Healthie.NET.AI
4.1.2
Prefix Reserved
See the version list below for details.
dotnet add package Healthie.NET.AI --version 4.1.2
NuGet\Install-Package Healthie.NET.AI -Version 4.1.2
<PackageReference Include="Healthie.NET.AI" Version="4.1.2" />
<PackageVersion Include="Healthie.NET.AI" Version="4.1.2" />
<PackageReference Include="Healthie.NET.AI" />
paket add Healthie.NET.AI --version 4.1.2
#r "nuget: Healthie.NET.AI, 4.1.2"
#:package Healthie.NET.AI@4.1.2
#addin nuget:?package=Healthie.NET.AI&version=4.1.2
#tool nuget:?package=Healthie.NET.AI&version=4.1.2

Healthie.NET.AI
▶ Live demo — board.healthie-dotnet.dev — a read-only Healthie.NET dashboard watching real status pages (Anthropic, OpenAI, GitHub, Cloudflare, and more), built from these packages. Full documentation at healthie-dotnet.dev.
Explains what a pulse checker's recent history shows: whether it is failing consistently or intermittently, when it started, and what the errors point to. Provider-agnostic -- it depends on Microsoft.Extensions.AI.Abstractions and nothing else, so the model is your choice.
Installation
dotnet add package Healthie.NET.AI
Usage
Register any IChatClient, then add the diagnostician:
using Healthie.AI;
// Anthropic, OpenAI, Azure OpenAI, or a local model via Ollama -- this package depends on none of them.
builder.Services.AddSingleton<IChatClient>(
new AnthropicClient().AsIChatClient("claude-opus-4-8"));
builder.Services.AddHealthieAI();
Then ask it about a checker:
var diagnosis = await diagnostician.DiagnoseAsync("MyApp.Pulses.DatabasePulseChecker");
Console.WriteLine(diagnosis.Summary);
Console.WriteLine($"Failure rate rose: {diagnosis.Anomaly.IsAnomalous}");
Key Types
| Type | Description |
|---|---|
IPulseDiagnostician |
DiagnoseAsync(name) returns a PulseDiagnosis for one checker. |
PulseDiagnosis |
The checker's Name, a plain-English Summary, and an Anomaly report. |
AnomalyReport |
IsAnomalous, RecentFailureRate, EarlierFailureRate. |
StartupExtensions.AddHealthieAI() |
Registers IPulseDiagnostician. Requires an IChatClient in DI. |
The Anomaly Report Is Not a Model Call
Anomaly compares the recent failure rate against the earlier one arithmetically, from the checker's own history. It is deterministic and testable, and it is worth trusting on its own -- including when no model is configured to write the summary.
What Leaves Your Process
Diagnosing a checker sends its name, its health, and the messages its checks reported to whichever model you configured. Those messages are written by your own checkers, which is a good reason to keep credentials out of check results. This package never mutates state.
You May Not Need This
If your only consumer is an AI agent, use Healthie.NET.Mcp instead: an agent talking to the MCP endpoint can read get_check_history and reason about it itself. This package is for the surfaces where there is no model in the loop -- a REST response, a dashboard, an alert.
See Also
| Product | Versions Compatible and additional computed target framework versions. |
|---|---|
| .NET | net8.0 is compatible. net8.0-android was computed. net8.0-browser was computed. net8.0-ios was computed. net8.0-maccatalyst was computed. net8.0-macos was computed. net8.0-tvos was computed. net8.0-windows was computed. net9.0 was computed. net9.0-android was computed. net9.0-browser was computed. net9.0-ios was computed. net9.0-maccatalyst was computed. net9.0-macos was computed. net9.0-tvos was computed. net9.0-windows was computed. net10.0 is compatible. net10.0-android was computed. net10.0-browser was computed. net10.0-ios was computed. net10.0-maccatalyst was computed. net10.0-macos was computed. net10.0-tvos was computed. net10.0-windows was computed. |
-
net10.0
- Healthie.NET.Abstractions (>= 4.1.2)
- Microsoft.Extensions.AI.Abstractions (>= 10.9.0)
-
net8.0
- Healthie.NET.Abstractions (>= 4.1.2)
- Microsoft.Extensions.AI.Abstractions (>= 10.9.0)
NuGet packages
This package is not used by any NuGet packages.
GitHub repositories
This package is not used by any popular GitHub repositories.
### Fixed
- The dashboard event-log and about dialogs now move focus inside, keep keyboard focus contained,
close with Escape, and restore focus to the control that opened them.
- Dashboard checker rows now expose selection as a real button alongside their action buttons,
removing nested interactive controls while preserving whole-row pointer selection.
- A dashboard state-provider failure now shows a recoverable unavailable state with a retry action
instead of looking like an empty installation with no registered checkers.
- The Blazor sample now keeps the dashboard stylesheet in its static host head, preventing an
unstyled flash while the interactive head reconnects on a slow network.
- HTTP checker display names and result messages now omit URI user information, query strings, and
fragments, server-controlled reason phrases, and transport exception details, preventing embedded
credentials and tokens from reaching stored history, alerts, logs, or the dashboard while requests
still use the complete configured URI.
- Alert-history persistence now loads durable history before its first post-restart write and
coalesces changes raised during a slow state-provider write into one latest-state follow-up. This
preserves earlier alerts without duplicating the new one and avoids an unbounded queue of tasks
and redundant full-history writes during alert bursts.
- The Web API sample maps its unauthenticated management API and mutation-enabled MCP endpoint only
in Development, so deploying the sample with a production environment no longer exposes those
local demonstration surfaces. Liveness and readiness probes remain available.
- History clearing and startup trimming now use optimistic concurrency, so they no longer overwrite
a check result or setting changed by another replica between the read and write.
- Leader election now serializes schedule changes with leadership transitions and reconciles every
registered checker's persisted active state and schedule through a bulk read on lease renewal. A
pause or cadence change made through another replica now reaches the leader, a fresh leader
removes inactive durable jobs left by the previous process, and reconciliation runs with bounded
concurrency while an independent lease heartbeat prevents slow scheduler calls from expiring an
otherwise healthy leader. Partial and cancelled transitions remain tracked until cleanup succeeds,
and shutdown does not release the lease over work it could not stop.
- CosmosDB multi-state reads now use `ReadManyItemsAsync`, avoiding one sequential network round
trip per checker during dashboard loads and leader renewal.
- The in-memory lease provider now returns exactly one winner when replicas contend concurrently.
- Coravel, Hangfire, Temporal, and the leader-election decorator now validate schedules before they
are persisted. Cronos expressions requiring an absent jitter seed are reported as validation
failures instead of escaping as server errors. The built-in timer also rejects periods outside
`PeriodicTimer`'s supported range before replacing a working schedule.
- Temporal now preserves Healthie.NET's six-field, leading-seconds cron meaning by translating it
to Temporal's seven-field seconds form. It rejects Cronos-only relative-day and descending-range
expressions whose Temporal meaning differs, and fixed periods outside Temporal's supported range,
before an existing schedule is removed.
- Uptime recording now starts a new observation segment on the first fresh result after a process
restart, even when its health matches persisted state. Time while no process was observing the
checker remains unknown, and a transition that finds the bounded queue full is retried by the next
fresh result.
Full changelog: https://github.com/ivanvyd/Healthie.NET/blob/v4.1.2/CHANGELOG.md