Skip to main content
MTE Relayv5

Faster. Leaner.Stronger.

Measured against v4 on the same hardware, with stronger key exchange, verified data, and one dashboard to manage it all.

90% less memory.Per Relay with 5 to 10 KB requests: about 150 MB in v5 against about 2 GB in v4.
37% more requests per second.On the same 2 CPU Relay with 10 KB requests.
39% lower latency.At light load, two Relays with Redis.
29% faster p99.Under full load, 256 connections.
16% higher throughput with high availability.Two 2 CPU Relays with Redis.
1 Redis instead of 2.A smaller footprint: one less instance to run, patch and pay for.

Measured October 1 to 2, 2026: v5 5.0.0 against the June 2026 v4 certification runs on the same workstation, median of four runs. How we measured

Want the technical details? Read the engineering guide

Raw performance.
Backed by results.​

We reran the load tests that certified v4 in June 2026 against v5 5.0.0: the same payloads, the same connection ladders, the same 60 second holds, on the same workstation, four times over. v5 moves more requests through the same cores in every configuration we compared, answers faster, and needs a fraction of the memory.

Requests per second on the same cores.More is better
v5v4
One 4 CPU Relay10 KB requests, 16 to 256 connections
9,152 req/s
7,729 req/s
One 2 CPU Relay10 KB requests, 16 to 256 connections
5,198 req/s
3,807 req/s
Two 2 CPU Relays5 KB requests, 256 connections
9,038 req/s
7,778 req/s
Two 2 CPU Relays10 KB requests, 256 connections
8,026 req/s
7,142 req/s
Requests per second at matching connection counts, 60 second holds, median of four v5 runs. The single Relay rows average the 16 to 256 connection levels, the way the v4 headline of 7,700 was formed.
Faster round trips at light load.Less is better
v5v4
Two 4 CPU Relays10 KB requests, one connection
1.13 ms
1.86 ms
Two 2 CPU Relays5 KB requests, one connection
1.06 ms
1.55 ms
One 4 CPU Relay10 KB requests, one connection
0.94 ms
1.03 ms
Mean round trip through the Relay pair with one connection, measured right after warm up.
A fraction of the memory.Less is better
v5v4
One 2 CPU Relay10 KB requests
152 MiB
1,949 MiB
Two Relays with Redis5 and 10 KB requests
124 MiB
1,165 MiB
Largest container working set per Relay sampled at 64 and 128 connections. With 2 KB requests v4 needed about 200 MB, so the gap grows with request size.

Measured October 1 to 2, 2026: v5 5.0.0 against the June 2026 v4 certification runs on the same workstation, median of four runs. The engineering guide describes the v5 run and lists every v5 figure.

Data verified.
In every frame.​

v4 streaming checks integrity when the stream ends. v5 validates each frame before decoding, then verifies its integrity before releasing any plaintext. The request method, response status, and headers are protected too. Every frame is checked. Only verified data reaches your application.

v5 framesEvery data frame cryptographically verified
Frame 1Frame 2Frame 3Frame 4Frame 5
v4 streamingCheck at the end
One continuous data streamFinal check
Only verified plaintext reaches the application.
Altered data is rejected before use.
Frame structure is validated before decoding. Integrity is verified during decoding, before plaintext is released. Frames stay in sequence; transport security is still required.

Stronger quantum resistant key exchange.​

Kyber 1024 strengthens the quantum resistant key exchange that establishes each protected session. Clients check compatibility before pairing, so both ends agree before your data moves.

2×
Stronger encryption security.NIST approved, quantum-resistant key exchange
CRYSTALS-KYBER 1024
  • ML-KEM
  • FIPS 203
  • FIPS 140-3 validated
2× compares the symmetric security targets in bits with v4. Kyber 1024 targets roughly AES 256 security. CRYSTALS-Kyber 1024 is standardized as ML-KEM-1024 in FIPS 203, and MTE uses FIPS 140-3 validated cryptography. Security definitions.

Your whole fleet. One control panel.​

See Relay health and configuration in one place. Change supported settings live, track who changed what, and roll back a bad configuration. Other settings apply on restart. The panel is optional; Relays keep serving their last good configuration if it becomes unreachable.

MTE Relay Control Plane dashboard showing three workers online, all three in sync, configuration version 19, and fleet and configuration controls.
The v5 control panel: fleet visibility, configuration, and version history. Select the image to view full size.

One frame.
Limitless transports.​

v5 introduces a new frame generation. One binary envelope and one parser, shared across HTTP, WebSocket, and TCP. The HTTP Relay is what this release delivers. SocketX v2 for WebSocket and the TCP Gateway are built on the same frame, each versioned and released on its own.

One protected frameOne parser, every transport
HTTPMTE Relay v5

Requests, responses, and SSE streams through the HTTP Relay.

This release
WebSocketSocketX v2

Live sessions for chat, dashboards, and other real time services.

Separate product
TCPTCP Gateway

Approved destinations such as SSH, mail, and databases.

Separate product
One frame layout and one parser across the three products. The v5 HTTP Relay advertises HTTP transport only. SocketX and the TCP Gateway are separate builds with their own release versions, and availability for each is announced separately.