Testing

IntoChat.OS.Testing is the test harness for IntoChat OS modules. It provides a TestBrain builder that runs the product recipe in process, a shared host per test assembly, signal probes, bounded waits, and fakes for the model endpoint, an OAuth token endpoint and telemetry.

Test tiers

Each module splits its tests by project suffix:

Suffix	Contents
*.Tests.Unit	No host, no network, no sleeps.
*.Tests.Integration	One BrainHostFixture per assembly.
*.Tests.Hosting	Aspire application model only (see IntoChat.OS.Aspire.Testing).
*.Tests.E2E, *.Tests.Browser	The full product on Aspire, run nightly (see IntoChat.OS.Testing.Aspire, IntoChat.OS.Testing.Playwright).

Quick start

await using var brain = await TestBrain.Create()
    .WithModule<MyModule>()
    .StartAsync(ct);

StartAsync boots a silo with the same recipe the product uses. WithoutHttp() skips the HTTP surface when a fact only talks to grains.

One host per assembly

Starting a host is the expensive part. Start one per test assembly and isolate each fact in a fresh person's brain:

[assembly: AssemblyFixture(typeof(MyHost))]

public sealed class MyHost : BrainHostFixture
{
    protected override TestBrain Configure(TestBrain brain) => brain.WithModule<MyModule>().WithoutHttp();
}

PersonAsync(name) creates a fresh principal, account and brain, and ObserveAsync<T>(source, owner, ct) subscribes a SignalProbe<T> as the owner. The package targets xunit v3.

Fakes

• FakeModelServer: an OpenAI-compatible endpoint with scripted replies.
• TokenEndpointStub: an OAuth token and profile endpoint for integration modules.
• InMemoryTelemetry: collects spans and log records through brain.WithInMemoryTelemetry.