JAVATEL Inc. has delivered surveillance camera systems,
video management, and media streaming infrastructure across Japan
since 1997.
■ Genetec's longest-tenured authorized partner in Japan, the global VMS leader
■ XEOMA AI NVR for distinctive small- and mid-scale deployments
■ Planet Networks (Taiwan) network gear
■ JAVATEL-built servers / workstations
■ DMS (Docokame Media Streaming) — redefined with a RESTful API and a modern UI.
Headquartered in Nishinomiya, Hyōgo. The Video Intelligence in our company logo is not a recent trend — it is the direction we have walked for more than 29 years. Our current frontier consists of three patent-pending pillars: the NormMAP commutator-norm-driven codec engine, deterministic cross-language transpilation, and phase-synchronized deterministic parallelism (PSDP).
4K, at 1.5 Mbps.
The eye cannot tell.
| H.264 plain | 25 – 50 Mbps |
| AV1 plain (AOM) | 12 – 15 Mbps |
| NormH.264 | 6 – 8 Mbps |
| NormAV1 | 1.79 Mbps |
Measurement: CityHall 3840×2160 60fps 10s, AV1 source transcoding, VMAF 84.36 (mean) / 79.22 (min). Perceptual limit 0.97 Mbps at VMAF 75.99 — see the live demo.
-
Commutator norm
Per-block ‖A·B − B·A‖F = perceptual saliency. Math finds where the eye looks. -
Bit redistribution
Take from flat regions, give to texture/edges. Per‑block QP offset, 8×8 granularity. -
Codec-native dispatch
libx264 ROI side-data + SVT‑AV1--roi-map-file. No new decoder, anywhere it plays. -
Source-aware transcoding
Preserves source artifact profile. Even already-AV1 input compresses 8× further at VMAF 84.
Switch demo: NormAV1 sweep · Live demo: dav1d WASM 4K AV1 decode
The measurement source CityHall is already AOM-AV1-compressed. NormAV1 squeezes a further 8.43× out of that already-compressed AV1 — visually transparent at VMAF 84.36. This is possible because commutator norm operates on a perceptual axis rather than the rate-control axis other codecs use.
JAVATEL Inc. has supplied surveillance camera systems, video management, and media streaming in Japan since 1997. Based in Nishinomiya, Hyōgo. Video intelligence is not a recent pivot — it is the direction we have walked for 29+ years, now delivered through the patent-pending NormMAP commutator-norm codec engine.
See the live switch demo → 30-day trial →
Patent: JP 2026-046898 / 046609 / 046625 (出願済み) · Paper: Commutator Norm as a Continuous Criterion for GOP Safety and Perceptual Quality Redistribution in Video Encoding — IEEE TCSVT v8 投稿.
DMS — Docokame Media Streaming
JAVATEL OEM of Unreal Media Streaming · client-server WebRTC + 48‑hour time-shift archive
Production-grade live streaming + recording platform. JAVATEL distributes and integrates Unreal Media Streaming under the DMS brand, with a modern REST API layer (DmsRestApi) and React management console (DMS Console) on top.
Client × server architecture
DMS Server ingests live RTSP / file / capture sources. DMS clients consume via WebRTC, MSE, HLS, or RTSP. One server handles dozens of simultaneous viewers and recorders without per-stream re-encode.
WebRTC sub-second latency
End-to-end browser playback under 1 second over WebRTC with no plugin. Live monitoring, drone feeds, broadcast IFB — all through Chrome / Firefox / Edge / Safari without any installs. Native MSE fallback for compatibility.
ShiftCast48 — 48h time-shift
Continuous 48-hour rolling archive on every channel. Absolute time seek: jump to "2026-05-06 14:32:11" directly — no scrubbing required. Live and replay coexist seamlessly through the same WebRTC pipe.
Introduction
SlimeTree is JAVATEL's umbrella brand for the Slime family — SlimeCodec (video codec), SlimeNENC (cross-language transpiler + PSDP same-language parallelization), SlimeDSP (acoustic anomaly detection). Pat. P 2025-183827.
SlimeCodec series — video codec research
JP Patent App. 2026-046609
Research framed by a single principle: smaller files, without smaller pictures. NormH.264 compresses 4K H.264 source to roughly one-quarter the size, visually indistinguishable from the original. NormAV1 (research preview) takes already-AV1 4K source down to 2.13 MB at 1.79 Mbps and VMAF 84.36 — visually transparent at LTE 4G bitrate — and to 1.15 MB / 0.97 Mbps at the perceptual limit. Switch and see for yourself.
Standard decoders work unchanged. Playback environments need not change.
4K AV1 browser-native player — plugin-free HW decode, automatic H.264 fallback for iPhone.
WASM 2-core dav1d decode — the TCSVT paper's “2C WASM 4K AV1 realtime” claim verified on three architecture classes: Apple A15 (iPhone 13, ARM 2C), Intel i5-1145G7 (Tiger Lake x86 mobile 2C), and AMD Ryzen 5 GHz (x86 desktop) all play 4K 30 fps smoothly. Single-machine number → cross-architecture universality.
SlimeNENC — legacy-language transpiler family
JP Patent App. 2026-046620 (Subset A)
An audit-grade transpiler family covering SlimeCOBOL / SlimeMUMPS / SlimePL/I / SlimeRPG / SlimeFORTRAN / SlimeJCL — 6 languages plus JCL on a shared S2-S5 / S7-S9 backbone. Targets primarily Java, with Rust, C#, Kotlin, Go, and Clojure also supported.
- SlimeCOBOL: 14 dialects, verified on NIST CCVS85 (501 programs), 5 emitters
- SlimeMUMPS: 9 dialects (GT.M / Caché / IRIS / YottaDB …), healthcare EHR (VistA / Epic) migration
- SlimePL/I: 5 dialects (IBM Enterprise / Open PL/I …), insurance / banking / aviation; S9 60/60 + byte-exact 60/60
- SlimeRPG: RPG IV free-format, IBM iSeries / AS/400, banking / manufacturing / distribution; S9 35/35 + byte-exact 35/35 + full file I/O
- SlimeFORTRAN: F77 fixed-form (col 1-72) → modern FORTRAN (F2018 free-form), HPC / numerical, NASA / ECMWF / Japan Meteorological Agency / JAEA; S9 25/25 + byte-exact 25/25 + gfortran compile 25/25
- SlimeJCL (2026-Q3): IBM z/OS / Fujitsu / Hitachi / NEC, to Spring Batch
- LLM-at-build only. Runtime is LLM-free and offline-capable.
- Full audit trail — every transformation chained by FNV1a64 hash. Six-line provenance per token. Deterministic, replayable.
- Tamper-proof — any post-hoc edit breaks the hash chain and is detected instantly. Audit log cannot lie.
- Stage-rollback — pipeline replays from any checkpoint, byte-for-byte. Failed transforms rewind without ambiguity.
No mainframe-modernization product on the market combines all three with bit-exact determinism. We believe this is industry-first.
+ 99.9995% determinism — 3 verification mechanisms 🔽
On NIST CCVS85's 501 programs (353,666 lines), measured pass rate was 100% (zero detected errors). The figure 99.9995% is the theoretical upper bound on residual undetected error — obtained by combining the independent miss-rates of three independent verification mechanisms:
- Algebraic acceptor (S5) — every transformation must satisfy a 64-dimensional invariant before it is emitted. Outputs that don't satisfy ker(dφ) are rejected upfront.
- Chained audit log (S7) — every accepted transformation is hashed into a 6-line provenance chain. Tampering breaks the chain; replay is byte-exact.
- Statistical guard (S8 Mini-PSV) — a bigram language model over generated Java catches silent-pass-but-meaning-wrong outputs. Trigram hallucination detection 95%, CORE silent-pass rate 98.9%.
Combined miss-rate ≈ (1−0.95) × (1−0.989) ≈ 5 in 1 million — the "99.9995% bound." Independent of LLM runtime availability; survives offline / air-gapped deployment.
This subset handles cross-language transpilation only. Same-language parallelization is offered separately as PSDP, below.
Trial: 5,000 LOC for 1 month, free. Optional add-on up to 10,000 LOC.
SlimeNENC — same-language parallelization (PSDP)
JP Patent App. 2026-046620 (Subset B)
Phase-synchronized deterministic parallelism. An existing program is re-emitted in the same language, with bit-exact 12-core parallelization.
- 23 converters / 12+ languages and
variants implemented:
C / C++ / C# / Go / Rust (+prime) / Swift / Python / Node / PHP / Lisp / Clojure / Erlang / Java 17 & 21 (incl. migration paths 8→17, 17→21, 8→21, plus prime) / Kotlin (+coroutines) / Scala / FORTRAN - 2.97–4.23× speedup (PSDP paper v5d, 5-language core measurement)
- 378 runs all bit-exact
- 90/90 fault-injection rollbacks, zero barrier violations
Applicable to any existing same-language program. Java migrated from COBOL via Subset A is a natural target. Same patent application — no contractual renegotiation required.
Trial: 5,000 LOC for 1 month (no paid add-on).