Three months, 153 commits, ten people. This is the first dev to master push since May and it is not a small one, so rather than list every commit I want to talk about the three things that will actually change how you build.
Points Rework Points has been a contentious topic, as we moved from the original RDC algorithm of raw per-entity values and tuning, to a new system designed by Ally to use traces to determine armor values, and finally onto the current system also designed by Ally to combine RDC’s per-entity point cost with an algorithmic approach to determining performance.
We also added a manufacturing cost you can mess around with! It’s currently purely for show, and will get more accurate as we introduce more components!
The actual specifics are rather complicated, so a technical changelist will be provided at the bottom of this post.
Scalable Explosives There is a new entity in ACE, the scalable explosive! This is a scalable entity with several shapes that explodes based on its size. This is an HE-based explosion, and can overpressure contraptions. These can also be manually detonated via an input, but watch out; they will explode when destroyed as well. We also added fixed-size aerial, barrel, and satchel explosives just for fun! Keep in mind, that they're configured to carry less HE compared to their volume. It is an experimental feature, so keep the feedback coming!
Quality of Life Improvements A few small things that we think the players will like.
- Your server will feel much nicer with the huge lag improvements to HE, ballistics traces, IRST, and extra crash protection.
- Ammo, fuel, and radar entities default to being active, so no need to wire active on start anymore.
- Guns now display their proper ammo types and ready status when out of ammo.
General Optimizations, Fixes, and Improvements We made a lot of improvements to the ACE code you won’t even see, but here are a few things you might notice:
- Rubber’s spall eating was heavily nerfed, removing it’s special behaviour. Instead: it eats a percentage more energy from spall than it would have as the equivalent thickness of other materials.
- Missiles now properly ignore visclips.
- Primitives now behave properly, including if clipped.
- Suppression is nerfed for weapons below 10mm.
RDC's patch, finally RDC wrote a set of balance changes back in April 2025 and never committed them. They sat. This cycle we pulled them out of his working copy and shipped them, because they were good and because leaving somebody's work to rot is a bad habit.
- Rubber is genuinely dangerous to shoot into now. Kinetic resilience up from 0.5 to 0.7, catch resilience from 1.0 to 1.25, and spall now runs through the ordinary kinetic resolver instead of a flat 0.75 multiplier.
- Ceramic damage reduced , and the ERA effective-thickness buff undone. Both were overperforming.
- AT rifles lost their magazines , with a small cost reduction to match. They are single-shot again.
- HVAP drag reduced — it holds velocity further out.
That is the balance half of what he had. There is more to come next dev cycle and we're really it's going to come sooner than this patch. Stay tuned.
New entities
- Kh-39 LMUR — a 2011-era Russian ATGM, 200mm, 105kg, top speed 141 m/s, 667mm max kinetic penetration. Inertial guidance and datalink, no ECCM. Guidance options are Dumb, Infrared, Top Attack IR and Laser, so you can build a designator setup around it. Ships with 1x and 2x racks.
LMUR Missile
- New rack model
LMUR Rack
- Zaslin APS — model and textures are in. APS logc to come later, feel free to build around with it.
ACE Zaslin APS model No more AI slop? — New contribution guidelines for contributors & developers. Seeing Wiremod large discussion also transfered towards us. You can read the statements from all sides here, however now. We do not prohibit the use of AI on our repo, however:
Authors must:
- Be honest and disclose when AI tools were used.
- Use AI only to help with coding. PR text and replies must be written by the author; maintainers need to be talking to the contributor, not an AI toolchain.
- Understand the submitted code well enough to explain it, justify it, audit it for side effects, and make requested changes themselves.
- Make requested changes that actually fix the problem, rather than blindly resubmitting tool output.
- Keep each contribution focused on one issue or a closely related set of issues per PR. Large all-in-one PRs are not accepted when AI is doing the coding.
AI use does not relax ACE's style, cleanliness, or review requirements. Slop is slop regardless of how it was made.
Maintainers are not LLM code auditors. Breaking these rules, or giving maintainers reason to believe they are being broken, may result in a repository ban.
Besides that, we're now moving towards making it easier to developer ACE or for ACE.
Here are some examples how we do that:
Public functions you add or change (on ACE. / ACF. , or entity/SWEP methods) get an LDoc --- doc-comment:
--- Computes kinetic penetration for a projectile impact.
-- @param velocity number Impact speed in source units/s.
-- @param mass number Projectile mass in kg.
-- @return number Penetration depth in mm RHA.
function ACE.CalcPenetration( velocity, mass ) And developers are now recommended to write tests, so we make sure that no new change breaks old logic harshly, so we detect such issues earlier.
What's coming Several things are sitting in open drafts right now:
- Sustainability (#266) — the big one. Electricity and oil as real systems: alternators, solar panels, transformers, power lines with real resistance, capacitors, breakers, and power collectors so overhead-wire trams and trains work. Plus oil pumps, refineries, fuel pipes that leak when damaged, fuel plugs and sockets so you can build actual gas stations, and a universal fuel tank that takes whatever you pour into it. The idea is to make building a base or a logistics chain worth doing — right now there is no reason to. However it came to a silient end, we need somebody who actually understands how these systems work irl, it takes a lot of time, so pretty much archived.
- RDC mobility rework (#270) — the radiator logic, customizable engines and exclusive torque curves to engines & the other half of RDC's shelved work. Requires RDC to come back to finish what he started
- In-place gun updates (#319) — change a gun in the menu without breaking your welds.
- Guns not hitting their own contraption (#302) . After so many years, that issue will get finally patched.
- Asset size reduction (#321) — the addon is large. This makes it considerably less large.
Important note: another change that is incoming is ACF -> ACE namespace switch. So if you're making an addon creator, then make sure that it will be supported in the upcoming ACE update in the next cycle, that one will be huge and will help us long term.
Technical Changelog Onto the longest section! A vast majority of the work we did was fully back-end.
- Points system behaviour moved into sh_ace_points_model.lua .
- Ammo is given value by round lethality, seeker type, and post-penetration.
- Guns and racks multiply that cost linearly by sustained RPM
- HE is now converted into HE-Equiv, which is a KE threat level equivalent number that is used in the math.
- Armor HP pricing is now linear per prop
- Shared helpers are exposed for every value.
- Manufacturing cost introduced in sh_ace_manufacturing.lua .
- Cost is based on armor, including mass and material.
- Engines get their armor base cost, plus a mobility cost based on horsepower and fuel.
- Gun costs are based on caliber squared and type of gun
- Rack cost is based on tube count.
- Ammunition cost is based on mass and warhead.
- Missiles’ cost is based on body mass plus seeker/guidance.
- Electronics cost is based the on ACE point tier of th component.
- There is also costs for crew seats, refill crates, and explosive charges
- The output is categorized into armor, powerpack, armament, ammunition, electronics, and crew.
- Contraption lifecycle is now event-driven
- Central invalidation/lifecycle boundary for contraptions added.
- Point state is invalidated on relevant changes; links, ammo, ROF, racks, physics state, or ownership
- Affected contraptions are deduplicated, with cache generation tracked
- We maintain this through CFW transitions to keep a consistent family
- Primitive armor is also captured and properly transferred
- ACE functions are moved to the most part to the ACE. namespace, but legacy aliases are still available for certain functions
- Ballistics is bounded and schedulable
- Active bullets are tracked in a registry
- Ballistic work is processed per-frame in bounded batches
- Damage queries have explicit limits (crash prevention)
- HE candidate selection has a fast path.
- Spall tracing is now bounded (crash prevention)
- Missile visibility/fuze tracing has explicit retry behaviour
- Scaleable explosives
- Insert section here
- ACE now has automated tests
- LuaJIT model tests have been added
- There are also now tests for various regressions and features, such as but not limited to:
- Missile launches
- Primitive armor registration
- Points invalidation
- GLuaTest cases have also been added for the following cases
- Entities
- Registries
- Lifecycle hooks
- Armor
- Spall
- General-use contracts
- Python scanners and performance runners have been added
- Repository structure and safety checks have been added
- ACE ammo, fuel tanks, missiles radars, and various other sensors are now active by default, with shared active-state handling.
- Miscellaneous new content
- LMUR missile with custom 1x and 2x racks
- Zaslin APS model
- Vignette effect assets
- Project and contribution infrastructure improved
- Contributor documentation added
- Agent-development rules added
- Manual-only canary deployment added
- Expanded automated quality checks
- More explicit test preparation and validation workflows
Website update Thought that it's better to provide those updates here, rather than an its own blog post on its own.
For a while now the website has been the place to find the ACE wiki, the builds library, and the dev docs. This update changes what the site is . The Discord bot officially joins the family — that alone is a lot — and alongside it comes build contests, community events, partner wikis, and a much faster homepage. Fair warning: this is a long one, because there's a lot to unpack.
The New ACE Bot — Now in Any Discord Server This is the headline. Before this update there was no bot at all; now there's a full-time ACE assistant that can live in your server, answer questions, keep an eye on your builds, and bridge the gap between Discord and the website.
Ask the knowledge base, in plain language The core trick: the bot can search the entire ACE knowledge base from inside Discord. Wiki articles, forum threads, Stuff Street builds, blog posts, dev docs, servers, even contests and events — one query searches all of it.
/ask how does penetration work gets you the top results as a clean card with buttons. But the best part is when it's confident: instead of links, you get the answer — a pulled wiki section, or a full function reference with signature, parameters, and a link straight to the source code.
The search is the same engine the website uses, so it understands the site's own quirks. Ask "find the AC-76 dupe" and it knows you're looking for a build. Add +wiki , +docs , +upload , +server to narrow to one category. Add +gmod and it searches the official Garry's Mod developer wiki instead. Please note, that it is not an "AI" or LLM by default. Perfomance may vary by the search query. Bigger DB = better result.
Just talk to it You don't need to learn slash commands.
- @mention it in any server: @ACE Bot what does this function do gets the same treatment as /ask . Mention pings and role pings never trigger it, so nobody accidentally summons the bot in a big announcement.
- DM it. The bot answers privately too, and it handles the small talk — say "hi" or "thanks" or "what is ace" and it responds like a person, because answering a search bar with "hi" would be weird. Even verification works in DMs: type verify in a private message and it issues your code right there.
Deep answers straight from the source code This one's a little magic. Append +ai (or +deep ) to any question and the bot reads the ACE source code itself to answer — the kind of question basic search can't handle most of the time, unless it exists on the wiki. it costs nothing and needs no key. In partner org servers, that answer comes from their repo instead — the bot feels like it belongs to that community.
A live board for every ACE server A live card for every ACE server showing players, map, gamemode, version, and a working Connect button that actually launches the game. The cards update by themselves, offline servers collapse into a quiet grey line, and if a new server appears it joins the board automatically. No human maintaining it.
There's also a simpler /list if you just want a compact rundown of every server, and single-server status embeds for the one you care about.
Builds flow between Discord and the site Stuff Street's audience is Discord, and the bridge goes both ways:
- Discord to site: post a build in a watched channel and it becomes a Stuff Street listing — first screenshot becomes the thumbnail, and your later file posts build up the version history automatically. Even your old posts get picked up if the bot was offline when you made them.
- Site to Discord: upload a build on the website and it appears in the watched channel, tagging you if you're verified.
- Ownership is protected. If you aren't verified yet, your build is parked under your name, not orphaned — run /verify once and every build reserved for you is claimed instantly. And if the auto-detection grabbed the wrong attachment (a screenshot instead of the dupe), right-click the message and pick "Use as build file."
Verification, done properly /verify links your Discord to your ACE account. The code is issued right in Discord, you paste it on the site, and from then on your account, your builds, and your notifications are connected. It also fixes a legacy problem: older accounts have a Discord handle that was never actually proven to belong to them.
Notifications, wherever you are /notifications on connects your ACE notifications to Discord. Mentions, replies, new content from people you follow, announcements — instead of only living on the website, they arrive as DMs. It's done thoughtfully: notifications are staggered and presence-aware, so if you're actively on the site right now you won't also get pinged in Discord. The bot delivers in-app, browser push, and Discord DM in order, skipping what you don't need.
The help forum and the issue tracker now talk to each other This is the quiet one that saves people's sanity. New threads in a watched help forum automatically become tracked issues on the website — no manual copy-paste. From inside the thread, devs can drive the whole lifecycle: /issue status , /issue resolve , /issue assign , /issue priority . Resolving applies the Solved tag and archives the thread; reopening brings it back. When a dev answers and it goes quiet, the bot pings the reporter to check if it's actually fixed; when it's fixed, it asks them to confirm. Solved on Discord means history, not backlog.
Each Discord tag can be configured to do its own thing — BUG threads auto-track, Suggestion threads wait for a dev to promote them, and some tags get auto-answered from the knowledge base. All of it is configurable from the admin panel, no code changes.
Keep your announcements channel alive /ace-updates points a channel at the site's activity feed — new builds, wiki edits, articles, forum threads — and you choose which types show up. A community can keep its #announcements channel fresh with site activity automatically.
Feedback buttons on everything Every bot answer carries "Did this help?" buttons. If the bot got you sorted, great; if not, your question is logged as a knowledge gap and routed to a place a human actually reads — and org teams can see the questions their community's players had trouble with, which is a direct list of what to document next. Basically supposed to create a large database of the issues players have had with bot answer that could help us work on new wiki articles and such. Potentially, even making a training data for existing llms to make it go public, so that your favorite llm would actually help you, instead of just straight up hallucinating.
Built to stay up A few things under the hood worth knowing about: the bot checks with the site before answering and fails open if the site is unreachable (a hiccup never bricks the bot), every query is rate-limited per user and per server, and suspicious queries get flagged and penalized rather than ignored.
Build Contests — A Real Stage for Your Builds The website finally has what a screenshot channel never could: a proper home for build contests, where every entry is a permanent gallery card with its own link.
Coming soon (it already exists, still beta)
Community Events — "Wait, that was today?" Solved Events have a proper page now, and the automation handles the parts that used to fall through the cracks. One page carries everything: gamemode, banner, host-written rules and schedule tabs, the play server's connect command, a Steam collection to preload, and the Discord invite.
Say you're going with one tap. An hour before start, everyone who said they were going gets reminded — even people who didn't subscribe, if they RSVP'd. When it goes live, the connect command jumps to the front of the page: one click from announcement to in-game. Afterward, anyone who showed up can drop screenshots or YouTube clips into the community gallery, so the event lives on as a page instead of evaporating into chat.
Can be hosted by partner communities, server owners.
A Bigger Wiki, and It's Not Just Ours Anymore The wiki's core experience got better, and then it got bigger than ACE.
Smarter browsing. The root view defaults to a flat, searchable grid (most people arrive looking for one specific page), with a "Folders" toggle for the organized view. Sub-pages live under their folder, and the folder counts include everything beneath them.
Partner wikis, mirrored automatically. Projects keep their docs on their own infrastructure — PAC3 on DokuWiki, StarfallEx on a GitHub wiki — and asking them to re-type everything here would never stay in sync. So we don't. The site pulls partner wikis in automatically and re-syncs them on a schedule, keeping their structure (DokuWiki's folder trees, GitHub's sidebar) intact. You read them with the same layout and search as the rest of the wiki; the project team edits where they always have, and the mirrored pages tell you honestly: "This is synced from X, contribute there." No more copy-paste docs rotting in place.
PAC3 wiki is now available on our website (acegmod.com) PAC3 wiki is now available on our website (acegmod.com)
StarfallEx docs, fully browsable. The entire StarfallEx standard library is now in the docs — every function from its Libraries, Types, and Hooks, organized into folders so a 2,000-function namespace isn't one endless list.
Real contributor roles. The old contributor application could approve someone who then hit a wall at "write" — the role granted nothing usable. Now there are two real tiers: Wiki Contributor can write and publish their own articles; Wiki Editor can additionally edit anyone else's ACE wiki article — deliberately scoped so it can't touch org content or non-wiki pages. Org staff maintain their own org's content through the same system, and admins always retain access.
Org spaces keep growing. Partner communities get their own branded slice: wiki, blog, build space, and now their own developer docs. And there's a nice search payoff: when a partner's Discord server uses the bot, that org's own content ranks higher in the results there — so a Wiremod server asking about Wiremod gets Wiremod answers first.
Expect to see more orgs soon.
The Homepage and Under-the-Hood Speed The homepage is rebuilt around what's happening now — a "Happening Now" panel that prioritizes live events and open contests, the community feed, and a cleaner showcase of recent videos and builds. The hero loads much faster thanks to smarter image handling.
Speed was a real focus: the site's own fonts were hosting ourselves with aggressive caching (the old external font chain cost over a second on mobile connections alone), large pieces of code are split so pages load only what they need, and search results come back with quick answers instead of bare links. Search engine visibility and accessibility both improved across the board. In short: the site feels meaningfully snappier, especially on phones.
It seems like website is bare minimum HTML website but it really isn't, super fast and comfortable to use.
The Boring But Important Stuff The unglamorous work that keeps the site running well:
- Role management was audited and tightened — a security review found the role-management authorization was too coarse, and it's now properly gated. JSON-LD output is HTML-escaped to prevent injection.
- Repeat offenders get handled. Accounts that rack up multiple critical security violations in a short window are now auto-banned, instead of accruing warnings until someone noticed.
- A dependency cleanup dropped the vulnerability count from 36 to 12, with a modern browser list to match. More fixes and updates to backend to come soon.
Thanks This cycle: Ally-for-Hire , Agentgollem , OrangeFox861 , KemGus , MartyX5555 , RDC , MistakenHyena and Cheezus .
As always — if something here breaks for you, open an issue here, or on GitHub or say so on the Discord. The sooner we hear about it, the sooner it is fixed.