Ananke.Federation 0.8.5

Ananke.Federation

Federation hub for Ananke. Provides the core abstractions, default implementations, and shared infrastructure that platform adapters (Ananke.Federation.Azure, Ananke.Federation.Google, Ananke.Federation.Anthropic) are built on.

Also ships the local design loop infrastructure — run and test workflows that declare ToolExecutionMode.PlatformNative capabilities on your developer machine or in CI without credentials. See Local design loop below, and for the full emulator catalogue.

What this package provides

Architecture

Federation follows the supervisor-only hybrid model:

• Remote cells are self-contained — no cross-boundary tool callbacks.
• The local OrganicHost supervises all cells (local and remote) through FederatedComplexityMonitor.
• FederatedDivisionPolicy wraps an inner IDivisionPolicy and sets TargetPlatform on child specs.
• PlatformDivisionApprovalGate always requires human oversight before any cross-boundary deployment.
• RemoteMetricsTracker emits OTEL gauges via the Ananke.Federation meter name.

Quick start

// 1. Validate a manifest against a target platform
var validator = new DeployabilityValidator();
DeployabilityReport report = await validator.ValidateAsync(manifest, "azure-ai");
if (report.HasErrors) { /* handle */ }

// 2. Deploy (via a platform adapter)
var deployer = new AzureAgentDeployer(credentials, registry);
DeploymentRecord record = await deployer.DeployAsync(manifest, toolKit, options);

// 3. Wrap your local monitor for federated complexity tracking
var monitor = new FederatedComplexityMonitor(
 localMonitor: myLocalMonitor,
 registry: registry,
 remoteMonitors: [new AzureRemoteCellMonitor()],
 metricsTracker: metricsTracker);

// 4. Use the nnke-platform CLI for day-2 operations
// nnke-platform validate manifest.ananke.yml
// nnke-platform deploy manifest.ananke.yml
// nnke-platform trends --deployment-id <id>
// nnke-platform analyze manifest.ananke.yml --deployment-id <id>

Local design loop

Workflows that target a managed platform can be authored, executed, and tested locally before any deployment. This enables offline authoring, CI without credentials, and fast iteration on platform-native capability usage.

How it works

1. Declare capabilities as usual with ToolExecutionMode.PlatformNative.
2. Register emulators from Ananke.Federation.LocalEmulators:

using Ananke.Federation.Execution;
using Ananke.Federation.LocalEmulators;

var registry = new PlatformNativeExecutorRegistry();
DefaultPlatformNativeExecutors.Register(registry);

// Patch a ToolKit so platform-native tools execute locally
registry.ApplyTo(toolKit, platform: "azure-ai");

3. Deploy locally using LocalFederationDeployer — no network, no credentials:

var deployer = new LocalFederationDeployer(new InMemoryDeploymentRegistry());
var record = await deployer.DeployAsync(manifest, toolKit, new DeployOptions { Platform = "local" });
// record.Platform == "local", record.Status == DeploymentStatus.Active

4. Validate capability coverage with LocalPlatformValidator:

var validator = new LocalPlatformValidator(registry, new DeployabilityValidator());
var report = await validator.ValidateAsync(manifest, toolKit, "local-emulated:azure-ai");
// FED061 — capability declared but no executor registered
// FED062 — capability is covered by a stub (deterministic, not real)

Routing tier

HybridRouter supports three routing targets:

// Pin a cell to local emulation of Foundry
var rule = RoutingRule.EmulateAll("foundry"); // target: "local-emulated:azure-ai"

Platform identifier aliases

The post-May-2026 platform names are accepted everywhere:

Existing manifests authored with azure-ai / vertex-ai continue to work unchanged.

Diagnostic codes (local loop)

Deployment profiles

A DeploymentProfile describes a named target environment and rebinds tools for platform-native execution. Profiles are declared in .ananke.yml under profiles: and parsed by the Ananke.Design manifest parser.

profiles:
 - id: prod-azure
 platform: azure-ai
 tools:
 - name: web_search
 binding: bing_grounding

OTEL integration

RemoteMetricsTracker emits observable gauge measurements. Add the meter to your OpenTelemetry pipeline to export trends to Prometheus, Grafana, or any OTLP backend:

builder.AddMeter(RemoteMetricsTracker.MeterName); // "Ananke.Federation"

Credentials

Platform credentials are never stored in manifests. See for the full credentials matrix: credential types, required scopes, rotation procedures, and per-platform IFederationCredentialProvider implementation status.

Release checklist

Before each release, update the platform capabilities list in :

1. Azure AI Agent Service — check the Azure.AI.Projects.Agents SDK for new tool types (*Tool classes). The SDK changelog and Azure AI Agent Service docs list newly GA'd capabilities.
2. Vertex AI — check the Google.Cloud.AIPlatform.V1 SDK and Vertex AI Agent Engine docs for new tool types.
3. Claude — check the Anthropic tool use docs for new built-in tools.

The JSON file is embedded as a resource and loaded by DeployabilityValidator at startup. Unknown capabilities produce a warning (FED003), not an error — the platform API validates at deploy time. This list only improves pre-deploy DX.