diff --git a/Memory/ARCHITECTURE.md b/Memory/ARCHITECTURE.md
index 7b731185..4994a48c 100644
--- a/Memory/ARCHITECTURE.md
+++ b/Memory/ARCHITECTURE.md
@@ -155,6 +155,13 @@ adapters at the LLM boundary. Models without tool capability receive no tools.
 Stop cancels the active stream and actions that have not started; steering
 messages are consumed at the next turn boundary.
 
+Model purposes are `chat`, `embedding`, `rerank`, `image_generation`, and
+`video_generation`. Only recognised purpose values from model discovery are
+asserted as reported metadata. Unknown values, including Anthropic's `model`
+object discriminator, leave catalog and name inference available and otherwise
+default to chat. Only chat models enter conversation pickers. Persisted unknown
+purpose strings also default to chat; there are no legacy purpose aliases.
+
 A turn's tool calls run concurrently only when every call in it is marked
 parallel; one sequential call makes the whole turn serial rather than being
 fenced into its own segment. Concurrency is declared per tool and may be
diff --git a/Termind.xcodeproj/project.pbxproj b/Termind.xcodeproj/project.pbxproj
index e8082250..0c15ac93 100644
--- a/Termind.xcodeproj/project.pbxproj
+++ b/Termind.xcodeproj/project.pbxproj
@@ -675,7 +675,7 @@
 			repositoryURL = "https://github.com/AkinoKaede/AgentKit.git";
 			requirement = {
 				kind = upToNextMajorVersion;
-				minimumVersion = 0.5.0;
+				minimumVersion = 0.6.0;
 			};
 		};
 /* End XCRemoteSwiftPackageReference section */
diff --git a/Termind.xcodeproj/project.xcworkspace/xcshareddata/swiftpm/Package.resolved b/Termind.xcodeproj/project.xcworkspace/xcshareddata/swiftpm/Package.resolved
index 34ed7bed..a4807c5d 100644
--- a/Termind.xcodeproj/project.xcworkspace/xcshareddata/swiftpm/Package.resolved
+++ b/Termind.xcodeproj/project.xcworkspace/xcshareddata/swiftpm/Package.resolved
@@ -6,8 +6,8 @@
       "kind" : "remoteSourceControl",
       "location" : "https://github.com/AkinoKaede/AgentKit.git",
       "state" : {
-        "revision" : "a7345ee68fd92b52589cecfc812517c089f39368",
-        "version" : "0.5.0"
+        "revision" : "ad0ab2f4f707657611faa671d41f229ff019d4e8",
+        "version" : "0.6.0"
       }
     },
     {
diff --git a/TermindTests/AIProviderPersistenceTests.swift b/TermindTests/AIProviderPersistenceTests.swift
index 3d8bd477..362b8796 100644
--- a/TermindTests/AIProviderPersistenceTests.swift
+++ b/TermindTests/AIProviderPersistenceTests.swift
@@ -336,6 +336,32 @@ struct AIProviderPersistenceTests {
         #expect(store.resolve(conversation: nil)?.model.id == hidden.id)
     }
 
+    @Test(arguments: ["image", "other", "unknown-purpose"])
+    func unrecognisedStoredModelKindsLoadAsChat(kind: String) throws {
+        let container = try makeContainer("UnknownModelKind-" + kind)
+        let provider = ModelProvider(name: "Gateway", models: [AIModel(id: "private-chat")])
+        let store = AIConfigurationModel(modelContext: container.mainContext, persistsSecrets: false)
+        store.upsert(provider)
+        let row = try #require(container.mainContext.fetch(FetchDescriptor<PersistentAIModel>()).first)
+        row.kind = kind
+        try container.mainContext.save()
+
+        let reloaded = AIConfigurationModel(modelContext: container.mainContext, persistsSecrets: false)
+        let restored = try #require(reloaded.modelProvider(provider.id))
+        #expect(restored.models.first?.kind == .chat)
+        #expect(restored.assistantModels.map(\.id) == ["private-chat"])
+    }
+
+    @Test(arguments: AIModel.Kind.allCases)
+    func modelPurposesRoundTripThroughPersistence(kind: AIModel.Kind) throws {
+        let container = try makeContainer("ModelPurpose-" + kind.rawValue)
+        let provider = ModelProvider(name: "Gateway", models: [AIModel(id: "custom", kind: kind)])
+        let store = AIConfigurationModel(modelContext: container.mainContext, persistsSecrets: false)
+        store.upsert(provider)
+        let reloaded = AIConfigurationModel(modelContext: container.mainContext, persistsSecrets: false)
+        #expect(reloaded.modelProvider(provider.id)?.models.first?.kind == kind)
+    }
+
     private func makeContainer(_ name: String) throws -> ModelContainer {
         let configuration = localModelConfiguration(
             name, schema: HostPersistence.schema, isStoredInMemoryOnly: true
@@ -368,6 +394,17 @@ private enum TestPersistenceError: Error {
 
 @Suite
 struct ModelCapabilityResolverTests {
+    @Test(arguments: ["model", "unknown-purpose", "", " \n"])
+    func unrecognisedWireTypeAllowsCatalogClassification(type: String) throws {
+        let data = try JSONSerialization.data(withJSONObject: [
+            "data": [["id": "gpt-5.4", "type": type]]
+        ])
+        let model = try #require(ModelCatalogClient.parseOpenAIPage(data).models.first)
+        #expect(model.kind == .chat)
+        #expect(model.contextLength == 272_000)
+        #expect(model.abilities.contains(.toolCall))
+    }
+
     @Test
     func bundledPiCatalogEnrichesKnownModelWithoutRegex() {
         let provider = ModelProvider(name: "OpenAI")
@@ -376,8 +413,8 @@ struct ModelCapabilityResolverTests {
             provider: provider
         )
 
-        #expect(ModelCapabilityResolver.catalogMetadata?.version == "0.84.3")
-        #expect(ModelCapabilityResolver.catalogMetadata?.modelCount == 1_312)
+        #expect(ModelCapabilityResolver.catalogMetadata?.version == "0.85.1")
+        #expect(ModelCapabilityResolver.catalogMetadata?.modelCount == 1_354)
         #expect(resolution.catalogSource == .init(provider: "openai", modelID: "gpt-5.4"))
         #expect(resolution.model.input == [.text, .image])
         // `webSearch` is the one ability that does come from the model id, and
@@ -542,7 +579,7 @@ struct ModelCapabilityResolverTests {
     func reportedFieldsWinOverCodexAutoReviewOverride() {
         let provider = ModelProvider(name: "OpenAI")
         var model = AIModel(id: "codex-auto-review")
-        model.kind = .other
+        model.kind = .embedding
         model.input = [.audio]
         model.output = [.image]
         model.abilities = []
@@ -555,7 +592,7 @@ struct ModelCapabilityResolverTests {
             reported: [.kind, .input, .output, .abilities]
         )
 
-        #expect(resolution.model.kind == .other)
+        #expect(resolution.model.kind == .embedding)
         #expect(resolution.model.input == [.audio])
         #expect(resolution.model.output == [.image])
         #expect(resolution.model.abilities.isEmpty)
@@ -568,7 +605,7 @@ struct ModelCapabilityResolverTests {
     func explicitlyReportedFieldsWinOverPiCatalog() {
         let provider = ModelProvider(name: "OpenAI")
         var model = AIModel(id: "gpt-5.4")
-        model.kind = .other
+        model.kind = .embedding
         model.input = [.audio]
         model.output = [.image]
         model.abilities = []
@@ -581,7 +618,7 @@ struct ModelCapabilityResolverTests {
             reported: [.kind, .input, .output, .abilities]
         )
 
-        #expect(resolution.model.kind == .other)
+        #expect(resolution.model.kind == .embedding)
         #expect(resolution.model.input == [.audio])
         #expect(resolution.model.output == [.image])
         // Reported abilities still win — `toolCall` and `reasoning` stay off