v0.14.2 · engine stable

Sub-microsecond state,
in every process.

Attos compiles your read-heavy state into a memory-mapped artifact and decodes it with an off-heap Apache Fury engine. Zero-allocation deserialization, O(1) lookups, and no network hop between your service and its data.

Start a proof of concept See the benchmarks GitHub

P99 lookup: 181 ns
Allocations / op: 0 B
Heap at 15M keys: 0 MB
Uptime SLA: 99.99%

state distribution pathlive

01Control planerevision r1482

02Attos Binary Compiler174 MB artifact

03Zonal replicaus-east1-b

04Page cachemmap, evictable

05Your processzero-copy read

last read181 ns

Compliance posture

AES-256-GCM

At rest + in transit

GDPR Compliant

EU data residency

99.99% Uptime SLA

Financially backed

90-day Key Rotation

HSM-backed KMS

VPC Peering

Single-tenant isolation

Benchmark Suite · 15M Keys

Built for Ultra-Low Latency & Zero OOM Risk at 15M+ Scale

515ns

P99 Tail Latency (Rust)

Sub-microsecond lookups

0MB

Hot-Path Heap Allocation

Zero GC pause risk

174MB

Total Pod RAM @ 15M Keys

Up to 15x smaller footprint

< 40ms

Cold-Start File Load Time

Instantaneous initialization

Total Pod RAM required for 15M keys

lower is better

Attos (Rust/Go)174 MB

ChronicleMap1,606 MB

BigCache / FreeCache1,946 – 2,610 MB

ConcurrentHashMap / DashMap1,745 – 2,867 MB

Explore full benchmark methodology & docs

Architecture

Zero-copy state distribution, end to end.

Six stages from a committed write to a nanosecond read. Select a stage to see what runs and what it costs.

Stage 01 / 06

Source of truth commits

Postgres, Kafka or your control plane emits a changeset. Attos treats every write as an immutable revision — there is no in-place mutation anywhere in the pipeline.

revision r1482

Capacity planning

Model the resident memory before you provision.

Attos pages stay in the OS page cache, which the kernel can evict under pressure — so they never count toward the cgroup limit that produces Exit Code 137.

Dataset size

10,000,000 keys

1M100M

Lookup latency (ns)

AttosO(1)

181ns

FastCacheO(1)

552ns

LMDBO(log N)

930ns

BBoltO(log N)

1,991ns

RedisO(1) + RTT