Nerve 2.0.0

Nerve ⚡

In-process publish/subscribe for .NET 10, shaped like MQTT and about as expensive as a method call. Topics with + and # wildcards, retained messages, request/reply, async streams and per-route statistics — 21 ns and zero allocations per published message.

Built by Gravicode Studios, led by Kang Fadhil.

English · Bahasa Indonesia · Documentation · Benchmarks

<a name="english"></a>

🇬🇧 English

Install

dotnet add package Nerve

Thirty seconds

using Nerve;

var nerve = new NerveHub();

// Subscribe. The handler runs on the thread that publishes.
using var reader = nerve.Subscribe<double>("sensor/tank-3/temperature",
    celsius => Console.WriteLine($"{celsius:N1} C"));

// One level with +, everything below with #.
using var all = nerve.Subscribe<double>("sensor/+/temperature", c => Log(c));

await nerve.PublishAsync("sensor/tank-3/temperature", 28.4);

// Ask a question and wait for the answer, over the same topics.
using var responder = nerve.Respond<string, int>("text/length", text => text.Length);
int length = await nerve.RequestAsync<string, int>("text/length", "gravicode");   // 9

What it does

Measured on this machine

.NET 10.0.11, Windows 11, 8 logical cores. Full output in docs/benchmark-run.txt.

Fan-out costs what fan-out costs — 8 subscribers is 83.9 ns, 32 subscribers is 300 ns — and a wildcard subscriber measures the same as an exact one, because the matching already happened.

Run it

dotnet run --project src/Nerve.Demo                        # every feature, end to end
dotnet run --project src/Nerve.AgentSim -- --demo 8        # the simulator below
dotnet test tests/Nerve.Tests                              # 79 tests, about a second
dotnet run --project src/Nerve.Benchmarks -c Release -- --quick

The agent coordination simulator

An orchestrator is handed a list of instructions. It plans each one from the words in it, dispatches the pieces to six specialists, and folds their answers back into a digest.

No agent holds a reference to another. Every arrow you see is a topic: the orchestrator publishes to agents/task/{specialty} and subscribes to agents/result/+, the specialists take work through StreamAsync so they genuinely run in parallel, and the panel itself is just a seventh subscriber watching the same traffic.

Every spark on the arbor is a message that was really published. Violet leaving the soma is a sub-task going out; a specialist's own colour returning is their answer coming back.

dotnet run --project src/Nerve.AgentSim -- --demo 8

Documentation

Bahasa Indonesia: docs/id/.

<a name="bahasa-indonesia"></a>

🇮🇩 Bahasa Indonesia

Instalasi

dotnet add package Nerve

Tiga puluh detik

using Nerve;

var nerve = new NerveHub();

// Berlangganan. Handler berjalan di thread yang mem-publish.
using var pembaca = nerve.Subscribe<double>("sensor/tank-3/temperature",
    celsius => Console.WriteLine($"{celsius:N1} C"));

// Satu level dengan +, seluruh sisanya dengan #.
using var semua = nerve.Subscribe<double>("sensor/+/temperature", c => Catat(c));

await nerve.PublishAsync("sensor/tank-3/temperature", 28.4);

// Bertanya dan menunggu jawabannya, lewat topik yang sama.
using var penjawab = nerve.Respond<string, int>("text/length", teks => teks.Length);
int panjang = await nerve.RequestAsync<string, int>("text/length", "gravicode");   // 9

Apa saja yang tersedia

Hasil pengukuran di mesin ini

.NET 10.0.11, Windows 11, 8 core logis. Output lengkapnya di docs/benchmark-run.txt.

Fan-out tetap ada biayanya — 8 subscriber 83,9 ns, 32 subscriber 300 ns — dan subscriber wildcard terukur sama dengan subscriber eksak, karena pencocokannya sudah selesai sebelumnya.

Menjalankannya

dotnet run --project src/Nerve.Demo                        # seluruh fitur, dari ujung ke ujung
dotnet run --project src/Nerve.AgentSim -- --demo 8        # simulator di bawah ini
dotnet test tests/Nerve.Tests                              # 79 tes, sekitar satu detik
dotnet run --project src/Nerve.Benchmarks -c Release -- --quick

Simulator koordinasi multi-agent

Sebuah orchestrator menerima daftar instruksi. Ia menyusun rencana dari kata-kata di dalamnya, mengirim potongan pekerjaannya ke enam agen spesialis, lalu melipat kembali jawaban mereka menjadi satu ringkasan.

Tidak ada satu agen pun yang memegang referensi ke agen lain. Setiap panah yang terlihat adalah sebuah topik: orchestrator mem-publish ke agents/task/{specialty} dan berlangganan agents/result/+, para spesialis mengambil pekerjaan lewat StreamAsync sehingga benar-benar berjalan paralel, dan panelnya sendiri hanyalah subscriber ketujuh yang menonton lalu lintas yang sama.

Setiap kilatan pada arbor adalah pesan yang sungguh-sungguh di-publish. Warna ungu yang meninggalkan soma adalah sub-tugas yang dikirim keluar; warna khas seorang spesialis yang kembali adalah jawabannya.

dotnet run --project src/Nerve.AgentSim -- --demo 8

Dokumentasi

Versi Bahasa Indonesia ada di docs/id/:

License

MIT. See LICENSE.

Built by Gravicode Studios, led by Kang Fadhil — studios.gravicode.com