Game Data Browser – Must have tool for every FiveM developer – search & live-preview 453,528 assets

Installation

1. Copy lastresort_gamedata into the resources directory.
2. Add this to server.cfg:

set onesync on
ensure lastresort_gamedata
add_ace group.admin lastresort_gamedata.browse allow
add_ace group.admin lastresort_gamedata.elevated allow

3. Restart the server, then open the browser with the configured command or key.

Configure

• Use Config.Permissions.Identifiers for individual access if you do not use ACE groups.
• Keep browse and elevated access separate. Elevated access controls weapons, explosions, interiors, and visible previews.
• Adjust the open key, search limits, preview settings, and theme in the config files.

Verify

1. Open the browser and search for a model or animation.
2. Run a safe preview and then Stop all to confirm cleanup.
3. Test an elevated category with an allowed admin and a regular player.
4. Use /gdb report if the browser, permissions, or data index does not load.

Important

• No framework or database is required.
• Set Config.Dev.Enabled and Config.Dev.BypassPermissions to false on a live server.
• Elevated access includes dangerous previews, so grant it only to trusted admins.

215 paths classified: 73 previewable, 5 search-only, 137 excluded. all 46 shipped categories are wired into shared/categories.lua.

RESEARCH — lastresort_gamedata

Milestone M0. Every number in this document was measured on the checked-out data, not estimated. Where something could not be verified it is marked ASSUMPTION and repeated in section 9.

1. Provenance and acquisition

Nothing is fetched at runtime and no raw dump ships inside the resource. The build pulls three repositories as blobless, no-checkout git clones and then checks out only the paths actually needed, per the product line data policy.

tools/fetch_dumps.sh performs the acquisition. It deliberately skips assets/, nodes.zip and ps3-builds.xml, which together are ~5 MB of blobs no shipped index draws from. objectslocations/ is fetched — it was skipped until the World Locations category was added; see section 2.

Build pipeline, in order:

./tools/fetch_dumps.sh
python3 tools/native_index.py --natives build-refs/natives --cfx build-refs/fivem/ext/native-decls --out tools/native_index.json
python3 tools/distill.py --dumps build-refs/dumps --out data/index --enums tools/native_enums.json
python3 tools/coverage.py --dumps build-refs/dumps --index data/index > docs/coverage-matrix.md
python3 tools/verify_snippets.py --index tools/native_index.json
python3 tools/check_natives.py --index tools/native_index.json

The shipped resource depends on none of it at runtime.

2. Coverage matrix

The full per-file matrix is generated, not hand-written: docs/coverage-matrix.md. tools/coverage.py lists the upstream repository itself and exits non-zero if any path lacks a verdict, so the matrix cannot drift when the dumps change.

All 215 paths are classified. The verdict split moved during development as categories were added — docs/coverage-matrix.md is generated from the code and is always current; the numbers below are illustrative of the shape, not a running total.

The remaining exclusions are less severe than the number suggests — 66 of them are the assets/ images:

speeches.json was initially excluded at 10.4 MB, then brought in under the same lazy shard scheme the animation clips use: 1,083 voices and 81,206 distinct line names, playable through PlayPedAmbientSpeechWithVoiceNative. Vehicle handling and waypoint recordings joined it.

objectslocations/ followed, and the reversal is worth recording because the original exclusion reason was sound for the question it answered. As props these are positions of models already browsable by name, which is what the first verdict said. As places to stand, a coordinate is the entire point, and 33,912 of them are exactly the data a scripter building a fuel, robbery or jukebox script has to source by hand today. Two measurements decided it: the same lazy shard scheme keeps the always-loaded index flat (+2,827 B for the 34 group rows, the 1.16 MB of placements fetched only when a group is opened), and every one of the 33,912 coordinates but four falls inside a bounding box in zones.json, so each gets a real map-zone label instead of a bare number.

3. Index sizes after distillation

Measured output of tools/distill.py. Categories are stored columnar, with low-cardinality string columns replaced by an integer index into a per-column palette.

Largest single categories: pedcomponents 1.23 MB, animdicts 743 KB, objects 566 KB.

The shard split is the decision that makes the budget work. Indexing all 269,414 clips, 81,206 speech lines and 33,912 world placements inline would more than quadruple the always-loaded payload for data that is only meaningful once a specific dictionary, voice or location group is selected. Clips and speech shard by first character (data/index/clips/<char>.json, data/index/speeches/<char>.json); placements shard by group (data/index/locations/<group>.json), because a group is exactly the unit the browser asks for. What stays always-loaded is the 34-row group index, 2,827 bytes.

4. Where the index lives, and the search approach

4.1 Transport

The index ships as plain files listed in files{} and is fetched by the NUI page directly. A FiveM NUI page is served from https://cfx-nui-<resource>/, so a relative fetch('data/index/vehicles.json') resolves to the resource's own files. This is verified against working code already in this repo: lastresort_vehicle_inspection/web/js/app.js:12 derives its resource name by stripping the cfx-nui- prefix from window.location.hostname.

This avoids the alternatives entirely: no SendNuiMessage payload carrying megabytes of JSON, no LoadResourceFile on the client, no server round-trip. The Lua side never parses the index unless a script asks for it through an export.

4.2 Search algorithm — measured, not assumed

Benchmarked in Node 22 over the real distilled index:

Cold start, measured on the 45-category 1.0 index: 36 ms to read and parse every core file, plus 25 ms to build the lowercase haystack — ~61 ms total. The 46th category added since costs 2,827 bytes and 34 rows, so the figure is unchanged in practice; the placement shards are not in the cold path at all.

Incremental narrowing measured separately: typing t → toolbox (7 keystrokes) costs 25 ms in total, because each keystroke filters the previous result set rather than rescanning.

Conclusion: a naive indexOf scan is fast enough. No trigram index, no inverted index, no Web Worker. The measurement says the simple implementation wins, and the anti-slop contract says to prefer it. Clips stay out of the default cross-category haystack and are searched in an explicit clip mode, which keeps the common path in the 1.5–5.6 ms band.

4.2b Rendering, not searching, was the real cost

Measured in Chromium against the live GUI, per keystroke, end to end:

The scan was never the bottleneck — building up to 1,000 DOM rows was. The results list now renders a 120-row window that follows the selection, which is 7× faster at the worst case and puts every keystroke inside one 60 fps frame.

A trigram index was considered and rejected on evidence: these names share a tiny alphabet of machine-generated morphemes (prop, ent_, _01, mp_), so only ~13,000 distinct trigrams exist across the whole corpus and the index has almost no discriminating power. On the most common real query in an object browser — prop_ — posting-list intersection is slower than the plain scan.

ASSUMPTION-1: these timings are Node 22 on the build machine. FiveM's CEF runs an older V8 and competes with the game for CPU. Even allowing a conservative 3–4× slowdown the core index stays well inside one 60 fps frame. To be confirmed with the in-game debug overlay at M1.

4.3 DLC filtering

Only some dumps carry a DlcName. Measured distinct values:

The brief's DLC filter therefore applies to those categories only. The GUI must hide the filter rather than show an empty one where the field does not exist.

5. Snippet verification

The brief's non-negotiable rule is that every generated example runs as pasted. tools/verify_snippets.py enforces the mechanical half of that as a build step, so the in-game paste test only ever has to confirm behaviour:

1. Placeholders are filled from the real distilled index, so what is verified is exactly what a customer copies.
2. Every native call is checked against the 6,089-signature index built from the CitizenFX docs — existence, exact spelling, and argument count.
3. The filled snippet is compiled with luac5.4, the runtime FiveM targets.
4. The filled snippet is executed against tools/fivem_stub.lua, catching undefined locals, nil arithmetic, missing load waits, and loops that never terminate.

Arity checking models a real FiveM rule: Lua callers omit pointer out-parameters, which come back as extra return values. char* is an input and stays. The checker therefore accepts a range rather than a single count.

Status: 42 templates, 0 failing.

The checker is proven to fail on bad input — five deliberately broken snippets are all caught with precise diagnostics:

Three real defects were found and fixed by this process during M0:

• LoadInterior does not exist. The correct native is PinInteriorInMemory.
• GetVehicleColours was called without accounting for its int* out-parameters.
• shaders.json silently distilled to zero entries because the generic string-list builder was pointed at a dump of objects. The builder now raises instead of emitting an empty category.

ASSUMPTION-2: the stub proves a snippet executes; it cannot prove the game does the visible thing. Per-category in-game paste tests are the M1–M3 acceptance gate and are listed in docs/TESTING.md.

6. Corrections to the upstream README

The shipped indexes follow the measured values. _lod entries are dropped because they are level-of-detail meshes, not props anyone spawns deliberately.

ASSUMPTION-3: the two upstream count mismatches are most likely because the free dump is a reduced variant of the paid one and because soundNames.json changed format upstream. Either way the shipped indexes describe the files as they actually are.

7. Data shape findings that change the product

Discovered while distilling, each one affects what the GUI can offer:

• Vehicle mods have no names. vehicleModKits.json mods carry only {Index, Type, Bone, Weight, IsWeapon}. There is nothing to substring-search. Mods are therefore browsed through their kit and mod type, and the kit name is the searchable string. Liveries are the exception — those are named, so they ship as their own searchable category (78 entries).
• Ped clothing is freemode-only. All 84 collections resolve to mp_m_freemode_01 or mp_f_freemode_01. Component ids are meaningless on other ped models, so the clothing preview requires a freemode mannequin and the snippet warns when the target ped is a different model.
• RelativeCollectionDrawableId differs from DrawableId. Both ship; the absolute id is what SetPedComponentVariation wants, and the relative id is what DLC-collection-aware code needs.
• Ped names are cased inconsistently upstream (A_C_Boar in peds.json, lowercase in most scripts). The index normalises to lowercase, which is what GetHashKey needs.
• Hashes ship signed. peds.json and vehicles.json expose both; weapons expose IntHash. All verified to match after conversion.

8. Repo conventions and sibling detection

Surveyed across all seven shipped lastresort_* resources. What this product adopts, and what the survey caught it getting wrong:

Four defects the survey caught in this resource, all fixed:

1. exports['lastresort_chat']:AddMessage(...) — the real export is lowercase addMessage, is server-side only, and takes (target, message). The client call would have failed silently forever. Now uses the chat:addMessage event, which lastresort_chat provides.
2. The GTA feed fallback was missing the mandatory 99-character clamp.
3. bridge/server.lua called framework exports with no pcall and returned a job name as an admin group. A job is not an admin group; the QBCore path is gone and the ESX path is wrapped.
4. ACE names were bare gamedata.browse / .elevated; house style is lastresort_gamedata.browse / .elevated.

The NUI focus rule

The most important convention in the repo, learned from a real bug: release focus unconditionally. Never gate SetNuiFocus(false, false) on your own open flag or on the shared owner bag.

SetNuiFocus is tracked per resource, so releasing can only ever drop your claim — there is nothing to protect anyone else from. Gating the release is how a cursor gets stuck for the rest of the session. The shared lrNuiOwner statebag is the only part that takes an ownership check, and only on clear, because that value is shared.

This product's first implementation had the gate. It was removed.

Collision check

gdb as a command prefix, F10 as a keybind, and lastresort_gamedata.* as ACE names collide with nothing in the suite. Exports are PascalCase and domain-explicit (Search, Lookup, GetCategories, GetSnippet, OpenBrowser) to stay clear of the shared vocabulary already in use (Notify, Open, Close, GetHistory, …).

This product overlaps no sibling's features, so unlike lastresort_npc_chat under lastresort_intel_npc it never self-shuts-down.

9. Open ASSUMPTION items

10. Preview method and cleanup guarantee, per kind

Every previewable category maps to a handler in client/preview/. The registry keeps exactly one preview per admin and journals everything it touches, so the guarantee is the same everywhere: switch, close, disconnect, resource stop and panic all drain the same journal.

11. Adversarial review

The preview methods were re-checked by an independent pass whose only brief was to find what is wrong. Three of its findings were confirmed against the native documentation and fixed in the shipped code:

1. SetVehicleDoorsLocked(vehicle, 4) does not lock a preview vehicle. State 4 is VEHICLELOCK_LOCKED_PLAYER_INSIDE — "locked once a player enters, preventing others from entering" — which is not what a display vehicle wants. Now 2, VEHICLELOCK_LOCKED.
2. The particle cleanup swept a 25 m radius unconditionally. RemoveParticleFxInRange destroys effects owned by other resources. It ran on every teardown, including the ones where the handle had already been stopped cleanly — so every preview would have silently killed other scripts' vehicle smoke, fire and weather effects within 25 m of the admin. The sweep is now the fallback for an orphaned handle only.
3. Preview start had no re-entrancy guard. start yields while streaming, so two rapid activations could interleave and the second would purge the journal of the first mid-load.

The first attempt at a fix was a busy latch that refused the second start, and it had a bug of its own: stop() cleared the very flag start() had just set, so the guard never engaged. tests/preview.test.lua caught that. The latch was then replaced outright, because refusing the newer preview is the wrong behaviour — an admin arrowing down a list expects the latest entry to win. The shipped guard is a monotonic generation token (client/preview/registry.lua): the second start proceeds, the first notices its token is stale, purges only what it created, and returns silently. tests/preview.test.lua covers all three fixes in 16 assertions.

Findings that did not apply to the shipped code, because the journal design already handles them: streamed assets are tracked before the bounded wait, so a load timeout releases rather than leaks; and onResourceStop, disconnect and panic all drain the same journal.

12. Status of this document

Complete. Sections 1–7 and 9 were verified directly against the checked-out data; section 8 against the seven shipped sibling resources; section 10 against the CitizenFX native documentation, with each trap traced to the specific native doc that records it.

STORE — lastresort_gamedata

Tebex listing copy. Nothing here uses Rockstar or Next GTA logos, artwork, screenshots or trademarks.

Title

Game Data Browser — search, preview and copy any GTA V asset, in game

Short description

Stop alt-tabbing to wiki lists. Search 453,562 asset names from inside your server, see any prop, ped, vehicle, animation or effect standing in front of you a second later, and copy a working code example with the names already filled in.

Long description

The loop this ends

You need a toolbox prop. So you open a wiki list, guess a name, alt-tab back, spawn it, get nothing, and guess again. Twenty minutes for one prop.

With this installed you type toolbox, arrow down, press Enter, and it is standing in front of you. Press Alt+C and a working snippet is on your clipboard. Seconds, not minutes, without leaving the game.

Everything, in one panel

• 21,561 props · 921 vehicles · 1,109 peds · 184 weapons
• 20,179 animation dictionaries holding 269,414 clips
• 1,083 speech voices holding 81,206 lines
• 2,907 particle effects · 1,088 timecycle modifiers · 173 screen effects
• 30,396 clothing components · 5,213 ped props · 3,373 tattoos and badges
• 602 mod kits · 838 handling profiles · 895 IPLs · 385 interiors
• Plus scenarios, clipsets, pickups, explosions, camera shakes, zones, garages, waypoint routes and the full audio set

Search that keeps up with you

Every keystroke scans the whole index in a few milliseconds. Results are ranked — exact match first, then prefix, then the rest — and grouped by category, so searching toolbox reads as 8 props and 1 animation dictionary, not nine undifferentiated rows.

No loading spinner. No pagination. Scroll as far as you like.

Preview in the world, not in a thumbnail

The interface sits down the sides of your screen on purpose: the middle is where the preview goes.

Props and vehicles spawn on the ground in front of you, frozen and collision-free, rotatable and zoomable, swapping instantly as you arrow through results. Previewing a ped turns you into it — arrow down the list and you change character every keypress, which is how you actually judge a model. Your own character is captured first and handed straight back when you stop. Animations and scenarios play on a mannequin — or on your own ped. Weapons are held, not dropped in the grass. Particle effects fire at a marked point. Timecycle filters apply live and revert on stop. Clothing cycles on a freemode mannequin. Handling profiles put you in the driver's seat of a car that actually uses them. Waypoint routes are drawn across the map.

Only you see it. Previews are local and hidden from the network unless you deliberately enable the visible-preview flag for collaborative scene work. The one exception is a ped preview, which changes your own character and is therefore visible to everyone, as any model change is — Config.Preview.PedsOnSelf = false puts it back to a separate figure standing in front of you.

Closing the panel leaves the preview standing, so you can walk around it.

33,912 places, sorted by how far away they are

Every ATM, gas pump, vending machine, bus stop, jukebox, dartboard, telescope, oil jack and park bench in Los Santos, at its real coordinate, pulled from Rockstar's own placement data. Open World Locations, pick a group, and the list is ordered nearest you first, each entry naming the map zone it sits in — 121 ATMs, 157 gas pumps, 345 vending machines, 48 bus stops, 6 jukeboxes. Press Enter and you are standing at it, in your car if you were driving.

This is the part you cannot get from a wiki: building a fuel script, a robbery script or a jukebox script means sourcing those coordinates by hand, one screenshot at a time. /gdb waypoint takes you to any pin you drop on the map, and /gdb back returns you to where you started.

Code examples that actually run

42 of the 46 categories ship a minimal, self-contained Lua example with the real names already filled in — load-and-play for an animation, create-and-place for a prop, asset-and-start for an effect. The other four are flag and reference sets that no native takes by name, and the entry says so instead of inventing one.

These are verified, not assumed. Before release, every native call in every template is checked against 6,089 signatures from the official CitizenFX documentation, then each filled snippet is compiled and executed. Five deliberately broken snippets are kept in the build to prove the check still fails when it should.

Pick entries as you go, favourite the ones you keep returning to.

Leak-proof, by design

Every preview journals what it touched — entities, streamed models, animation dictionaries, particle handles, weapon assets, IPLs, pinned interiors, waypoint recordings, audio banks, screen effects, world toggles, and your own character down to its clothing, head blend and face features — and reverts all of it when you switch entries, hit Stop all, disconnect, or stop the resource.

/gdb debug shows that journal live, so you can watch it return to zero rather than take our word for it.

Built to fit

• Standalone. ESX, QBCore and QBox are detected automatically and never required. No database.
• 0.00 ms when the browser is closed and nothing is previewing.
• Two permission levels, so juniors can browse props while weapons, explosions and world edits stay locked. Permissions are decided server-side, and everything that leaves the machine is re-checked on arrival.
• Every category can be switched off in config and vanishes from the GUI.
• Recolour it from config to match your own branding, no CSS editing.
• Scales to your screen — 1366×768 through 4K.
• Fully keyboard-operable, with a visible focus ring on every control.
• Open source. No escrow, no obfuscation, no fxap. Read every line.
• A documented query interface so your own scripts can search the same index.
• Rebuildable. When a new GTA V build lands, the distillation tooling ships with the resource.

Feature bullets

• 453,562 asset names across 46 categories, searchable instantly in game
• 41 categories preview in the world; the other five say so up front
• 33,912 real world positions on top of those — ATMs, gas pumps, bus stops, jukeboxes — reached by picking their group and listed nearest you first, with the map zone each one stands in
• One-click preview with rotate, zoom and instant result swapping
• A verified copy-ready Lua example for 42 of the 46 categories
• Picks, favourites and a live activity timeline
• Panic stop that clears every preview, effect and filter server-wide
• Live debug overlay and an AI-ready diagnostic report command
• Demo tour mode for recording your own trailer
• Standalone with automatic ESX / QBCore / QBox detection
• Open source, no escrow

Suggested trailer shots

/gdb demo tour is a single-take scripted sequence built for exactly this.

1. Cold open. Standing in an empty lot. /gdb — the panels fade in, the dusk gradient catching the top edge. Two seconds, no narration.
2. The promise, in one shot. Type toolbox letter by letter; the count falls from 102,942 to 9 as the fingers move. Press Enter. The prop appears in the empty centre of the screen.
3. Flipping pages. Hold the down arrow through the props category — a new prop in the world every frame, like flipping a catalogue. This is the shot that sells the product.
4. The snippet. Cut to the detail pane, press Alt+C. Overlay: "pasted straight into a fresh resource — it runs." Then show it running.
5. Scale montage. Vehicles (a supercar drops in), Peds (you become each one as you arrow down), Animations (a mannequin starts dancing), Particles (fire erupts), Timecycles (the world turns amber).
6. Clothing cycle. Arrow through ped components; the mannequin changes outfit each keypress.
7. The panic. Several effects running, screen filter on, particles going. Press Delete. Everything stops in one frame. Cut to /gdb debug showing the journal at zero.
8. Close. Wordmark on black with the dusk gradient. "Game Data Browser — LastResort Development."

Record at 1080p or higher; the GUI scales with the screen and stays legible when downscaled for a store page.

Tags

fivem developer tools admin props animations particle effects standalone esx qbcore qbox open source data browser

Suggested price band

Developer tooling with a one-off install and no per-player runtime cost. Price against admin and developer utilities rather than gameplay scripts.

Support notes for the listing

• Requires OneSync. No database, no dependencies.
• Ships with the ACE lines a buyer needs to paste into server.cfg.
• /gdb report produces a self-contained diagnostic file; asking for that file answers most support questions in one round.

TESTING — lastresort_gamedata

Each item is a command plus the observable result it must produce. Everything runs with default config and no arguments.

M0 — Build verification

M0 ships no runtime code, so its checklist verifies the build pipeline and the data it produces. Run from the resource root.

1. Acquire the reference data

./tools/fetch_dumps.sh

Expected: build-refs/dumps, build-refs/natives and build-refs/fivem exist. build-refs/dumps contains no assets/, objectslocations/, nodes.zip or ps3-builds.xml — those are deliberately not checked out.

2. Build the native signature index

python3 tools/native_index.py --natives build-refs/natives --cfx build-refs/fivem/ext/native-decls --out tools/native_index.json

Expected: 6049 natives indexed (±, as CitizenFX documents more), and tools/native_enums.json: eExplosionTag (85). The command exits non-zero if fewer than 5000 natives parse, which means the clones are incomplete.

3. Distill the indexes

python3 tools/distill.py --dumps build-refs/dumps --out data/index --enums tools/native_enums.json

Expected, per the shipped build:

• 45 categories, no category reporting 0 entries — World Locations is the 46th and is written by the next step
• total 12248035 B raw, 1567622 B gzipped
• the warning 11 explosion tags have no documented eExplosionTag value
• animclips 269414 entries across 23 shards

3b. Build the World Locations index

python3 tools/locations.py build-refs/dumps data/index

Expected:

• locations 34 groups 2827 B 1442 B gz
• locationplaces 33912 entries 1211789 B (34 shards, loaded on demand)
• skipped empty dumps: worldExtraVendingMachines
• a +zone line for each of ipls (892/895 named), interiors (375/385), garages (32/32) and staticemitters (908/930)
• total 13471752 B raw, 2003949 B gzipped, across 46 categories

Re-running it is a no-op: the zone columns are added only once, and a second run leaves every file byte-identical.

4. Prove the coverage matrix is exhaustive

python3 tools/coverage.py --dumps build-refs/dumps --index data/index

Expected: exits 0, ends with a 215 paths classified: … line, and confirms all 46 shipped categories are wired into shared/categories.lua.

Exits non-zero if the upstream repository gained a file with no verdict, or if a shipped category has no dump mapped to it.

5. Verify every snippet

python3 tools/verify_snippets.py --index tools/native_index.json

Expected: exits 0, ending with

42 templates, 0 failing
46 shipped categories: 42 with snippets, 4 documented as snippet-less

6. Prove the verifier actually fails

The check above is only worth trusting if it rejects bad input. Each of these must be reported, and each must make the command exit non-zero:

Restore the file afterwards and confirm the command exits 0 again.

7. Confirm the runtime tables agree

python3 tools/check_natives.py --index tools/native_index.json
lua5.4 tools/check_defaults.lua

Expected: 0 unknown native calls across the resource's own Lua, and defaults.lua and categories.lua agree. Both exit 0; the second fails if a category is added to shared/categories.lua without a default, which is how a new category silently vanishes on a server that kept its own config.lua.

8. Confirm the distilled data round-trips

python3 - <<'EOF'
import json, os
index = "data/index"
manifest = json.load(open(os.path.join(index, "manifest.json")))
source = {d["DictionaryName"]: d["Animations"]
          for d in json.load(open("build-refs/dumps/animDictsCompact.json"))}
total = 0
for shard in manifest["clipShards"]:
    for name, clips in json.load(open(f"{index}/clips/{shard}.json")).items():
        assert source[name] == clips, name
        total += len(clips)
print("clip shards round-trip exactly:", total)
EOF

Expected: clip shards round-trip exactly: 269414.

9. Run the unit tests

for t in tests/*.test.lua; do lua5.4 "$t"; done

Expected: four files, 46 assertions, none failing. They cover the notification fallback, the permission ladder, preview supersede and cleanup, and ground placement per entity type.

M1–M3 — In-game checklist

Run with default config on a bare server. Grant yourself both ACE permissions first (see INSTALLATION.md).

Opening and search

Preview and cleanup

World locations and teleporting

The panic stop

Snippets

Permissions

Demo, report and siblings

Performance

Changelog

2026-09-02

• 1.1.0: World Locations, and the engine's own teleport.

2026-08-22

• Make Config.Keybind.Open actually move the key, and open the ACE gate.
• Document that the keybind default is only a default.

2026-08-17

• 1.0.2: ped previews become you; fix the model-swap frame.
• Stop the black boxes, scale with the screen, and stop typing from walking.
• Three defects from the adversarial review, and tests that hold them.
• Remember a broken notification provider, and index the CFX natives we were missing.
• Two ways past ACE, and move the keybind off F7.

2026-08-11

• Initial release with one searchable browser for GTA models, animations, effects, interiors, audio, and world data.
• Added ranked global search with grouped results and on-demand loading for large animation and speech datasets.
• Added live previews for props, vehicles, peds, weapons, animations, particles, timecycles, interiors, zones, and audio.
• Added verified, copy-ready Lua examples for supported categories.
• Added session history, persistent admin favorites, and server-wide preview cleanup.
• Added /gdb stopall, /gdb debug, /gdb demo, and /gdb report tools.
• Added client and server exports for search, lookup, category data, snippets, and browser control.
• Added automatic ESX, QBCore, QBox, notification, and sibling-resource detection.
• Added ACE, identifier allowlist, and separate elevated permissions.
• Added keyboard navigation, theme configuration, rate limits, and validation for client payloads.

1. v1.1.0LatestSeptember 2, 2026
   • 1.1.0: World Locations, and the engine's own teleport.

Must have tool for every FiveM developer! Easy to use!

YouTubein actionLoads from YouTube when you press play

Hi everyone! My next big script is made for developers, for faster and more detailed server development. You press F10 (or whatever key/command you set) and you can instantly search locations, effects, animations, sounds, props, vehicles, clothes — 453,528 assets with detailed data. Build your server from native GTA content: it’s already in the game, it doesn’t eat extra resources, and it’s what enriches the gameplay the most!

• 453,528 asset names in 45 categories, instant in-game search
• Live preview in the world for 40 categories — rotate, zoom, instant swap as you move through the results
• You see the code right away: Alt+C copies working Lua code with the real names already filled in (41 categories, verified against 6,089 native signatures from the official CitizenFX docs)
• Pick list + favorites + Lua export for your current project, no database needed
• Full reset with one key — you don’t litter while testing
• Standalone, auto-detects ESX / QBCore / QBox, 0.00 ms while closed
• Two permission levels with server-side checks
• Built-in LastResort Notify integration (ox_lib and chat fallback)
• Full, unencrypted source — no escrow, no obfuscation, no fxap

Commands in short: F10 open · Enter preview · Alt+C copy code · Alt+T pick list · Alt+F favorites · Delete full reset · more /gdb commands in the docs

Requirements: OneSync. No database and no dependencies needed (lastresort_notify is optional).