VoidSentryUltimate

IAMNOTULTRA3 · August 22, 2026, 4:38am

Github
Docs
Wally
Pesde
Npm
Creator Store

VoidSentryUltimate is a schematized serdes library, unlike my previous serdes libraries, this one is mainly focused on pure performance, and it has a lot of cool optimizations

Code example

local VoidSentryUltimate = require(path.to.VoidSentryUltimate)
local Types = VoidSentryUltimate.Types

local PlayerState = Types.Struct({
    Health = Types.U16,
    Name = Types.String8,
    Position = Types.Vector,
})

local encoded = VoidSentryUltimate.Serialize(PlayerState, {
    Health = 100,
    Name = "Builder",
    Position = Vector3.new(4, 8, 15),
})

local decoded = VoidSentryUltimate.Deserialize(PlayerState, encoded)
print(decoded.Name, decoded.Health)

According to my benchmarks, VoidSentryUltimate is 1.5-3x faster than sera on non-struct types(its a bit unfair), and pretty much equal on struct types

Why Use VoidSentryUltimate?

Rich Types Support – More than 40 types
Very Fast – Optimized for peak performance (bench results; benchmark)
Straightforward API

Why not use VoidSentryUltimate?

No schemaless serialization – You have to define the exact schema (if you want schemaless serialization, you might like VoidSentry2)
No runtime validations

Types Catalog

Primitives

Type	Description
`U8` / `I8`	1-byte unsigned / signed integer
`U16` / `I16`	2-byte unsigned / signed integer
`U24` / `I24`	3-byte unsigned / signed integer
`U32` / `I32`	4-byte unsigned / signed integer
`F16`	2-byte reduced-precision float (lossy)
`F24`	3-byte reduced-precision float (lossy)
`F32`	4-byte float
`F64`	8-byte float
`Bool`	One boolean in 1 byte
`BoolPacked`	Exactly 8 booleans packed into 1 byte
`String`	2-byte length prefix, then 0–65,535 bytes
`String8`	1-byte length prefix, then 0–255 bytes
`StringFixed(length)`	Exactly `length` bytes, no prefix
`Buffer`	2-byte length prefix, then 0–65,535 bytes
`Buffer8`	1-byte length prefix, then 0–255 bytes
`Buffer24`	3-byte length prefix, then 0–16,777,215 bytes
`BufferFixed(length)`	Exactly `length` bytes, no prefix
`Vector`	`Vector3`, three `F32` (12 bytes)
`VectorF16`	`Vector3`, three `F16` (6 bytes)
`VectorF24`	`Vector3`, three `F24` (9 bytes)
`VectorU8` / `VectorI8`	`Vector3`, three integers (3 bytes)
`VectorU16` / `VectorI16`	`Vector3`, three integers (6 bytes)
`VectorU24` / `VectorI24`	`Vector3`, three integers (9 bytes)

Collections

Type	Description
`Array(node)`	Array with 2-byte element count
`Array8(node)`	Array with 1-byte element count
`Array24(node)`	Array with 3-byte element count
`ArrayFixed(node, length)`	Fixed length, no count prefix
`Map(key, value)`	Map with 2-byte entry count
`Map8(key, value)`	Map with 1-byte entry count
`Map24(key, value)`	Map with 3-byte entry count
`MapFixed(key, value, length)`	Fixed length, no count prefix
`Struct(fields)`	Dense named fields, no struct prefix
`DeltaStruct(fields)`	Sparse struct; `u8` count/ids (max 255 present)
`DeltaStruct16(fields)`	Sparse struct; `u16` count/ids (max 65,535 present)
`Optional(node)`	1-byte presence marker + value when present

Roblox Types

Type	Description
`Vector2`	Two `F32` components (8 bytes)
`Vector2F16`	Two `F16` components (4 bytes)
`Vector2F24`	Two `F24` components (6 bytes)
`Vector2U8` / `Vector2I8`	Two integers (2 bytes)
`Vector2U16` / `Vector2I16`	Two integers (4 bytes)
`Vector2U24` / `Vector2I24`	Two integers (6 bytes)
`UDim2`	Scale + offset as four `F32` (16 bytes)
`UDim2Scale`	Scale only as two `F32` (8 bytes)
`UDim2Offset`	Offset only as two `F32` (8 bytes)
`UDim2OffsetI16`	Offset only as two `I16` (4 bytes)
`UDim2ScaleF24`	Scale only as two `F24` (6 bytes)
`Color3`	RGB as three bytes
`Enum(enumType)`	Enum item value in 2 bytes
`Enum8(enumType)`	Enum item value in 1 byte
`CFrame`	Position + 3×3 matrix, 12×`F32` (48 bytes)
`CFrameF16`	Same layout with `F16` (24 bytes)
`CFrameF24`	Same layout with `F24` (36 bytes)
`QCFrame`	Position + quaternion, 7×`F32` (28 bytes)
`QCFrameF16`	Same layout with `F16` (14 bytes)
`QCFrameF24`	Same layout with `F24` (21 bytes)
`Instance`	2-byte `_VSID` reference
`Instance24`	3-byte `_VSID` reference
`SerInstance(className, schema)`	Serialize listed properties; deserialize via `Instance.new`

API Catalog

API	Description
`Serialize(schema, value)`	Encode into a new exactly sized buffer
`Deserialize(schema, buffer)`	Decode from byte 0
`SerializeWithOffset(schema, value, offset)`	Encode into a new buffer with `offset` zero bytes reserved up front
`DeserializeWithOffset(schema, buffer, offset)`	Decode starting at `offset`
`Push(schema, value, buffer, offset?)`	Encode into an existing buffer (default offset 0); buffer must already be large enough
`SetWriteBufferSize(size)`	Resize the reusable serialize scratch buffer (default 1,000,000 bytes)

IAMNOTULTRA3 · August 23, 2026, 2:56am

Just benched VoidSentryUltimate against Squash, Sera, Z, VoidSentry2, BlazeSentry, results:
Bench Repo

All libraries comparison (structs, primitives, cframes)

__________________________________________________
* RANKINGS (lower median μs is faster)
______________________________________
* TOP 6 OVERALL (all tests)
** #1 VoidSentryUltimate — 0.24μs med avg (57 tests)
** #2 Z — 0.25μs med avg (57 tests)
** #3 BlazeSentry — 0.29μs med avg (57 tests)
** #4 Sera — 0.30μs med avg (57 tests)
** #5 VoidSentry2 — 0.40μs med avg (57 tests)
** #6 Squash — 0.48μs med avg (57 tests)
______________________________________
* TOP 6 OVERALL — RoundTrip only
** #1 VoidSentryUltimate — 0.36μs med avg (19 tests)
** #2 Z — 0.38μs med avg (19 tests)
** #3 BlazeSentry — 0.44μs med avg (19 tests)
** #4 Sera — 0.45μs med avg (19 tests)
** #5 VoidSentry2 — 0.60μs med avg (19 tests)
** #6 Squash — 0.70μs med avg (19 tests)
______________________________________
* TOP 6 OVERALL — Serialize only
** #1 VoidSentryUltimate — 0.15μs med avg (19 tests)
** #2 BlazeSentry — 0.18μs med avg (19 tests)
** #3 Z — 0.18μs med avg (19 tests)
** #4 Sera — 0.21μs med avg (19 tests)
** #5 VoidSentry2 — 0.32μs med avg (19 tests)
** #6 Squash — 0.40μs med avg (19 tests)
______________________________________
* TOP 6 OVERALL — Deserialize only
** #1 Z — 0.20μs med avg (19 tests)
** #2 VoidSentryUltimate — 0.20μs med avg (19 tests)
** #3 Sera — 0.23μs med avg (19 tests)
** #4 BlazeSentry — 0.25μs med avg (19 tests)
** #5 VoidSentry2 — 0.28μs med avg (19 tests)
** #6 Squash — 0.32μs med avg (19 tests)
______________________________________
* TOP 6 — Number Types
** #1 VoidSentryUltimate — 0.05μs med avg (24 tests)
** #2 Z — 0.07μs med avg (24 tests)
** #3 BlazeSentry — 0.08μs med avg (24 tests)
** #4 VoidSentry2 — 0.09μs med avg (24 tests)
** #5 Sera — 0.10μs med avg (24 tests)
** #6 Squash — 0.15μs med avg (24 tests)
______________________________________
* TOP 6 — Number Types (RoundTrip)
** #1 VoidSentryUltimate — 0.08μs med avg (8 tests)
** #2 Z — 0.11μs med avg (8 tests)
** #3 BlazeSentry — 0.12μs med avg (8 tests)
** #4 VoidSentry2 — 0.15μs med avg (8 tests)
** #5 Sera — 0.15μs med avg (8 tests)
** #6 Squash — 0.20μs med avg (8 tests)
______________________________________
* TOP 6 — Other Primitives
** #1 VoidSentryUltimate — 0.08μs med avg (15 tests)
** #2 BlazeSentry — 0.11μs med avg (15 tests)
** #3 Z — 0.11μs med avg (15 tests)
** #4 VoidSentry2 — 0.12μs med avg (15 tests)
** #5 Sera — 0.13μs med avg (15 tests)
** #6 Squash — 0.22μs med avg (15 tests)
______________________________________
* TOP 6 — Other Primitives (RoundTrip)
** #1 VoidSentryUltimate — 0.12μs med avg (5 tests)
** #2 Z — 0.16μs med avg (5 tests)
** #3 BlazeSentry — 0.17μs med avg (5 tests)
** #4 VoidSentry2 — 0.18μs med avg (5 tests)
** #5 Sera — 0.20μs med avg (5 tests)
** #6 Squash — 0.31μs med avg (5 tests)
______________________________________
* TOP 6 — Vector Types
** #1 VoidSentryUltimate — 0.05μs med avg (3 tests)
** #2 Z — 0.07μs med avg (3 tests)
** #3 BlazeSentry — 0.08μs med avg (3 tests)
** #4 VoidSentry2 — 0.09μs med avg (3 tests)
** #5 Sera — 0.10μs med avg (3 tests)
** #6 Squash — 0.18μs med avg (3 tests)
______________________________________
* TOP 6 — Vector Types (RoundTrip)
** #1 VoidSentryUltimate — 0.07μs med avg (1 tests)
** #2 Z — 0.10μs med avg (1 tests)
** #3 BlazeSentry — 0.12μs med avg (1 tests)
** #4 VoidSentry2 — 0.14μs med avg (1 tests)
** #5 Sera — 0.14μs med avg (1 tests)
** #6 Squash — 0.27μs med avg (1 tests)
______________________________________
* TOP 6 — CFrame Types
** #1 VoidSentryUltimate — 0.10μs med avg (6 tests)
** #2 Z — 0.12μs med avg (6 tests)
** #3 BlazeSentry — 0.15μs med avg (6 tests)
** #4 Sera — 0.15μs med avg (6 tests)
** #5 VoidSentry2 — 0.17μs med avg (6 tests)
** #6 Squash — 0.32μs med avg (6 tests)
______________________________________
* TOP 6 — CFrame Types (RoundTrip)
** #1 VoidSentryUltimate — 0.14μs med avg (2 tests)
** #2 Z — 0.18μs med avg (2 tests)
** #3 BlazeSentry — 0.23μs med avg (2 tests)
** #4 Sera — 0.23μs med avg (2 tests)
** #5 VoidSentry2 — 0.25μs med avg (2 tests)
** #6 Squash — 0.48μs med avg (2 tests)
______________________________________
* TOP 6 — Struct Types
** #1 Z — 1.12μs med avg (9 tests)
** #2 VoidSentryUltimate — 1.16μs med avg (9 tests)
** #3 Sera — 1.24μs med avg (9 tests)
** #4 BlazeSentry — 1.29μs med avg (9 tests)
** #5 VoidSentry2 — 1.93μs med avg (9 tests)
** #6 Squash — 1.98μs med avg (9 tests)
______________________________________
* TOP 6 — Struct Types (RoundTrip)
** #1 Z — 1.69μs med avg (3 tests)
** #2 VoidSentryUltimate — 1.75μs med avg (3 tests)
** #3 Sera — 1.88μs med avg (3 tests)
** #4 BlazeSentry — 1.98μs med avg (3 tests)
** #5 VoidSentry2 — 2.91μs med avg (3 tests)
** #6 Squash — 2.99μs med avg (3 tests)
__________________________________________________

Maps & Arrays comparison (Sera is excluded since it doesnt support them)

__________________________________________________
* RANKINGS (lower median μs is faster)
______________________________________
* TOP 5 OVERALL (all tests)
** #1 VoidSentryUltimate — 0.80μs med avg (18 tests)
** #2 Z — 0.80μs med avg (18 tests)
** #3 BlazeSentry — 0.89μs med avg (18 tests)
** #4 VoidSentry2 — 1.56μs med avg (18 tests)
** #5 Squash — 1.75μs med avg (18 tests)
______________________________________
* TOP 5 OVERALL — RoundTrip only
** #1 VoidSentryUltimate — 1.20μs med avg (6 tests)
** #2 Z — 1.21μs med avg (6 tests)
** #3 BlazeSentry — 1.34μs med avg (6 tests)
** #4 VoidSentry2 — 2.37μs med avg (6 tests)
** #5 Squash — 2.64μs med avg (6 tests)
______________________________________
* TOP 5 OVERALL — Serialize only
** #1 VoidSentryUltimate — 0.54μs med avg (6 tests)
** #2 Z — 0.58μs med avg (6 tests)
** #3 BlazeSentry — 0.61μs med avg (6 tests)
** #4 VoidSentry2 — 1.43μs med avg (6 tests)
** #5 Squash — 1.57μs med avg (6 tests)
______________________________________
* TOP 5 OVERALL — Deserialize only
** #1 Z — 0.61μs med avg (6 tests)
** #2 VoidSentryUltimate — 0.66μs med avg (6 tests)
** #3 BlazeSentry — 0.70μs med avg (6 tests)
** #4 VoidSentry2 — 0.88μs med avg (6 tests)
** #5 Squash — 1.04μs med avg (6 tests)
______________________________________
* TOP 5 — Array Types
** #1 VoidSentryUltimate — 0.55μs med avg (9 tests)
** #2 BlazeSentry — 0.59μs med avg (9 tests)
** #3 Z — 0.60μs med avg (9 tests)
** #4 VoidSentry2 — 1.04μs med avg (9 tests)
** #5 Squash — 1.08μs med avg (9 tests)
______________________________________
* TOP 5 — Array Types (RoundTrip)
** #1 VoidSentryUltimate — 0.82μs med avg (3 tests)
** #2 BlazeSentry — 0.89μs med avg (3 tests)
** #3 Z — 0.89μs med avg (3 tests)
** #4 VoidSentry2 — 1.58μs med avg (3 tests)
** #5 Squash — 1.63μs med avg (3 tests)
______________________________________
* TOP 5 — Map Types
** #1 Z — 1.01μs med avg (9 tests)
** #2 VoidSentryUltimate — 1.05μs med avg (9 tests)
** #3 BlazeSentry — 1.18μs med avg (9 tests)
** #4 VoidSentry2 — 2.08μs med avg (9 tests)
** #5 Squash — 2.42μs med avg (9 tests)
______________________________________
* TOP 5 — Map Types (RoundTrip)
** #1 Z — 1.52μs med avg (3 tests)
** #2 VoidSentryUltimate — 1.58μs med avg (3 tests)
** #3 BlazeSentry — 1.80μs med avg (3 tests)
** #4 VoidSentry2 — 3.15μs med avg (3 tests)
** #5 Squash — 3.66μs med avg (3 tests)
__________________________________________________

TLDR; VoidSentryUltimate is overall fastest, Z slightly beats it in Structs and Maps

Yarik_superpro · August 23, 2026, 3:43am

ANYTHING but learning to use buffer library atp
Its not that hard to divide stuff by 8 mentally

scooter_user · August 23, 2026, 3:46am

as an average low-level programmer this topic is made entirely out of words i do not understand

IAMNOTULTRA3 · August 23, 2026, 4:01am

what are those words you dont understand?

3xp0x3d · August 23, 2026, 4:08am

Good work, I’ll definitely use this in my upcoming projects.

scooter_user · August 23, 2026, 2:22pm

all of them, like i just said

cybercrazyyy · August 23, 2026, 5:39pm

what kinda name is VoidSentryUltimate tho

IAMNOTULTRA3 · August 23, 2026, 11:59pm

I just randomly came up with Voidsentry, and this library is pretty much a third rewrite of the original VoidSentry so I added Ultimate

IAMNOTULTRA3 · August 29, 2026, 11:00am

v1.0.1

Added DeltaStruct, DeltaStruct16 types
Micro improvements

soggybird0 · September 1, 2026, 1:02am

nice work, ill use this for sure

IAMNOTULTRA3 · September 3, 2026, 3:09pm

V1.0.2:

Deserialize, DeserializeWithOffset, Push now return the cursor

soggybird0 · September 4, 2026, 10:38pm

pretty useful ill keep that in mind

IAMNOTULTRA3 · September 8, 2026, 12:11pm

im thinking on maybe adding more precision types (such as more quantized cframes, udim2, etc.), new schema object and micro optimizations (i missed some)