Aspire
1 / 46

Inside Aspire

The Architecture Behind the AI Loop

How we help AI get Aspire right.

Start presentation
David Pine's actual public profile portrait
David PineMicrosoft's Aspire team

Wisconsin .NET UG
November 17, 2026 at 6pm Central

What is Aspire?

Aspire is an agent‑ready, code‑first tool to compose, debug, and deploy any distributed app.

aspire.dev

Aspire describes your whole app in code.

apphost.cs
var builder = DistributedApplication.CreateBuilder(args);

var cache = builder.AddRedis("cache");

var api = builder.AddDotnetProject("api", "../api")
    .WithReference(cache)
    .WithHttpHealthCheck("/health");

builder.AddViteApp("web", "../web")
    .WithReference(api);

builder.Build().Run();
  • DescribeYour services, databases, and how they connect. One file.
  • RunOne command starts all of it, with a dashboard for logs and traces.
  • DeployThe same model publishes to Docker Compose, Kubernetes, or Azure.
aspire.dev

Meet Aspire. Three commands.

Live demoFrom nothing to a running app
aspire --version --banner
One CLI, installed once.
aspire new
Pick a starter. You pick.
aspire run
Everything starts. Dashboard included.

Ask AI about Aspire.

  1. Let's add .NET Aspire to your solution.Aspire
  2. Then start it with dotnet run.aspire run
  3. Docs: learn.microsoft.comaspire.dev
  4. .WithHealthCheckEndpoint().WithHttpHealthCheck()

Three were out of date. One never existed.

Models learned the old Aspire.

What models read.NET Aspire8.0 · 9.0 · 9.1 · 9.2 · 9.3 · 9.4 · 9.5
renamed
What's true nowAspire13.0 · 13.1 · … · today

Anything newer than the training data is a guess.

So we built loops.

One of many loops. Let's zoom in.

Say you want browser logs in your app.

WithBrowserLogs

Console output, exceptions, network events, and screenshots become evidence.

Experimental browser logging can be configured at runtime, without restarting the AppHost.

That was one loop. Here are six.

The browser logs page started as a pull request.

GitHubGitHub
Excerpt of merged microsoft/aspire PR 16531, Add BrowserLogs configuration command: David Fowler merges commit 49e4797 into main on April 29, 2026, and aspire-repo-bot comments: Pull request created: #775, generated by PR Documentation Check.Excerpt of merged microsoft/aspire PR 16531, Add BrowserLogs configuration command: David Fowler merges commit 49e4797 into main on April 29, 2026, and aspire-repo-bot comments: Pull request created: #775, generated by PR Documentation Check.
The feature PR merges, then the docs bot links the PR it opened.
  1. microsoft/aspire#16531The feature merges.
  2. pr-docs-check run 250926952413 seconds later, the docs workflow starts.
  3. microsoft/aspire.dev#7759 minutes after the merge, a bot opens the docs pull request. It merges after 2 days.
  4. browser-logs.mdPublished. The CLI serves it.

The feature's author refined the bot's draft.

Excerpt of merged docs PR microsoft/aspire.dev 775, Add browser logs documentation: aspire-repo-bot opens it to document microsoft/aspire#16531 by @davidfowl, requests a review from David Fowler, and David Fowler adds the commit: Move browser logs to integrations/devtools and address review feedback.Excerpt of merged docs PR microsoft/aspire.dev 775, Add browser logs documentation: aspire-repo-bot opens it to document microsoft/aspire#16531 by @davidfowl, requests a review from David Fowler, and David Fowler adds the commit: Move browser logs to integrations/devtools and address review feedback.
microsoft/aspire.dev#775: drafted by a bot, refined by David Fowler, then merged.

Under two hours after the merge, David Fowler fixed the bot’s draft.

Who wrote that first draft?

Docs reach the CLI after review and release.

The CLI reads published docs. No model is trained on them.

The workflow drafts. People merge.

GitHubGitHub
pr-docs-check.mdgh aw compilepr-docs-check.lock.yml
pr-docs-check.md
on:
  pull_request:
    types: [closed]
    branches:
      - main
      - release/*
safe-outputs:
  create-pull-request:
    title-prefix: "[docs] "
    draft: true
    target-repo:
      "microsoft/aspire.dev"
    protected-files: blocked
pr-docs-check.lock.yml · 1,543 lines
jobs:
  agent:
    permissions:
      contents: read
      pull-requests: read
  detection:
    needs: [activation, agent]
  safe_outputs:
    needs: [activation, agent, detection]
    if: needs.detection.result == 'success'
    permissions:
      contents: write
      pull-requests: write

github.github.com/gh-aw

Docs ship as fast as Aspire does.

Ten releases in under ten weeks.

Guides

Written in MDX with agent skills, then reviewed by people.

  • doc-writer
  • doc-tester
  • doc-pr-reviewer
  • twoslash-validator

API reference

Generated from every shipped package, per version.

  • C#
  • TypeScript
  • Go

What’s new

Release notes now run through one skill: research, draft, critique, validate.

  • whatsnew

How does AI find it?

The AI asks the CLI to search the docs.

Recorded CLI run
aspire docs search "Configure tracked browser" --format Json
[
  {
    "title": "Browser logs",
    "slug": "browser-logs",
    "content": "Learn how to capture and configure browser console logs in Aspire using the BrowserLogs resource.",
    "section": "Add browser logs to a resource",
    "score": 172.43004
  },
  {
    "title": "Browser app configuration",
    "slug": "browser-app-configuration",
    "content": "Configure generic JavaScript browser apps to send OpenTelemetry data to the Aspire dashboard with OTLP HTTP exporters.",
    "section": "Example browser telemetry code",
    "score": 119.291626
  },
  {
    "title": "Enable browser telemetry",
    "slug": "enable-browser-telemetry",
    "content": "Enable browser telemetry in the Aspire dashboard to capture client-side OpenTelemetry logs, traces, and metrics from front-end JavaScript apps.",
    "section": "Dashboard configuration",
    "score": 101.06361
  },
  {
    "title": "Set up Grafana k6 in the Aspire AppHost",
    "slug": "set-up-grafana-k6-in-the-aspire-apphost",
    "content": "Configure the Community Toolkit k6 resource in Aspire, including scripts, browser testing, endpoints, health checks, and OpenTelemetry.",
    "section": "Use a fixed port or browser-enabled image",
    "score": 73.46021
  },
  {
    "title": "Compiler Error ASPIREBROWSERLOGS001",
    "slug": "compiler-error-aspirebrowserlogs001",
    "content": "Learn what causes the Aspire compiler error ASPIREBROWSERLOGS001 and how to fix it so your AppHost builds cleanly.",
    "section": "To correct this error",
    "score": 71.18428
  }
]

Then it reads the section it found.

Recorded CLI run
aspire docs get browser-logs --section "Configure browser logs"
Loading documentation...
Configure browser logs

The Configure tracked browser command opens an interactive settings dialog directly in the Aspire dashboard. Use it to adjust the browser session at runtime without restarting the AppHost.

The dialog presents the following options:

• Scope — Apply changes to the current resource only, or globally to all browser log resources in the AppHost.
• Browser — Choose a detected Chromium browser (Edge or Chrome) or specify a custom executable path.
• User data mode — Shared (cross-AppHost persistent data directory) or Isolated (per-AppHost persistent data directory).
• Profile — Select a discovered Chromium profile from the Aspire-managed browser hive (available in Shared mode only).
• Save to user secrets — Persist the selected settings to the AppHost’s user secrets so they’re restored on the next run. This option is only shown when the AppHost project has a user secrets ID configured. Run aspire secret set (https://aspire.dev/reference/cli/commands/aspire-secret-set/) once in the AppHost project to initialize a user secrets store; subsequent Save to user secrets clicks then write the browser configuration into it.

Note

The Configure tracked browser command is only available when the Aspire dashboard’s interaction service is active (that is, when the dashboard is connected to the AppHost). It is disabled in standalone dashboard mode.

After confirming the dialog, the new configuration takes effect immediately: the current browser session restarts with the updated settings.

Then it gets that member's full reference.

Recorded CLI run
aspire docs api get "csharp/aspire.hosting.browsers/browserlogsbuilderextensions/methods#withbrowserlogs"
Loading API documentation...
BrowserLogsBuilderExtensions Methods

• Package: Aspire.Hosting.Browsers (https://aspire.dev/reference/api/csharp/aspire.hosting.browsers.md)
• Type: BrowserLogsBuilderExtensions (https://aspire.dev/reference/api/csharp/aspire.hosting.browsers/browserlogsbuilderextensions.md)
• Kind: Methods
• Members: 1

Extension methods for adding tracked browser log resources to browser-based application resources.

WithBrowserLogs(IResourceBuilder, string?, string?, BrowserUserDataMode?)

> Experimental: ASPIREBROWSERLOGS001 - Learn more (https://aspire.dev/diagnostics/aspirebrowserlogs001/)

• Name: WithBrowserLogs(IResourceBuilder<T>, string?, string?, BrowserUserDataMode?)
• Modifiers: extension
• Returns: IResourceBuilder<T>
• Source: GitHub (https://github.com/microsoft/aspire/blob/c89352703f78ae46de90fbf751ed8b52b77b89bf/src/Aspire.Hosting.Browsers/BrowserLogsBuilderExtensions.cs#L122-L289)

Adds a child resource that can open the application's primary browser endpoint in a tracked browser session, surface browser diagnostics, and capture screenshots.

public static class BrowserLogsBuilderExtensions
{
    public static IResourceBuilder<T> WithBrowserLogs<T>(
        this IResourceBuilder<T> builder,
        string? browser = null,
        string? profile = null,
        BrowserUserDataMode? userDataMode = null)
    {
        // ...
    }
}

Parameters

• builder (IResourceBuilder<T>)
  The resource builder.
• browser (string?) optional
  The browser to launch. When not specified, the tracked browser uses the configured value from Aspire:Hosting:BrowserLogs and otherwise prefers an installed "msedge" browser in shared user data mode, an installed "chrome" browser in isolated user data mode, and finally falls back to "chrome". Supported values include logical browser names such as "msedge" and "chrome", or an explicit browser executable path.
• profile (string?) optional
  Optional Chromium profile name or directory name to use. Only valid when the effective user data mode is BrowserUserDataMode.Shared (https://aspire.dev/reference/api/csharp/aspire.hosting.browsers/browseruserdatamode/fields.md). When not specified, the tracked browser uses the configured value from Aspire:Hosting:BrowserLogs if present.
• userDataMode (BrowserUserDataMode? (https://aspire.dev/reference/api/csharp/aspire.hosting.browsers/browseruserdatamode.md)) optional
  Optional BrowserUserDataMode (https://aspire.dev/reference/api/csharp/aspire.hosting.browsers/browseruserdatamode.md) that selects whether the tracked browser launches against a persistent Aspire-managed user data directory shared across all AppHosts on the machine ( BrowserUserDataMode.Shared (https://aspire.dev/reference/api/csharp/aspire.hosting.browsers/browseruserdatamode/fields.md), the default) or a per-AppHost persistent user data directory ( BrowserUserDataMode.Isolated (https://aspire.dev/reference/api/csharp/aspire.hosting.browsers/browseruserdatamode/fields.md)). Both modes use Aspire-managed paths under %LocalAppData%\Aspire\BrowserData on Windows (or platform equivalents); the user's normal browser profile is never used. When not specified, the tracked browser uses the configured value from Aspire:Hosting:BrowserLogs and otherwise defaults to BrowserUserDataMode.Shared (https://aspire.dev/reference/api/csharp/aspire.hosting.browsers/browseruserdatamode/fields.md).

Returns

IResourceBuilder<T> -- A reference to the original ApplicationModel.IResourceBuilder1` for further chaining.

Remarks

This method adds a child browser logs resource beneath the parent resource represented by builder. The child resource exposes a dashboard command that launches a Chromium-based browser in a tracked mode, attaches to the browser's debugging protocol, forwards browser console, error, exception, and network output to the child resource's console log stream, and can capture screenshots as command artifacts.

The tracked browser session uses the Chrome DevTools Protocol (CDP) (https://chromedevtools.github.io/devtools-protocol/) to subscribe to browser runtime, log, page, and network events.

The parent resource must expose at least one HTTP or HTTPS endpoint. HTTPS endpoints are preferred over HTTP endpoints when selecting the browser target URL.

Browser, profile, and user data mode settings can also be supplied from configuration using Aspire:Hosting:BrowserLogs:Browser, Aspire:Hosting:BrowserLogs:Profile, and Aspire:Hosting:BrowserLogs:UserDataMode, or scoped to a specific resource with Aspire:Hosting:BrowserLogs:{ResourceName}:Browser, Aspire:Hosting:BrowserLogs:{ResourceName}:Profile, and Aspire:Hosting:BrowserLogs:{ResourceName}:UserDataMode. Explicit method arguments override configuration.

Examples

Add tracked browser logs for a web front end:

var builder = DistributedApplication.CreateBuilder(args);

builder.AddProject<Projects.WebFrontend>("web")
    .WithExternalHttpEndpoints()
    .WithBrowserLogs();

ATS metadata

ATS export

• Available to Polyglot AppHosts through the Aspire Type System.

The member ID is the page's address.

The CLI's response
{
  "id": "csharp/aspire.hosting.browsers/browserlogsbuilderextensions/methods#withbrowserlogs",
  "name": "WithBrowserLogs(IResourceBuilder<T>, string?, string?, BrowserUserDataMode?)",
  "language": "csharp",
  "kind": "member"
}

It spells out the language, package, type, and member kind, then the member.

Skills add determinism to AI.

Like registered MCP tools, a skill maps intent to an exact Aspire CLI command.

You askSkillAspire CLI runs
“Is the API ready?”aspire-orchestrationaspire wait api
“What's running, and where?”aspire-monitoringaspire describe
“Why was that request slow?”aspire-monitoringaspire otel traces api

The model still reasons. The skill decides what it runs.

Skills ship as a plugin.

Set up and migrate

aspireaspire-initaspireifyaspire-project-v2-migration

Run, observe, and deploy

aspire-orchestrationaspire-monitoringaspire-deploymentaspire-terminals

Aspire CLIaspire agent init

Plugin marketplace/plugin install aspire@aspire-skills

github.com/microsoft/aspire-skills

Trace one search from command to result.

Tracingaspire docs search "browser logs" --format Json

DocsSearchCommand.cs Receive the command
var query = parseResult.GetValue(s_queryArgument)!;
var format = parseResult.GetValue(s_formatOption);
var limit = Math.Clamp(parseResult.GetValue(s_limitOption) ?? 5, 1, 10);

var response = await InteractionService.ShowStatusAsync(
    DocsCommandStrings.LoadingDocumentation,
    async () => await _docsSearchService.SearchAsync(query, limit, cancellationToken));
DocsSearchService.cs Delegate to the index
var searchResults = await _docsIndexService.SearchAsync(query, topK, cancellationToken)
    .ConfigureAwait(false);
DocsIndexService.cs Prepare before searching
await EnsureIndexedAsync(cancellationToken).ConfigureAwait(false);

if (_indexedDocuments is null or { Count: 0 }
    || string.IsNullOrWhiteSpace(query))
{
    return [];
}

var queryTokens = Tokenize(query);
DocsIndexService.cs One index per process
if (_indexedDocuments is not null)
{
    return;
}

await _indexLock.WaitAsync(cancellationToken).ConfigureAwait(false);

// Inside try; finally releases the lock.
if (_indexedDocuments is not null)
{
    return;
}
DocsIndexService.cs Read local state, then refresh
var cachedDocuments = await _docsCache.GetIndexAsync(cancellationToken)
    .ConfigureAwait(false);

var cachedFingerprint = await _docsCache.GetIndexSourceFingerprintAsync(cancellationToken)
    .ConfigureAwait(false);
DocsCache.cs Persistent parsed index
// GetIndexAsync:
var documents = await _contentCache.GetJsonAsync(
    _indexCacheKey, JsonSourceGenerationContext.Default.LlmsDocumentArray,
    cancellationToken: cancellationToken).ConfigureAwait(false);

// GetIndexSourceFingerprintAsync:
var fingerprint = await _contentCache.GetAsync(
    _indexSourceFingerprintCacheKey, cancellationToken).ConfigureAwait(false);
DocsFetcher.cs Hand off to HTTP
public async Task<string?> FetchDocsAsync(CancellationToken cancellationToken = default)
{
    await MigrateLegacyCacheAsync(cancellationToken).ConfigureAwait(false);

    return await CachedHttpDocumentFetcher.FetchAsync(
        _httpClient, _cache, _llmsTxtUrl, _cacheKey, _logger, cancellationToken)
        .ConfigureAwait(false);
}

Turn on Debug. Watch the cache work.

Docs changed 200 OK
[dbug] DocsFetcher: Fetching content from https://aspire.dev/llms-full.txt, cached ETag: "Dbk3…"
[dbug] DocsFetcher: Stored new ETag for …/llms-full.txt: "KLdZ…"
[info] DocsFetcher: Fetched content from …/llms-full.txt, length: 7658674 chars
[info] DocsIndexService: Indexed 602 documents from aspire.dev in 01.901 seconds.
Nothing changed 304 Not Modified
[dbug] DocsCache: ETag cache disk hit for key: llms-full
[dbug] DocsFetcher: Fetching content from https://aspire.dev/llms-full.txt, cached ETag: "KLdZ…"
[dbug] DocsFetcher: Server returned 304 Not Modified for https://aspire.dev/llms-full.txt
[info] DocsIndexService: Loaded 602 documents from cache in 01.219 seconds.

Same search with --log-level Debug. The 304 skips the 7.7 MB download and the parse.

Where a word matched. How rare it is.

DocsIndexService.cs Narrative-doc weights
private const float TitleWeight = 10.0f;
private const float SummaryWeight = 8.0f;
private const float HeadingWeight = 6.0f;
private const float CodeWeight = 5.0f;
private const float BodyWeight = 1.0f;
DocsIndexService.cs Rarity (IDF)
var documentCount = documents.Count;

foreach (var (token, count) in documentFrequency)
{
    // IDF (inverse document frequency): the rarer the token, the higher its weight.
    idf[token] = MathF.Log(1.0f +
        (documentCount - count + 0.5f) / (count + 0.5f));
}
DocsIndexService.cs Keep the best matches
var (score, matchedSection) = ScoreDocument(
    doc, queryTokens, queryTokenWeights, queryAsSlug, queryAsPhrase);

if (score > 0)
{
    AddTopResult(results, new SearchCandidate(
        doc, matchedSection, score, documentIndex), topK);
}
DocsSearchService.cs Return the response shape
var results = searchResults.Select(r => new SearchResult
{
    Title = r.Title,
    Slug = r.Slug,
    Content = r.Summary ?? r.Title,
    Section = r.MatchedSection,
    Score = r.Score
}).ToList();

return new DocsSearchResponse { Query = query, Results = results };
DocsSearchCommand.cs Back to stdout
if (format is OutputFormat.Json)
{
    var json = JsonSerializer.Serialize(
        response.Results.ToArray(),
        JsonSourceGenerationContext.RelaxedEscaping.SearchResultArray);

    InteractionService.DisplayRawText(json, ConsoleOutput.Standard);
}

A title hit outweighs body text. A rare term like addredis outweighs a common one like apphost. No embeddings, no model training.

Now the AI has to act.

The CLI waits until the API is ready.

Recorded CLI run
aspire wait api --apphost 'demo\host\apphost.cs' --status healthy --timeout 120 --non-interactive
Waiting for resource 'api' to be healthy...

✅ Resource 'api' is healthy. (0.0s)

The CLI reports what is running and how healthy it is.

Recorded CLI run
aspire describe --apphost 'demo\host\apphost.cs' --format Json --non-interactive
{
  "resources": [
    {
      "name": "api-kewwzmak",
      "displayName": "api",
      "resourceType": "Executable",
      "state": "Running",
      "creationTimestamp": "2026-10-12T11:24:41+00:00",
      "startTimestamp": "2026-10-12T11:24:41.8+00:00",
      "source": "Weather.Api.exe",
      "healthStatus": "Healthy",
      "dashboardUrl": "[redacted dashboardUrl]",
      "relationships": [
        {
          "type": "Reference",
          "resourceName": "open-meteo"
        },
        {
          "type": "Reference",
          "resourceName": "reverse-geocoding"
        },
        {
          "type": "Reference",
          "resourceName": "geocoding"
        },
        {
          "type": "Reference",
          "resourceName": "open-meteo"
        },
        {
          "type": "Reference",
          "resourceName": "reverse-geocoding"
        },
        {
          "type": "Reference",
          "resourceName": "geocoding"
        }
      ],
      "urls": [
        {
          "name": "http",
          "url": "http://localhost:62869"
        }
      ],
      "volumes": [],
      "properties": {
        "executable.args": null,
        "executable.path": "[redacted local filesystem path]",
        "executable.pid": 40208,
        "executable.workDir": "[redacted local filesystem path]",
        "resource.appArgs": [],
        "resource.appArgsSensitivity": []
      },
      "environment": "[redacted environment]",
      "healthReports": {
        "api_http_/health_200_check": {
          "status": "Healthy"
        }
      },
      "commands": {
        "restart": {
          "displayName": "Restart",
          "description": "Restart resource",
          "state": "Enabled",
          "sortOrder": 2
        },
        "stop": {
          "displayName": "Stop",
          "description": "Stop resource",
          "state": "Enabled",
          "sortOrder": 1
        }
      }
    },
    {
      "name": "frontend-installer-trmtsfbm",
      "displayName": "frontend-installer",
      "resourceType": "Executable",
      "state": "NotStarted",
      "source": "npm",
      "dashboardUrl": "[redacted dashboardUrl]",
      "relationships": [
        {
          "type": "Parent",
          "resourceName": "frontend-quwryuxr"
        }
      ],
      "urls": [],
      "volumes": [],
      "properties": {
        "executable.args": null,
        "executable.path": "npm",
        "executable.pid": null,
        "executable.workDir": "[redacted local filesystem path]",
        "resource.appArgs": null,
        "resource.appArgsSensitivity": null,
        "resource.parentName": "frontend-quwryuxr"
      },
      "environment": "[redacted environment]",
      "healthReports": {},
      "commands": {
        "start": {
          "displayName": "Start",
          "description": "Start resource",
          "state": "Enabled",
          "sortOrder": 0
        }
      }
    },
    {
      "name": "frontend-quwryuxr",
      "displayName": "frontend",
      "resourceType": "Executable",
      "state": "Running",
      "creationTimestamp": "2026-10-12T11:24:42+00:00",
      "startTimestamp": "2026-10-12T11:24:42.205+00:00",
      "source": "npm",
      "healthStatus": "Unhealthy",
      "dashboardUrl": "[redacted dashboardUrl]",
      "relationships": [
        {
          "type": "WaitFor",
          "resourceName": "frontend-installer-trmtsfbm"
        },
        {
          "type": "Reference",
          "resourceName": "api-kewwzmak"
        },
        {
          "type": "WaitFor",
          "resourceName": "api-kewwzmak"
        },
        {
          "type": "Reference",
          "resourceName": "api-kewwzmak"
        }
      ],
      "urls": [
        {
          "name": "http",
          "url": "http://localhost:62871"
        }
      ],
      "volumes": [],
      "properties": {
        "executable.args": null,
        "executable.path": "npm",
        "executable.pid": 14900,
        "executable.workDir": "[redacted local filesystem path]",
        "resource.appArgs": [
          "run",
          "dev",
          "--",
          "--port",
          "11338"
        ],
        "resource.appArgsSensitivity": [
          0,
          0,
          0,
          0,
          0
        ],
        "resource.launchConfigurationType": "node"
      },
      "environment": "[redacted environment]",
      "healthReports": {
        "frontend_http_/_200_check": {}
      },
      "commands": {
        "restart": {
          "displayName": "Restart",
          "description": "Restart resource",
          "state": "Enabled",
          "sortOrder": 2
        },
        "stop": {
          "displayName": "Stop",
          "description": "Stop resource",
          "state": "Enabled",
          "sortOrder": 1
        }
      }
    },
    {
      "name": "geocoding",
      "displayName": "geocoding",
      "resourceType": "ExternalService",
      "state": "Running",
      "source": "geocoding-api.open-meteo.com",
      "healthStatus": "Healthy",
      "dashboardUrl": "[redacted dashboardUrl]",
      "relationships": [],
      "urls": [
        {
          "name": "default",
          "url": "https://geocoding-api.open-meteo.com/v1/search"
        }
      ],
      "volumes": [],
      "properties": {
        "resource.source": "geocoding-api.open-meteo.com"
      },
      "environment": "[redacted environment]",
      "healthReports": {
        "geocoding_external_/v1/search?name=Seattle&count=1&language=en&format=json_200_check": {
          "status": "Healthy"
        }
      },
      "commands": {}
    },
    {
      "name": "open-meteo",
      "displayName": "open-meteo",
      "resourceType": "ExternalService",
      "state": "Running",
      "source": "api.open-meteo.com",
      "healthStatus": "Healthy",
      "dashboardUrl": "[redacted dashboardUrl]",
      "relationships": [],
      "urls": [
        {
          "name": "default",
          "url": "https://api.open-meteo.com/"
        }
      ],
      "volumes": [],
      "properties": {
        "resource.source": "api.open-meteo.com"
      },
      "environment": "[redacted environment]",
      "healthReports": {
        "open-meteo_external_/v1/forecast?latitude=0&longitude=0&current=temperature_2m_200_check": {
          "status": "Healthy"
        }
      },
      "commands": {}
    },
    {
      "name": "reverse-geocoding",
      "displayName": "reverse-geocoding",
      "resourceType": "ExternalService",
      "state": "Running",
      "source": "api.bigdatacloud.net",
      "healthStatus": "Healthy",
      "dashboardUrl": "[redacted dashboardUrl]",
      "relationships": [],
      "urls": [
        {
          "name": "default",
          "url": "https://api.bigdatacloud.net/data/reverse-geocode-client"
        }
      ],
      "volumes": [],
      "properties": {
        "resource.source": "api.bigdatacloud.net"
      },
      "environment": "[redacted environment]",
      "healthReports": {
        "reverse-geocoding_external_/data/reverse-geocode-client?latitude=0&longitude=0&localityLanguage=en_200_check": {
          "status": "Healthy"
        }
      },
      "commands": {}
    }
  ]
}

The CLI returns the request's trace, span by span.

Recorded CLI run
aspire otel traces api --apphost 'demo\host\apphost.cs' --format Json --non-interactive
[
  {
    "traceId": "7030bd343ae293c5941aec96764b60a5",
    "durationMs": 13,
    "title": "GET /health",
    "spans": [
      {
        "traceId": "7030bd343ae293c5941aec96764b60a5",
        "spanId": "049b066236a5e0d2",
        "kind": "Server",
        "name": "GET /health",
        "source": "api",
        "destination": "localhost:62869",
        "durationMs": 13,
        "timestamp": "2026-10-12T11:24:42.1521661Z",
        "attributes": {
          "http.request.method": "GET",
          "server.address": "localhost",
          "server.port": "62869",
          "url.scheme": "http",
          "url.path": "/health",
          "network.protocol.version": "1.1",
          "http.route": "/health",
          "http.response.status_code": "200 OK"
        }
      }
    ],
    "hasError": false,
    "timestamp": "2026-10-12T11:24:42.1521661Z",
    "dashboardUrl": "[redacted dashboardUrl]"
  },
  {
    "traceId": "7e37e007d193c9d60fa813ee5f4df1ca",
    "durationMs": 523,
    "title": "GET /api/weather",
    "spans": [
      {
        "traceId": "7e37e007d193c9d60fa813ee5f4df1ca",
        "spanId": "d5cdad39eba8b8fe",
        "kind": "Server",
        "name": "GET /api/weather",
        "source": "api",
        "destination": "localhost:62869",
        "durationMs": 523,
        "timestamp": "2026-10-12T11:24:44.6316734Z",
        "attributes": {
          "http.request.method": "GET",
          "server.address": "localhost",
          "server.port": "62869",
          "user_agent.original": "Mozilla/5.0 (Windows NT 10.0; Microsoft Windows 10.0.26200; en-US) PowerShell/7.6.6",
          "url.scheme": "http",
          "url.path": "/api/weather",
          "url.query": "?lat=Redacted&lon=Redacted",
          "network.protocol.version": "1.1",
          "http.route": "/api/weather",
          "http.response.status_code": "200 OK"
        }
      },
      {
        "traceId": "7e37e007d193c9d60fa813ee5f4df1ca",
        "spanId": "e2d52233242d655c",
        "parentSpanId": "d5cdad39eba8b8fe",
        "kind": "Internal",
        "name": "weather.forecast",
        "source": "api",
        "durationMs": 521,
        "timestamp": "2026-10-12T11:24:44.6323495Z",
        "attributes": {}
      },
      {
        "traceId": "7e37e007d193c9d60fa813ee5f4df1ca",
        "spanId": "2ebb083081915206",
        "parentSpanId": "e2d52233242d655c",
        "kind": "Client",
        "name": "GET",
        "source": "api",
        "destination": "api.bigdatacloud.net:443",
        "durationMs": 209,
        "timestamp": "2026-10-12T11:24:44.6331319Z",
        "attributes": {
          "http.request.method": "GET",
          "server.address": "api.bigdatacloud.net",
          "server.port": "443",
          "url.full": "https://api.bigdatacloud.net/data/reverse-geocode-client?*",
          "http.response.status_code": "200 OK",
          "network.protocol.version": "1.1"
        }
      },
      {
        "traceId": "7e37e007d193c9d60fa813ee5f4df1ca",
        "spanId": "5da9f92c5b2662bb",
        "parentSpanId": "e2d52233242d655c",
        "kind": "Client",
        "name": "GET",
        "source": "api",
        "destination": "api.open-meteo.com:443",
        "durationMs": 520,
        "timestamp": "2026-10-12T11:24:44.6331352Z",
        "attributes": {
          "http.request.method": "GET",
          "server.address": "api.open-meteo.com",
          "server.port": "443",
          "url.full": "https://api.open-meteo.com/v1/forecast?*",
          "http.response.status_code": "200 OK",
          "network.protocol.version": "1.1"
        }
      }
    ],
    "hasError": false,
    "timestamp": "2026-10-12T11:24:44.6316734Z",
    "dashboardUrl": "[redacted dashboardUrl]"
  }
]

Look up, act on one resource, check the result.

skills → CLI → resources → evidence → next decision

Then everyone sees what happened.

Follow one request through your app.

Resource health, traces, logs and metrics in one place.

Weather APIGET /api/weather
  • ForecastParallel provider call
  • Reverse geocodingParallel provider call
Traces
Where time went
Logs
What happened
Metrics
Patterns over time

Find the slow call and the logs that explain it.

An AppHost starts its dashboard.

The Aspire 17.0 dashboard home page listing AspireShop resources with state, source, endpoints, incoming request counts, and p95 latency. Dark theme. The Aspire 17.0 dashboard home page listing AspireShop resources with state, source, endpoints, incoming request counts, and p95 latency. Light theme.
Every resource: state, endpoints, request counts, p95 latency.

Aspire is what these loops produce.

Ask again, with the skills installed.

  1. Let's add Aspire to your solution.aspire docs search
  2. Then start it with aspire run.aspire wait
  3. Docs: aspire.devaspire docs get
  4. .WithHttpHealthCheck()aspire docs api search

Same questions. It checked first.

Stop guessing.
Start checking.

Try it

  • aspire agent init (install the skills)
  • aspire docs search health
  • aspire describe (with an AppHost running)
Take these with youSlides · source · rehearsal

From MCP tools to CLI commands.

Earlier post

llms-small.txt and MCP-first list/search/get tools.

The C# architecture story

This talk

llms-full.txt, CLI docs commands, and a separate API-reference index.

Current pinned implementation

The fetching, parsing, indexing, and retrieval carried over to the new surface.

The first step: microsoft/aspire#14028, merged in January 2026.

C# AppHost running a Go service

AppHosts can be authored in C#, TypeScript, or Go. Here, a C# AppHost runs a Go HTTP service. Workloads can use other languages too.

AppHost.cs
using Aspire.Hosting;

var builder = DistributedApplication.CreateBuilder(args);

builder.AddGoApp("api", "../go-api")
    .WithHttpEndpoint(port: 8080, env: "PORT")
    .WithExternalHttpEndpoints();

builder.Build().Run();
main.go
package main

import (
    "fmt"
    "log"
    "net/http"
    "os"
)

func main() {
    port := os.Getenv("PORT")
    if port == "" { port = "8080" }

    http.HandleFunc("/", func(w http.ResponseWriter, _ *http.Request) {
        fmt.Fprintln(w, "Hello from Go")
    })

    log.Fatal(http.ListenAndServe(":"+port, nil))
}

This example needs a C# AppHost with Aspire.Hosting.Go, the Go toolchain, and a Go module at the illustrative path. Authoring versus workload languages / pinned integration source. A TypeScript AppHost is apphost.mts. Illustrative; not executed here.

Reported outcomes, May 3 to June 2, 2026.

82feature-docs PRs merged

44.8hmedian docs time-to-merge

Historical, human-reviewed figures as reported, not a service-level promise.

Source and measurement context

The CLI, dashboard, and API are NativeAOT.

The AppHost stays managed .NET. The browser runs React; Node is only for frontend development and build.

Deploy the model you already run.

Describe services and dependencies once; use them in development and deployment.

AppHost modelServices, configuration, relationships
  • RunLocal processes: native API + Vite
  • Publish + deployCompose: API / React + dashboard

The AppHost produces the deployment; the deployed app runs independently.

Add the target to the same model.

AppHost modelTarget integrationEnvironment resource
Docker integration
aspire add docker --version 13.6.4 --non-interactive
Compose environment
builder.AddDockerComposeEnvironment("talk-compose")
    .WithDashboard(dashboard => dashboard
        .WithImage("aspire/dashboard", "13.6.2")
        .WithImageRegistry("mcr.microsoft.com"));

The native API serves the bundled React app.

Publish writes the deployment files.

Recorded CLI run
aspire publish --apphost demo\host\apphost.cs `
  --environment Talk --output-path .artifacts\chapter-demos\compose `
  --non-interactive
🛠️ Executing step publish
🛠️ Building AppHost... demo\host\apphost.cs
12:40:16 (validate-compute-environments) → Starting validate-compute-environments...
12:40:16 (validate-compute-environments) ✓ validate-compute-environments completed successfully
12:40:16 (prepare-deployment-targets-talk-compose) → Starting prepare-deployment-targets-talk-compose...
12:40:16 (prepare-deployment-targets-talk-compose) i [INF] Creating Docker Compose resource for talk-compose-dashboard
12:40:16 (prepare-deployment-targets-talk-compose) i [INF] Creating Docker Compose resource for api
12:40:16 (prepare-deployment-targets-talk-compose) ✓ prepare-deployment-targets-talk-compose completed successfully
12:40:16 (validate-docker-compose) → Starting validate-docker-compose...
12:40:16 (validate-docker-compose) ✓ validate-docker-compose completed successfully
12:40:16 (before-start) → Starting before-start...
12:40:16 (before-start) ✓ before-start completed successfully
12:40:16 (pipeline-execution) → Starting pipeline-execution...
12:40:16 (publish-talk-compose) → Starting publish-talk-compose...
12:40:16 (validate-javascript-dockerfile-run-script-frontend) → Starting validate-javascript-dockerfile-run-script-frontend...
12:40:16 (validate-javascript-dockerfile-run-script-frontend) ✓ validate-javascript-dockerfile-run-script-frontend completed successfully
12:40:16 (publish-talk-compose) i [INF] Generating Compose output
12:40:16 (publish-talk-compose) → Writing the Docker Compose file to the output path.
12:40:16 (publish-talk-compose) ✓ Docker Compose file written successfully to
<WORKTREE>\.artifacts\chapter-demos\compose\docker-compose.yaml. (0.0s)
12:40:16 (publish-talk-compose) ✓ publish-talk-compose completed successfully
12:40:16 (publish) → Starting publish...
12:40:16 (publish) ✓ publish completed successfully
12:40:16 (pipeline-execution) ✓ Completed successfully
------------------------------------------------------------
✓ 8/8 steps succeeded • Total time: 36.98ms

Steps Summary:
                Step timeline:                                      0ms                    10.82ms
                                                                    │───────┬──────┬─────┬───────│
      1.11ms  ✓ validate-compute-environments                       │╶──╴                        │
      2.72ms  ✓   prepare-deployment-targets-talk-compose           │   ╶───────╴                │
      0.40ms  ✓     before-start                                    │                ╴           │
      0.40ms  ✓ validate-docker-compose                             │              ╴             │
      3.80ms  ✓ pipeline-execution                                  │                 ╶─────────╴│
      2.44ms  ✓ publish-talk-compose                                │                  ╶──────╴  │
      0.32ms  ✓   publish                                           │                          ╴ │
      0.35ms  ✓ validate-javascript-dockerfile-run-script-frontend  │                    ╴       │

✓ Pipeline succeeded
------------------------------------------------------------

The output is docker-compose.yaml and .env placeholders. No images build and no containers start.

Deploy runs the app, not the AppHost.

Recorded CLI run
aspire deploy --apphost demo\host\apphost.cs `
  --environment Talk --output-path .artifacts\chapter-demos\compose `
  --non-interactive
🛠️ Executing step deploy
🛠️ Building AppHost... demo\host\apphost.cs
12:42:19 (validate-compute-environments) → Starting validate-compute-environments...
12:42:19 (validate-compute-environments) ✓ validate-compute-environments completed successfully
12:42:19 (prepare-deployment-targets-talk-compose) → Starting prepare-deployment-targets-talk-compose...
12:42:19 (prepare-deployment-targets-talk-compose) i [INF] Creating Docker Compose resource for talk-compose-dashboard
12:42:19 (prepare-deployment-targets-talk-compose) i [INF] Creating Docker Compose resource for api
12:42:19 (prepare-deployment-targets-talk-compose) ✓ prepare-deployment-targets-talk-compose completed successfully
12:42:19 (validate-docker-compose) → Starting validate-docker-compose...
12:42:19 (validate-docker-compose) ✓ validate-docker-compose completed successfully
12:42:19 (before-start) → Starting before-start...
12:42:19 (before-start) ✓ before-start completed successfully
12:42:19 (pipeline-execution) → Starting pipeline-execution...
12:42:19 (validate-javascript-dockerfile-run-script-frontend) → Starting validate-javascript-dockerfile-run-script-frontend...
12:42:19 (process-parameters) → Starting process-parameters...
12:42:19 (validate-compute-environments) → Starting validate-compute-environments...
12:42:19 (publish-talk-compose) → Starting publish-talk-compose...
12:42:19 (validate-compute-environments) ✓ validate-compute-environments completed successfully
12:42:19 (check-container-runtime) → Starting check-container-runtime...
12:42:19 (validate-build-only-container-references) → Starting validate-build-only-container-references...
12:42:19 (push-prereq) → Starting push-prereq...
12:42:19 (validate-javascript-dockerfile-run-script-frontend) ✓ validate-javascript-dockerfile-run-script-frontend completed successfully
12:42:19 (validate-build-only-container-references) ✓ validate-build-only-container-references completed successfully
12:42:19 (push-prereq) ✓ push-prereq completed successfully
12:42:19 (publish-talk-compose) i [INF] Generating Compose output
12:42:19 (process-parameters) ✓ process-parameters completed successfully
12:42:19 (build-prereq) → Starting build-prereq...
12:42:19 (deploy-prereq) → Starting deploy-prereq...
12:42:19 (build-prereq) ✓ build-prereq completed successfully
12:42:19 (deploy-prereq) i [INF] Initializing deployment for environment 'Talk'
12:42:19 (deploy-prereq) i [INF] Setting default deploy tag 'aspire-deploy-20261105184219' for compute resource(s).
12:42:19 (deploy-prereq) ✓ deploy-prereq completed successfully
12:42:19 (publish-talk-compose) → Writing the Docker Compose file to the output path.
12:42:19 (publish-talk-compose) ✓ Docker Compose file written successfully to
<WORKTREE>\.artifacts\chapter-demos\compose\docker-compose.yaml. (0.0s)
12:42:19 (publish-talk-compose) ✓ publish-talk-compose completed successfully
12:42:19 (publish) → Starting publish...
12:42:19 (publish) ✓ publish completed successfully
12:42:19 (check-container-runtime) i [INF] Container runtime auto-detected: Docker (docker).
12:42:19 (check-container-runtime) ✓ check-container-runtime completed successfully
12:42:19 (build-api) → Starting build-api...
12:42:19 (build-api) i [INF] Building container image for resource api
12:42:19 (build-api) i [INF] Building image: api
12:43:43 (build-api) i [INF] Building image for api completed
12:43:43 (build-api) ✓ build-api completed successfully
12:43:43 (build) → Starting build...
12:43:43 (build) ✓ build completed successfully
12:43:43 (prepare-talk-compose) → Starting prepare-talk-compose...
12:43:43 (push-api) → Starting push-api...
12:43:43 (prepare-talk-compose) ✓ prepare-talk-compose completed successfully
12:43:43 (push-api) → Tagging api for local use
12:43:43 (push-api) i [INF] Docker tag for api:a213d6d515beca1dc295332bb130df5e4f913d22 -> api:aspire-deploy-20261105184219 succeeded.
12:43:43 (push-api) ✓ Successfully tagged api as api:aspire-deploy-20261105184219 (0.2s)
12:43:43 (push-api) ✓ push-api completed successfully
12:43:43 (docker-compose-up-talk-compose) → Starting docker-compose-up-talk-compose...
12:43:43 (docker-compose-up-talk-compose) → Running compose up for talk-compose using Docker
12:43:43 (docker-compose-up-talk-compose) i [INF] Using container runtime 'docker' for compose operations.
12:43:51 (docker-compose-up-talk-compose) ✓ Service talk-compose is now running with Docker Compose locally (runtime: Docker) (7.9s)
12:43:51 (docker-compose-up-talk-compose) ✓ docker-compose-up-talk-compose completed successfully
12:43:51 (print-api-summary) → Starting print-api-summary...
12:43:51 (print-talk-compose-dashboard-summary) → Starting print-talk-compose-dashboard-summary...
12:43:51 (print-talk-compose-dashboard-summary) i [INF] Successfully deployed talk-compose-dashboard to http://localhost:62881.
12:43:51 (print-talk-compose-dashboard-summary) ✓ print-talk-compose-dashboard-summary completed successfully
12:43:51 (print-api-summary) i [INF] Successfully deployed api to http://localhost:62880.
12:43:51 (print-api-summary) ✓ print-api-summary completed successfully
12:43:51 (deploy) → Starting deploy...
12:43:51 (deploy) ✓ deploy completed successfully
12:43:51 (pipeline-execution) ✓ Completed successfully
------------------------------------------------------------
✓ 23/23 steps succeeded • Total time: 1m 33s

Steps Summary:
                Step timeline:                                      0s                      1m 33s
                                                                    │───────┬──────┬─────┬───────│
      1.35ms  ✓ validate-compute-environments                       │╴                           │
      2.76ms  ✓   prepare-deployment-targets-talk-compose           │╴                           │
      0.36ms  ✓     before-start                                    │╴                           │
      0.41ms  ✓ validate-docker-compose                             │╴                           │
      1m 33s  ✓ pipeline-execution                                  │╶──────────────────────────╴│
      2.95ms  ✓ validate-javascript-dockerfile-run-script-frontend  │╴                           │
      3.94ms  ✓ process-parameters                                  │╴                           │
      0.73ms  ✓   build-prereq                                      │╴                           │
      1m 23s  ✓     build-api                                       │╶────────────────────────╴  │
      0.53ms  ✓       build                                         │                        ╴   │
      3.99ms  ✓         prepare-talk-compose                        │                        ╴   │
       7.91s  ✓           docker-compose-up-talk-compose            │                        ╶──╴│
       0.63s  ✓             print-api-summary                       │                           ╴│
      0.46ms  ✓               deploy                                │                           ╴│
       0.63s  ✓             print-talk-compose-dashboard-summary    │                           ╴│
       0.20s  ✓       push-api                                      │                        ╴   │
      1.52ms  ✓   deploy-prereq                                     │╴                           │
      0.74ms  ✓ validate-compute-environments                       │╴                           │
      6.77ms  ✓ publish-talk-compose                                │╴                           │
      0.39ms  ✓   publish                                           │╴                           │
       0.43s  ✓ check-container-runtime                             │╴                           │
      1.06ms  ✓ validate-build-only-container-references            │╴                           │
      1.04ms  ✓ push-prereq                                         │╴                           │

✓ Pipeline succeeded

  talk-compose-dashboard: http://localhost:62881
  api: http://localhost:62880
------------------------------------------------------------

The current model builds the native image, resolves parameters and starts Compose. The API still serves the React app.

Each target needs three decisions.

Target

Pick a supported integration and environment.

Permission

Provide parameters, secrets, and credentials, and approve side effects.

Verification

Check the deployed resources and their health.

The aspire-deployment skill guides the steps. The CLI and the target's provider carry them out.

Same viewer. No AppHost required.

Standalone CLI
aspire dashboard run --non-interactive
Standalone OCI image
docker run --rm `
  -p "127.0.0.1:18888:18888" `
  -p "127.0.0.1:4317:18889" `
  -p "127.0.0.1:4318:18890" `
  mcr.microsoft.com/aspire/dashboard:latest

Anything that exports OTLP can send telemetry here. The standalone dashboard shows it; it doesn't start or manage your processes.

Our demo: a weather app.

The actual weather app showing Madison, Wisconsin, live weather, five forecast days, and a loaded map with provider attribution in dark mode. The actual weather app showing Madison, Wisconsin, live weather, five forecast days, and a loaded map with provider attribution in light mode.
Madison, Wisconsin, from the running app. Rehearsal capture; conditions change. Data providers credited in the app. Original MIT sample.

Sources and reuse