua bridge modules (realism, RPG loader)
build_autorun.py	Builds build/autorun.lua from bridge_src/
powder_toy_mcp.py	Scripting/automation server exposing game state and actions over a local HTTP bridge, for external tooling (python powder_toy_mcp.py)
powder_ext/	Automation tool implementations used by the scripting server
knowledge/	Design docs, modules, structures, build lessons (local-only, .gitignored — not part of this published repo)
releases/	Per-version release notes
DEVELOPMENT.md	Development protocol and house rules for this project

Clone this repo, build the engine (meson + ninja — see wiki), run python build_autorun.py, then play from build/PowderRPG.exe.

Table of contents

• Powder RPG
• Life support & machines
• New physics & elements
• Sandbox quality-of-life
• Getting the update while you play
• Vision
• Roadmap
• Running from source
• Credit & license

Powder RPG

A continuous, side-scrolling world built entirely out of the sandbox's own particles — every block you stand on, mine, or die to is a real simulated element, and every survival system runs on the engine's actual physics rather than a scripted approximation of it.

• A real atmosphere, not a meter. Digging opens a genuine low-pressure pocket that the sim's own air physics fills from above; the surface holds a real, permanent positive pressure baseline pushing back down. Oxygen is a real, visible particle whose density scales with depth — thin and sparse far underground, abundant at the surface — and breathing consumes it and exhales real CO2 next to you. Sealed rooms only get dangerous once they're genuinely deep, not the instant you close a door at ground level.
• Survival needs: hunger and thirst drain over time and do real damage at zero; a Canteen lets you carry and boil water, and simply standing in a real lake quenches thirst directly. Taking damage (falls, burns, radiation, hunger, suffocation) draws real blood that isn't walkable, so you can't casually stand on your own puddle.
• A permanent companion — not a summonable pet: she's present from the moment a new world generates, pathfinds and fights alongside you with her own HP and death/revive cycle, reacts to what she sees in the world, and answers you in a chat box. Her behaviour is hand-authored against the simulation, not delegated to a model.
• Fallout-style radiation — uranium and plutonium contaminate the ground around them, which decays slowly over real time, and your character accumulates a long-term dose that keeps doing damage even after you've walked away. Lead blocks the buildup.
• Day/night cycle and weather, with an adjustable day length, rain that can strike lightning, and a toggleable enemy spawner that sends stick-figure attackers your way.
• Deep crafting progression — workbench, furnace, and anvil crafting stations feed into a Research Bench and an Advanced Lab past that, gated behind real material costs; ores smelt into bars, bars forge into tools across wood → stone → iron → steel tiers, each with its own damage/mining-power stats and an always-visible description.
• Native drawing tools in survival — Shift-drag draws a straight line and Ctrl+Shift drags out and fills a whole box from your own inventory, the same gestures the sandbox's own tools use, now working against what you've actually mined and carry.
• A live-tunable world engine — day length, cave frequency, ore rarity, gravity, jump height, and move speed are all real sliders in the Esc/Options menu, not fixed constants, so the feel of the world is something you dial in rather than accept.
• A real title screen — Play/Settings/Quit before world generation, with "quit to menu" reachable mid-game without killing the process or losing your in-memory world.
• A 14-step guided quest chain — from "chop 10 wood" through smelting, forging, biome exploration, and a 50m depth milestone, up to steel tools and diamond rewards, so a new player always has a next goal instead of an empty sandbox.
• Persistent saves, a chat/log feed, and a minimap.
• Built-in bug reporting — press F8 in-game to send a bug report or suggestion straight to the dev's Discord, with automatic context (seed, day, frame) attached, no alt-tabbing or GitHub account required.
• Self-updating — the game checks this repo's releases on startup. If you're behind, it shows a changelog dialog — now with a real scrollbar you can drag, not just the mouse wheel — listing everything you've missed across every skipped release, and lets you install with one keypress; it downloads, swaps its own files out, and relaunches on its own.

Life support & machines

Underneath the survival layer is a real energy/life-support progression: descending requires sustaining yourself with machines that chain into each other, not just better gear.

• Real atmosphere engineering — bellows, air lines, compressors, and life-support panels that report whether a room is genuinely sealed or open to sky, all reading and feeding the same pressure/oxygen physics the player breathes.
• A full energy tree — solar, wind, geothermal, fuel cells, and electrolysis, plus boilers and turbines running on real steam physics (measured live: water climbs through a real 373K phase change, boils dry on schedule with the fuel that's lit under it, and a piped connection to a turbine produces real, measurable output).
• Chemistry, fluids, and logistics machines — electrolysis cells, acid synthesis, sorters and splitters, silos and quarries, all built from real reagents and real particle behavior rather than abstract recipes.
• Hazard atmosphere — carbon monoxide from fire in a sealed room, CO2 from smelting, explosive methane from swamp/oil pockets, and depth radiation all demand real mitigation (scrubbers, ventilation, lead shielding) instead of just better armor.

New physics & elements

64 custom elements on top of the 195 stock ones (measured directly from pbx-custom-elements.json and POWDER_TOY_MATERIAL_INDEX.json), several backed by real physical constants rather than arbitrary game numbers.

Antimatter & magnetism showcase — a from-scratch physics demo (inspired by Veritasium's antimatter video) with four live bays: antimatter production, magnetic containment, an annihilation chain reaction, and a shielded-vs-unshielded comparison. AMTR annihilates on contact, releasing its full mass as heat, light, and a shockwave — unless it's inside a live MGNT electromagnetic field.

Real rare-earth magnetism — GADO (gadolinium) is a genuine permanent ferromagnet below its real Curie point (~20°C) and loses magnetism above it; FEFL (iron filings) is light enough to actually be pulled and clumped by a magnetic field, unlike a solid iron bar; LUTE (lutetium) rounds out the set as the densest, hardest rare earth.

Everything else, grouped by what it's for:

Category	Elements
Reactive chemistry	Chlorine, sodium, phosphorus, acetylene, nitrogen/liquid nitrogen, ammonia paths, graphite oxidation
Nuclear & radiation	Promethium, radon, quark-gluon plasma, radiation-shielding fabric (Demron)
Electronics & power	Customizable wire, LEDs, lithium-ion batteries, thermostats, tmp/ctype/gravity sensors, digital signs, powered converters, turbines
Materials processing	Generic gas/liquid/powder/cast-solid forms so any material can boil, melt, pulverize, or cast
Structural & defensive	Load-bearing structure blocks that collapse without support, defence foam, indestructible elemental walls, rubber
Creatures & robots	A bee that pollinates and defends its hive, a robot pet that follows and fights alongside you
Weapons & mechanisms	Guided missiles, powered projectiles, custom explosives, mechanical wheels

Sandbox quality-of-life

• Sandbox / RPG-menu toggle — flip the native element menu, HUD, and toolbar on or off from one place, so the RPG's own input handling never fights the classic sandbox tools underneath it.
• Favorites wheel and custom brush shapes for faster building.
• A local, loopback-only HTTP bridge (net.listen) that exposes safe, read-only and queued-mutation actions for external tooling — the same bridge this whole feature list was verified through.

Getting the update while you play

The version you're running is always shown bottom-left. When a newer build exists, a changelog window lists exactly what changed in every release you've missed — press U to install immediately (it restarts on its own in a few seconds), or Esc to keep playing and update later; the reminder stays on screen either way. Vision

The pillar this project is built around: a survival RPG where the physics is real, not simulated-looking. Every system below — atmosphere, fire, radiation, energy — is meant to behave the way the underlying particle sim actually computes it, not a game-design approximation layered on top. That constraint is deliberate and won't loosen as the game grows; a new mechanic has to be a real consequence of real physics before it ships.

The identity, stated plainly: you are not just surviving against a hostile world — you are building a small industrial ecosystem that converts energy into breathable air, drinking water, and food, and the tech tree is the story of that ecosystem getting bigger, safer, and more self-sufficient. That's not a pitch for something planned — it's already true of the current build, assembled piece by piece across many sessions without ever being named out loud until now. Concretely, today, in shipped code: power (turbines, reactors, solar) feeds electrolysis, which splits real water into real oxygen for life support and hydrogen as a byproduct; a gas turbine or fuel cell burns that hydrogen for more power (explicitly a closed loop back to electrolysis); the water side feeds real crop farming and a desalinator turns brine into more of it. Food and water are real depleting stats (R.need.food, R.need.water), and this is the same loop that keeps them (and you) alive — closer to a Factorio / Oxygen Not Included hybrid than a generic survival-craft game, and already real.

The tech tree already reflects this: Workbench → Furnace → Anvil (where the power chain begins — O2GENKIT, turbines, and fuel cells are Anvil recipes, several gated on a real "100W generated" production threshold, not just proximity to a bench) → Research Bench → Advanced Lab, each tier requiring the previous tier's own output as an ingredient, so a real production-rate gate exists alongside every station gate. The next rungs follow the same shape, and the material work backing them is already shipped or scoped:

• Reactor tier — turns the existing fission-plant-scale UO2 / ZIRC / GRPH / B4C chain (already real, already simulated with a genuine neutron economy — UO2 emits real NEUT via spontaneous fission, ZIRC is a real neutron-transparent cladding, GRPH is a real high-conductivity neutron moderator, B4C is a real neutron absorber for control rods) into a repeatable Advanced-Lab recipe chain instead of a one-off demo build. Each intermediate is its own craftable item (Fuel Rod: UO2 + ZIRC cladding; Moderator Block: GRPH; Control Rod: B4C + STEL actuator) wrapped in a CNCR-shielded Reactor Core structure that runs the real neutron-economy simulation already proven out in the existing FSN-2 / PLUT precedent builds and wires its output to the same turbine / power-grid system everything else uses. The real design work already exists in the project's community-plant notes — this tier is packaging proven physics into a repeatable player-facing recipe chain, not inventing new reactor physics.
• Sealed-base life support — a designated enclosed area where life-support machines maintain O2 / food / water automatically for time spent there, so the base-building loop matters mechanically (not just cosmetically) and late-game power generation has an actual sink. Implementation deliberately reuses the same cheap multi-ray roomSealed(wx, wy) heuristic the game already runs every tick for the player's breathing check, sampled at a life-support machine's location plus a small ring around it — not an O(area) flood-fill, which risks exactly the oxygen-spawn-lag bug the project has already learned the hard way not to repeat. While sealed, the controller tops up a persistent base O2 / food / water pool the player draws from while inside, reusing R.o2 / R.need.food / R.need.water and their existing regen logic.
• Fluid / gas logistics tier — pipes moving WATR / HYGN / OXYG between machines without the player manually carrying it. A CONVEYOR machine already exists (buildConveyor, pushes solid / powder materials along a belt); the actual gap is narrower than "build a logistics tier from scratch," because fluids / gases don't have automation yet — existing "pipes" (e.g. the boiler's steam-takeoff pipe) are fixed structural channels built as part of one specific machine, not a general player-placeable pipe connecting arbitrary machines. Concrete proposal: placeable pipe segments (visually similar to the conveyor's belt-segment pattern) that, once connected between two machines with matching fluid ports (e.g. an O2GENKIT's HYGN output and a turbine's fuel input), move a bounded amount of that fluid per tick — the same per-tick solid-pushing shape CONVEYOR already uses, generalized to liquids and gases. This is what actually completes the closed energy / oxygen / water loop into something a player builds once and leaves running automatically, instead of manually re-carrying water and hydrogen between machines forever.

Roadmap

This section reflects the real, current state of the project — what's actually shipped and verified (grounded in R.CHANGELOG / CHANGELOG.md, not aspiration), what's designed but not yet built, and what's still just a proposal. The full running wishlist of proposed systems, including newer ones not yet promoted into this roadmap, lives in knowledge/feature-wishlist.md (internal notes, not part of this published repo) and is meant to keep growing across sessions — check there for the latest state if this section looks stale.

Shipped (selected — see CHANGELOG.md for the full version-by-version list)

• Cave generation, "empty vertical tunnels going straight down" — root-caused with real measured numbers (an entrance-tunnel centerline formula covered too little of its own noise cycle to wind naturally for the first 20–40 depth units below the surface) and fixed with a second, fast-decaying noise layer active only near the surface; deep-cave behavior is numerically unchanged. Shipped v1.15.0.
• Respawn-path-drift bug class closed — four-version saga (v1.15.7 → v1.15.10) fixing state (companion position, pollution stats, overhead cleanup) that silently didn't apply on the bed-respawn path because it bypassed the main R.spawnPlayer() entry point; closed with two shared helpers (R._resetSpawnState(), R._teleportCompanionAndClear()) called from every spawn path. Full retrospective: releases/v1.15.7-v1.15.10-retrospective.md.
• Tree water routing + underground aquifer soak (v1.15.42 → v1.15.65) — rain and standing water now visibly route down hollow tree trunks as vein lines, soak into a root-zone moisture band that feeds a small underground aquifer layer (treeAquiferAt / treeAquiferSpreadTick), and drain correctly out of the gaps between adjacent trunks instead of pooling on canopy leaves. Trunk/canopy no longer crumbles to sawdust from nearby digging or hot-reloads (v1.15.65) — though the underlying support check was still wrong at that point and was re-fixed in v1.15.74, which found a direct collapse call on a healthy surface-supported trunk destroying 397 wood cells against a limit of 25. This is real, live, bridge-verified plumbing — but it's scoped to tree-adjacent moisture, not a world-wide water table or player-facing flooding/draining/pumping system; see knowledge/current-progress.md (internal notes) for the verification detail and knowledge/feature-wishlist.md #1 for the full-aquifer-system item this precedent feeds into.
• Underground environment feel (v1.15.60 → v1.15.62) — real depth-scaled geothermal gradient, biome-specific surface temperatures, and per-cave-pocket microclimates, with the TEMP/PRESS HUD reading the full depth column instead of a flat ~72°F everywhere.
• Mining ventilation physics (v1.15.28 → v1.15.58) — newly dug cavities no longer fill with breathable air instantly; ventilation diffuses in from adjacent open air over ~50–90 ticks, mined-cavity water no longer floats mid-air, and dig-pressure lag was root-caused and optimized.
• HUD legibility pass (v1.15.24 → v1.15.47) — multiple rounds closing real overlap bugs (Day/GOAL text collision, minimap crowding, Esc-menu column bleed, TEMP/PRESS gauge labeling) rather than a single redesign; each round is its own verified fix.
• Vehicles: minecart + rail + mine-lift kits (v1.15.26) — buildable rail network (LMB-drag snapped track), rideable minecart, and a powered vertical mine-lift cage.
• Inventory UX overhaul (v1.15.28, UI-INV-1/2/3) — Carried tab rebuilt from a scrollable text list into a real bordered grid with Terraria-style left-click pick/place semantics and a stack-merging SORT.
• Fail-closed visual verification harness (v1.15.33 → v1.15.38) — goal_verify.py / verify_claim.py with a mandatory screenshot gate, because lastErr == nil was repeatedly proven not sufficient on its own (three real bugs — drawMenu, wrap, HUD row overlap — all shipped with lastErr=nil and passed the old, bridge-only checks). See knowledge/verification-system.md (internal notes).

Near-term (designed or actively in flight)

• Full aquifer system — extend the tree-water precedent above into a real world-wide water table (connected cavern volume below one global depth fills with real WATR via TPT's own hydrostatic settling, replacing today's isolated noise-placed deep-cavern pockets), tunnels that fill/drain in response to player digging, and wells/pumps as a craftable hook into it. Proposed, not yet designed in full — see knowledge/feature-wishlist.md #1 (internal notes).
• Real geological layering for subsoil / bedrock — currently the bottom of the world keeps defaulting to fired brick, which is structurally fine but cosmetically wrong. Researched against actual soil science (topsoil → subsoil / regolith → bedrock) and this engine's real element physics. The recommended V1 fix is a single substance swap: BSLT (real basalt, id 503), which is verified genuinely solid (Falldown = 0 AND bit.band(props, elem.TYPE_SOLID) ~= 0) and is already a real registered element, so no new elements.allocate cost. An earlier attempt swapped in STNE instead and the generated terrain collapsed — caught and reverted in the same session (STNE is Falldown = 1, i.e. a falling powder, not solid rock). The BSLT swap itself is researched and spec'd but not yet applied to the world-gen fallback; full fix awaits a re-verified code change.
• Structural cave-ins — wide mining tunnels become genuinely dangerous: ceiling blocks over a large unsupported span convert to a falling-powder copy after a short warning delay, using TPT's own unsupported-granular-fall physics, no new elements. Proposed — see knowledge/feature-wishlist.md #5 (internal notes).
• Flash-flood caverns — some deep-cavern water pockets are sealed behind a thin rock membrane; breaking through with a pick lets TPT's real pressure/gravity sim flood the newly opened tunnel instead of just revealing static water. Proposed — see knowledge/feature-wishlist.md #6 (internal notes).
• Behavior-kind persistence — a few of the power / reactor elements (turbine, thermoelectric, reactive concrete) are defined through custom behavior kinds that are not yet re-registered on restart, so they currently only work in a live dev session rather than a fresh launch — everything else in the elements table above survives a restart intact. This is being driven from a single apply_realism_modules call against the lab instance so it can be re-tested end-to-end before claiming it's fixed.
• Item quality system — crafted tools rolling a quality tier that affects their stats, not just their tier.

Mid-term (proposed, grounded in existing mechanics, not yet designed)

• Landmarks / points of interest — placement shipped, discovery not. A 30-structure library (cabins, wells, campsites, watchtowers, mineshaft junctions, shrines, sealed vaults, reactor ruins, crystal chambers, plus 10 small props), authored from 42 analysed community saves, had been wired into nothing; v1.15.72 loads and places all 30, and v1.15.78 retuned flatness tolerance so they survive the new mountain terrain (3.7 buildings + 8.9 props per 3000px). What remains is making them worth finding — rewards and a discovery hook — e.g. a rare deep-cave VIRS outbreak pocket (cured by fire, chest reward gated on infection clearing) or a PSCN/NSCN logic-gate puzzle vault, both built entirely from existing element behaviors. See knowledge/feature-wishlist.md #2 (internal notes).
• Biome variety expansion — visual and mechanical variety beyond the current forest/desert/snow/swamp set, pairing real geological strata bands (granite, sandstone, limestone with genuine acid reactivity) with per-biome survival mechanics instead of just re-skinned terrain. See knowledge/feature-wishlist.md #3 (internal notes).
• Mountains / vertical terrain (v1.15.70) — shipped as the predicted low-frequency surface-height noise pass: a long-wavelength layer (period 1100px) hard-thresholded to zero across most of the map, so plains stay plains while 22% of columns rise into real ranges. Measured on seed 7, total relief went 61px → 280px across 8000 columns with the steepest slope still 2px per column, so peaks stay walkable without digging. Topsoil was deepened 20px → 50px in the same bump so a hillside can actually be dug into. Applies to newly generated terrain only. See knowledge/feature-wishlist.md #4 (internal notes).
• Portal-pipe item logistics network — placeable PRTI/PRTO portal pairs (real stock teleport-by-channel mechanic) for Terraria-style long-distance item pipes, complementing the fluid/gas logistics tier below. See knowledge/feature-wishlist.md #7 (internal notes).
• Snow-biome hypothermia + insulated bases — real cold-drain tick outside snow-biome shelter, countered by real insulator material (AERO) and a TEG that passively generates power at a warm/cold boundary. See knowledge/feature-wishlist.md #8 (internal notes).
• Meteor strike events — rare scheduled falling hot-rock event that craters terrain and leaves a rare-ore impact site, giving the world something that can happen during play rather than only be discovered. See knowledge/feature-wishlist.md #9 (internal notes).
• Reactor tier, sealed-base life support, fluid/gas logistics tier — the next three rungs of the industrial-ecosystem tech tree; see the Vision section above for the full grounding (each already reuses existing, live-tested element physics rather than inventing new ones).
• More biomes and quest content past the current 14-step starter chain.

Long-term (real architecture research done, large scope, not started)

• Co-op multiplayer — real architecture research done, not just an ask. This class of falling-sand simulation cannot do peer-to-peer lockstep (physics is chaotic and iteration-order-dependent — two machines running the "same" step on the same data diverge within a few ticks from float rounding and update-order differences alone, and exhaustive deterministic-math engineering is fragile and expensive). The two real precedents confirm this: Noita Together deliberately did NOT attempt shared-simulation multiplayer (each player keeps a separate world; see the wiki / GitHub), and Noita Entangled Worlds (GitHub) only ships a proxy-relay plus per-pixel delta sync — i.e. exactly host-authoritative, not lockstep. Recommended shape for this codebase: host-authoritative — one machine runs the real simulation untouched, every other player connects as a thin client sending only input (movePlayer / useTool / placeAt, all already exist) and receiving back a bounded region per client reusing the existing camera-scroll windowing system (the same windowed approach already used by the tile-cache / shiftCam infrastructure, so no new windowing scheme is invented). V1 scope: LAN-only, host + one remote player, no client-side prediction / rollback — accept host-latency lag on remote inputs in exchange for a much smaller, ship-able surface. Each player is genuinely just a second instance of the existing P-like state table driven by network input instead of local mouse / keyboard — which is the same actor-command shape companion.lua already separates ("decide what to do" vs "execute the action"), so the actor layer is real, existing, reusable infrastructure, not a new one. Not started.
• Physics-driven character overhaul + dismemberment — real technique identified: Verlet integration with distance ("stick") constraints, the same method Happy Wheels and Source-engine ragdolls use (going back to Thomas Jakobsen's paper for Hitman: Codename 47; reference: Tuts+ verlet ragdoll tutorial). Body is ~11 point masses (head, chest, pelvis, 2× upper arm, 2× forearm, 2× thigh, 2× shin); each point stores current and previous position (Verlet needs no explicit velocity — it is implicit as pos − prevPos); rigid areas get multiple constraints per joint so they don't fold, loose areas (elbows, knees) get exactly one. 4–8 constraint-relaxation iterations per frame looks convincingly rigid; collision reuses the same solid-style check the player already uses today, just once per point instead of once for the whole player box. Dismemberment is just a conditional constraint break: each stick gets a breakForce, and the simulation removes a stick whose current stretched length exceeds its rest length by more than breakForce for the rest of the ragdoll's life. The two sides then drift away under existing Verlet motion — this is "limb tears off" without any special-case code path, and decapitation is the neck stick breaking with the same mechanism; real BLD (blood) particles already in the codebase spawn at the break point for free. Recommended V1 cut: ragdoll only activates on death or a heavy hit, not full-time movement control — normal walking / running / jumping stays exactly as it is today (the current sprite + velocity model already feels responsive, and replacing it wholesale is real risk for no clear gain). Respawn resets back to normal sprite control. Explicitly out of V1: full-time ragdoll-driven walking, a real skeletal / inverse-kinematics rig, organs as separate simulated bodies (spawn as particle effects / decals at death instead — cheaper, same visual payoff). Not started; awaiting confirmation that "death-only ragdoll" matches the intended scope vs. always-on ragdoll movement.
• Real biome-varied geology (V2) — properly distinct layers per biome: granite under mountains, sandstone under deserts, limestone / shale variation, a real saprolite transition band at the topsoil / bedrock boundary. Each new layer would be its own custom element following the same elements.allocate("RPG", "NAME") pattern as GRSS / BLD, and every one would need its own Falldown == 0 AND TYPE_SOLID verification before being trusted as structural fill — the STNE regression generalized to a real rule. Flagged as the V2 scope, not V1, per the design doc.

For the respawn-path-drift saga spanning v1.15.7 → v1.15.10 (one bug class, four versions, helper extraction pattern established), see the cross-version retrospective at releases/v1.15.7-v1.15.10-retrospective.md and the individual release pages linked from its summary table.

Full version-by-version history lives in CHANGELOG.md. The full, growing wishlist of proposed systems (including anything added after this roadmap pass) lives in knowledge/feature-wishlist.md (internal notes, not part of this published repo).

Running from source

Standard Powder Toy build (Meson + Ninja); see the Powder Toy Development Help wiki page for toolchain setup. The RPG layer is pure Lua (scripts/lua/, autorun-compiled from bridge_src/) and need no rebuild — only the new elements above require a native rebuild.

Credit & license

Built on The Powder Toy by Stanislaw K Skowronek and the TPT team — see powdertoy.co.uk and the official forum for the original game, online saves, and the wider community. Distributed, like the original, under the GNU General Public License v3.

Windows quick start

Download the PowderRPG-v1.18.1-windows-x64.zip asset, extract the entire folder, and run PowderRPG.exe. Keep the DLL files beside the executable. The RPG Guide opens with L and closes with L or Esc.