@otto-code/brain 0.8.10 → 0.8.13

package/dist/config/schema.d.ts

@@ -6,6 +6,34 @@
  */
 import { z } from "zod";
 export declare const DEFAULT_REASONING_MESSAGE = "Enough analysis. Write the complete answer now.";
+/**
+ * Model-native names for reasoning controls exposed by its chat template.
+ * These are catalog metadata, not caller-provided flags: the host owns the
+ * translation from the OpenAI-compatible request into template arguments.
+ */
+export declare const ReasoningTemplateSchema: z.ZodObject<{
+    enableThinkingArgument: z.ZodString;
+    effortArgument: z.ZodString;
+}, "strict", z.ZodTypeAny, {
+    enableThinkingArgument: string;
+    effortArgument: string;
+}, {
+    enableThinkingArgument: string;
+    effortArgument: string;
+}>;
+export type ReasoningTemplate = z.infer<typeof ReasoningTemplateSchema>;
+/** A template's native spelling for Brain's provider-neutral preservation setting. */
+export declare const ReasoningPreservationSchema: z.ZodObject<{
+    templateArgument: z.ZodString;
+    default: z.ZodOptional<z.ZodBoolean>;
+}, "strict", z.ZodTypeAny, {
+    templateArgument: string;
+    default?: boolean | undefined;
+}, {
+    templateArgument: string;
+    default?: boolean | undefined;
+}>;
+export type ReasoningPreservation = z.infer<typeof ReasoningPreservationSchema>;
 export declare const ProfileSchema: z.ZodObject<{
     modelId: z.ZodDefault<z.ZodNullable<z.ZodString>>;
     modelPath: z.ZodDefault<z.ZodNullable<z.ZodString>>;
@@ -22,7 +50,31 @@ export declare const ProfileSchema: z.ZodObject<{
     vision: z.ZodDefault<z.ZodBoolean>;
     reasoningBudget: z.ZodDefault<z.ZodNumber>;
     reasoningBudgetMessage: z.ZodDefault<z.ZodString>;
+    /**
+     * Tri-state, matching llama-server's own `--reasoning-preserve` /
+     * `--no-reasoning-preserve` / unset: `true` keeps the reasoning trace in the
+     * whole history, `false` forces it to the last assistant message only, and
+     * null/undefined leaves the template's own default alone. Null is the
+     * default precisely so an untouched profile emits no flag and launches
+     * exactly as it did before this field existed.
+     */
+    preserveReasoning: z.ZodOptional<z.ZodNullable<z.ZodBoolean>>;
+    temperature: z.ZodDefault<z.ZodNumber>;
+    topP: z.ZodDefault<z.ZodNumber>;
+    topK: z.ZodDefault<z.ZodNumber>;
+    minP: z.ZodDefault<z.ZodNumber>;
+    presencePenalty: z.ZodDefault<z.ZodNumber>;
+    repeatPenalty: z.ZodDefault<z.ZodNumber>;
     parallelSlots: z.ZodDefault<z.ZodNumber>;
+    /**
+     * How many chats' KV state llama-server may park in system RAM when they
+     * lose their GPU slot, so returning to one costs a bulk copy instead of a
+     * full re-prefill. Stored as a count, not a size: the byte budget it turns
+     * into (`--cache-ram`) depends on the measured KV bytes/token and the
+     * per-slot context, both of which move when other fields are edited.
+     * 0 means "leave llama.cpp's own default alone" - the flag is not emitted.
+     */
+    cachedChats: z.ZodDefault<z.ZodNumber>;
     /** RoPE extension factor; 1 keeps the GGUF-native context window. */
     contextMultiplier: z.ZodDefault<z.ZodNumber>;
     /** Cleared only by a successful calibration of this saved model profile. */
@@ -68,7 +120,31 @@ export declare const ProfileSchema: z.ZodObject<{
     vision: z.ZodDefault<z.ZodBoolean>;
     reasoningBudget: z.ZodDefault<z.ZodNumber>;
     reasoningBudgetMessage: z.ZodDefault<z.ZodString>;
+    /**
+     * Tri-state, matching llama-server's own `--reasoning-preserve` /
+     * `--no-reasoning-preserve` / unset: `true` keeps the reasoning trace in the
+     * whole history, `false` forces it to the last assistant message only, and
+     * null/undefined leaves the template's own default alone. Null is the
+     * default precisely so an untouched profile emits no flag and launches
+     * exactly as it did before this field existed.
+     */
+    preserveReasoning: z.ZodOptional<z.ZodNullable<z.ZodBoolean>>;
+    temperature: z.ZodDefault<z.ZodNumber>;
+    topP: z.ZodDefault<z.ZodNumber>;
+    topK: z.ZodDefault<z.ZodNumber>;
+    minP: z.ZodDefault<z.ZodNumber>;
+    presencePenalty: z.ZodDefault<z.ZodNumber>;
+    repeatPenalty: z.ZodDefault<z.ZodNumber>;
     parallelSlots: z.ZodDefault<z.ZodNumber>;
+    /**
+     * How many chats' KV state llama-server may park in system RAM when they
+     * lose their GPU slot, so returning to one costs a bulk copy instead of a
+     * full re-prefill. Stored as a count, not a size: the byte budget it turns
+     * into (`--cache-ram`) depends on the measured KV bytes/token and the
+     * per-slot context, both of which move when other fields are edited.
+     * 0 means "leave llama.cpp's own default alone" - the flag is not emitted.
+     */
+    cachedChats: z.ZodDefault<z.ZodNumber>;
     /** RoPE extension factor; 1 keeps the GGUF-native context window. */
     contextMultiplier: z.ZodDefault<z.ZodNumber>;
     /** Cleared only by a successful calibration of this saved model profile. */
@@ -114,7 +190,31 @@ export declare const ProfileSchema: z.ZodObject<{
     vision: z.ZodDefault<z.ZodBoolean>;
     reasoningBudget: z.ZodDefault<z.ZodNumber>;
     reasoningBudgetMessage: z.ZodDefault<z.ZodString>;
+    /**
+     * Tri-state, matching llama-server's own `--reasoning-preserve` /
+     * `--no-reasoning-preserve` / unset: `true` keeps the reasoning trace in the
+     * whole history, `false` forces it to the last assistant message only, and
+     * null/undefined leaves the template's own default alone. Null is the
+     * default precisely so an untouched profile emits no flag and launches
+     * exactly as it did before this field existed.
+     */
+    preserveReasoning: z.ZodOptional<z.ZodNullable<z.ZodBoolean>>;
+    temperature: z.ZodDefault<z.ZodNumber>;
+    topP: z.ZodDefault<z.ZodNumber>;
+    topK: z.ZodDefault<z.ZodNumber>;
+    minP: z.ZodDefault<z.ZodNumber>;
+    presencePenalty: z.ZodDefault<z.ZodNumber>;
+    repeatPenalty: z.ZodDefault<z.ZodNumber>;
     parallelSlots: z.ZodDefault<z.ZodNumber>;
+    /**
+     * How many chats' KV state llama-server may park in system RAM when they
+     * lose their GPU slot, so returning to one costs a bulk copy instead of a
+     * full re-prefill. Stored as a count, not a size: the byte budget it turns
+     * into (`--cache-ram`) depends on the measured KV bytes/token and the
+     * per-slot context, both of which move when other fields are edited.
+     * 0 means "leave llama.cpp's own default alone" - the flag is not emitted.
+     */
+    cachedChats: z.ZodDefault<z.ZodNumber>;
     /** RoPE extension factor; 1 keeps the GGUF-native context window. */
     contextMultiplier: z.ZodDefault<z.ZodNumber>;
     /** Cleared only by a successful calibration of this saved model profile. */
@@ -287,7 +387,31 @@ export declare const ProfilesStoreSchema: z.ZodObject<{
         vision: z.ZodDefault<z.ZodBoolean>;
         reasoningBudget: z.ZodDefault<z.ZodNumber>;
         reasoningBudgetMessage: z.ZodDefault<z.ZodString>;
+        /**
+         * Tri-state, matching llama-server's own `--reasoning-preserve` /
+         * `--no-reasoning-preserve` / unset: `true` keeps the reasoning trace in the
+         * whole history, `false` forces it to the last assistant message only, and
+         * null/undefined leaves the template's own default alone. Null is the
+         * default precisely so an untouched profile emits no flag and launches
+         * exactly as it did before this field existed.
+         */
+        preserveReasoning: z.ZodOptional<z.ZodNullable<z.ZodBoolean>>;
+        temperature: z.ZodDefault<z.ZodNumber>;
+        topP: z.ZodDefault<z.ZodNumber>;
+        topK: z.ZodDefault<z.ZodNumber>;
+        minP: z.ZodDefault<z.ZodNumber>;
+        presencePenalty: z.ZodDefault<z.ZodNumber>;
+        repeatPenalty: z.ZodDefault<z.ZodNumber>;
         parallelSlots: z.ZodDefault<z.ZodNumber>;
+        /**
+         * How many chats' KV state llama-server may park in system RAM when they
+         * lose their GPU slot, so returning to one costs a bulk copy instead of a
+         * full re-prefill. Stored as a count, not a size: the byte budget it turns
+         * into (`--cache-ram`) depends on the measured KV bytes/token and the
+         * per-slot context, both of which move when other fields are edited.
+         * 0 means "leave llama.cpp's own default alone" - the flag is not emitted.
+         */
+        cachedChats: z.ZodDefault<z.ZodNumber>;
         /** RoPE extension factor; 1 keeps the GGUF-native context window. */
         contextMultiplier: z.ZodDefault<z.ZodNumber>;
         /** Cleared only by a successful calibration of this saved model profile. */
@@ -333,7 +457,31 @@ export declare const ProfilesStoreSchema: z.ZodObject<{
         vision: z.ZodDefault<z.ZodBoolean>;
         reasoningBudget: z.ZodDefault<z.ZodNumber>;
         reasoningBudgetMessage: z.ZodDefault<z.ZodString>;
+        /**
+         * Tri-state, matching llama-server's own `--reasoning-preserve` /
+         * `--no-reasoning-preserve` / unset: `true` keeps the reasoning trace in the
+         * whole history, `false` forces it to the last assistant message only, and
+         * null/undefined leaves the template's own default alone. Null is the
+         * default precisely so an untouched profile emits no flag and launches
+         * exactly as it did before this field existed.
+         */
+        preserveReasoning: z.ZodOptional<z.ZodNullable<z.ZodBoolean>>;
+        temperature: z.ZodDefault<z.ZodNumber>;
+        topP: z.ZodDefault<z.ZodNumber>;
+        topK: z.ZodDefault<z.ZodNumber>;
+        minP: z.ZodDefault<z.ZodNumber>;
+        presencePenalty: z.ZodDefault<z.ZodNumber>;
+        repeatPenalty: z.ZodDefault<z.ZodNumber>;
         parallelSlots: z.ZodDefault<z.ZodNumber>;
+        /**
+         * How many chats' KV state llama-server may park in system RAM when they
+         * lose their GPU slot, so returning to one costs a bulk copy instead of a
+         * full re-prefill. Stored as a count, not a size: the byte budget it turns
+         * into (`--cache-ram`) depends on the measured KV bytes/token and the
+         * per-slot context, both of which move when other fields are edited.
+         * 0 means "leave llama.cpp's own default alone" - the flag is not emitted.
+         */
+        cachedChats: z.ZodDefault<z.ZodNumber>;
         /** RoPE extension factor; 1 keeps the GGUF-native context window. */
         contextMultiplier: z.ZodDefault<z.ZodNumber>;
         /** Cleared only by a successful calibration of this saved model profile. */
@@ -379,7 +527,31 @@ export declare const ProfilesStoreSchema: z.ZodObject<{
         vision: z.ZodDefault<z.ZodBoolean>;
         reasoningBudget: z.ZodDefault<z.ZodNumber>;
         reasoningBudgetMessage: z.ZodDefault<z.ZodString>;
+        /**
+         * Tri-state, matching llama-server's own `--reasoning-preserve` /
+         * `--no-reasoning-preserve` / unset: `true` keeps the reasoning trace in the
+         * whole history, `false` forces it to the last assistant message only, and
+         * null/undefined leaves the template's own default alone. Null is the
+         * default precisely so an untouched profile emits no flag and launches
+         * exactly as it did before this field existed.
+         */
+        preserveReasoning: z.ZodOptional<z.ZodNullable<z.ZodBoolean>>;
+        temperature: z.ZodDefault<z.ZodNumber>;
+        topP: z.ZodDefault<z.ZodNumber>;
+        topK: z.ZodDefault<z.ZodNumber>;
+        minP: z.ZodDefault<z.ZodNumber>;
+        presencePenalty: z.ZodDefault<z.ZodNumber>;
+        repeatPenalty: z.ZodDefault<z.ZodNumber>;
         parallelSlots: z.ZodDefault<z.ZodNumber>;
+        /**
+         * How many chats' KV state llama-server may park in system RAM when they
+         * lose their GPU slot, so returning to one costs a bulk copy instead of a
+         * full re-prefill. Stored as a count, not a size: the byte budget it turns
+         * into (`--cache-ram`) depends on the measured KV bytes/token and the
+         * per-slot context, both of which move when other fields are edited.
+         * 0 means "leave llama.cpp's own default alone" - the flag is not emitted.
+         */
+        cachedChats: z.ZodDefault<z.ZodNumber>;
         /** RoPE extension factor; 1 keeps the GGUF-native context window. */
         contextMultiplier: z.ZodDefault<z.ZodNumber>;
         /** Cleared only by a successful calibration of this saved model profile. */
@@ -539,7 +711,31 @@ export declare const ProfilesStoreSchema: z.ZodObject<{
         vision: z.ZodDefault<z.ZodBoolean>;
         reasoningBudget: z.ZodDefault<z.ZodNumber>;
         reasoningBudgetMessage: z.ZodDefault<z.ZodString>;
+        /**
+         * Tri-state, matching llama-server's own `--reasoning-preserve` /
+         * `--no-reasoning-preserve` / unset: `true` keeps the reasoning trace in the
+         * whole history, `false` forces it to the last assistant message only, and
+         * null/undefined leaves the template's own default alone. Null is the
+         * default precisely so an untouched profile emits no flag and launches
+         * exactly as it did before this field existed.
+         */
+        preserveReasoning: z.ZodOptional<z.ZodNullable<z.ZodBoolean>>;
+        temperature: z.ZodDefault<z.ZodNumber>;
+        topP: z.ZodDefault<z.ZodNumber>;
+        topK: z.ZodDefault<z.ZodNumber>;
+        minP: z.ZodDefault<z.ZodNumber>;
+        presencePenalty: z.ZodDefault<z.ZodNumber>;
+        repeatPenalty: z.ZodDefault<z.ZodNumber>;
         parallelSlots: z.ZodDefault<z.ZodNumber>;
+        /**
+         * How many chats' KV state llama-server may park in system RAM when they
+         * lose their GPU slot, so returning to one costs a bulk copy instead of a
+         * full re-prefill. Stored as a count, not a size: the byte budget it turns
+         * into (`--cache-ram`) depends on the measured KV bytes/token and the
+         * per-slot context, both of which move when other fields are edited.
+         * 0 means "leave llama.cpp's own default alone" - the flag is not emitted.
+         */
+        cachedChats: z.ZodDefault<z.ZodNumber>;
         /** RoPE extension factor; 1 keeps the GGUF-native context window. */
         contextMultiplier: z.ZodDefault<z.ZodNumber>;
         /** Cleared only by a successful calibration of this saved model profile. */
@@ -585,7 +781,31 @@ export declare const ProfilesStoreSchema: z.ZodObject<{
         vision: z.ZodDefault<z.ZodBoolean>;
         reasoningBudget: z.ZodDefault<z.ZodNumber>;
         reasoningBudgetMessage: z.ZodDefault<z.ZodString>;
+        /**
+         * Tri-state, matching llama-server's own `--reasoning-preserve` /
+         * `--no-reasoning-preserve` / unset: `true` keeps the reasoning trace in the
+         * whole history, `false` forces it to the last assistant message only, and
+         * null/undefined leaves the template's own default alone. Null is the
+         * default precisely so an untouched profile emits no flag and launches
+         * exactly as it did before this field existed.
+         */
+        preserveReasoning: z.ZodOptional<z.ZodNullable<z.ZodBoolean>>;
+        temperature: z.ZodDefault<z.ZodNumber>;
+        topP: z.ZodDefault<z.ZodNumber>;
+        topK: z.ZodDefault<z.ZodNumber>;
+        minP: z.ZodDefault<z.ZodNumber>;
+        presencePenalty: z.ZodDefault<z.ZodNumber>;
+        repeatPenalty: z.ZodDefault<z.ZodNumber>;
         parallelSlots: z.ZodDefault<z.ZodNumber>;
+        /**
+         * How many chats' KV state llama-server may park in system RAM when they
+         * lose their GPU slot, so returning to one costs a bulk copy instead of a
+         * full re-prefill. Stored as a count, not a size: the byte budget it turns
+         * into (`--cache-ram`) depends on the measured KV bytes/token and the
+         * per-slot context, both of which move when other fields are edited.
+         * 0 means "leave llama.cpp's own default alone" - the flag is not emitted.
+         */
+        cachedChats: z.ZodDefault<z.ZodNumber>;
         /** RoPE extension factor; 1 keeps the GGUF-native context window. */
         contextMultiplier: z.ZodDefault<z.ZodNumber>;
         /** Cleared only by a successful calibration of this saved model profile. */
@@ -631,7 +851,31 @@ export declare const ProfilesStoreSchema: z.ZodObject<{
         vision: z.ZodDefault<z.ZodBoolean>;
         reasoningBudget: z.ZodDefault<z.ZodNumber>;
         reasoningBudgetMessage: z.ZodDefault<z.ZodString>;
+        /**
+         * Tri-state, matching llama-server's own `--reasoning-preserve` /
+         * `--no-reasoning-preserve` / unset: `true` keeps the reasoning trace in the
+         * whole history, `false` forces it to the last assistant message only, and
+         * null/undefined leaves the template's own default alone. Null is the
+         * default precisely so an untouched profile emits no flag and launches
+         * exactly as it did before this field existed.
+         */
+        preserveReasoning: z.ZodOptional<z.ZodNullable<z.ZodBoolean>>;
+        temperature: z.ZodDefault<z.ZodNumber>;
+        topP: z.ZodDefault<z.ZodNumber>;
+        topK: z.ZodDefault<z.ZodNumber>;
+        minP: z.ZodDefault<z.ZodNumber>;
+        presencePenalty: z.ZodDefault<z.ZodNumber>;
+        repeatPenalty: z.ZodDefault<z.ZodNumber>;
         parallelSlots: z.ZodDefault<z.ZodNumber>;
+        /**
+         * How many chats' KV state llama-server may park in system RAM when they
+         * lose their GPU slot, so returning to one costs a bulk copy instead of a
+         * full re-prefill. Stored as a count, not a size: the byte budget it turns
+         * into (`--cache-ram`) depends on the measured KV bytes/token and the
+         * per-slot context, both of which move when other fields are edited.
+         * 0 means "leave llama.cpp's own default alone" - the flag is not emitted.
+         */
+        cachedChats: z.ZodDefault<z.ZodNumber>;
         /** RoPE extension factor; 1 keeps the GGUF-native context window. */
         contextMultiplier: z.ZodDefault<z.ZodNumber>;
         /** Cleared only by a successful calibration of this saved model profile. */
@@ -791,7 +1035,31 @@ export declare const ProfilesStoreSchema: z.ZodObject<{
         vision: z.ZodDefault<z.ZodBoolean>;
         reasoningBudget: z.ZodDefault<z.ZodNumber>;
         reasoningBudgetMessage: z.ZodDefault<z.ZodString>;
+        /**
+         * Tri-state, matching llama-server's own `--reasoning-preserve` /
+         * `--no-reasoning-preserve` / unset: `true` keeps the reasoning trace in the
+         * whole history, `false` forces it to the last assistant message only, and
+         * null/undefined leaves the template's own default alone. Null is the
+         * default precisely so an untouched profile emits no flag and launches
+         * exactly as it did before this field existed.
+         */
+        preserveReasoning: z.ZodOptional<z.ZodNullable<z.ZodBoolean>>;
+        temperature: z.ZodDefault<z.ZodNumber>;
+        topP: z.ZodDefault<z.ZodNumber>;
+        topK: z.ZodDefault<z.ZodNumber>;
+        minP: z.ZodDefault<z.ZodNumber>;
+        presencePenalty: z.ZodDefault<z.ZodNumber>;
+        repeatPenalty: z.ZodDefault<z.ZodNumber>;
         parallelSlots: z.ZodDefault<z.ZodNumber>;
+        /**
+         * How many chats' KV state llama-server may park in system RAM when they
+         * lose their GPU slot, so returning to one costs a bulk copy instead of a
+         * full re-prefill. Stored as a count, not a size: the byte budget it turns
+         * into (`--cache-ram`) depends on the measured KV bytes/token and the
+         * per-slot context, both of which move when other fields are edited.
+         * 0 means "leave llama.cpp's own default alone" - the flag is not emitted.
+         */
+        cachedChats: z.ZodDefault<z.ZodNumber>;
         /** RoPE extension factor; 1 keeps the GGUF-native context window. */
         contextMultiplier: z.ZodDefault<z.ZodNumber>;
         /** Cleared only by a successful calibration of this saved model profile. */
@@ -837,7 +1105,31 @@ export declare const ProfilesStoreSchema: z.ZodObject<{
         vision: z.ZodDefault<z.ZodBoolean>;
         reasoningBudget: z.ZodDefault<z.ZodNumber>;
         reasoningBudgetMessage: z.ZodDefault<z.ZodString>;
+        /**
+         * Tri-state, matching llama-server's own `--reasoning-preserve` /
+         * `--no-reasoning-preserve` / unset: `true` keeps the reasoning trace in the
+         * whole history, `false` forces it to the last assistant message only, and
+         * null/undefined leaves the template's own default alone. Null is the
+         * default precisely so an untouched profile emits no flag and launches
+         * exactly as it did before this field existed.
+         */
+        preserveReasoning: z.ZodOptional<z.ZodNullable<z.ZodBoolean>>;
+        temperature: z.ZodDefault<z.ZodNumber>;
+        topP: z.ZodDefault<z.ZodNumber>;
+        topK: z.ZodDefault<z.ZodNumber>;
+        minP: z.ZodDefault<z.ZodNumber>;
+        presencePenalty: z.ZodDefault<z.ZodNumber>;
+        repeatPenalty: z.ZodDefault<z.ZodNumber>;
         parallelSlots: z.ZodDefault<z.ZodNumber>;
+        /**
+         * How many chats' KV state llama-server may park in system RAM when they
+         * lose their GPU slot, so returning to one costs a bulk copy instead of a
+         * full re-prefill. Stored as a count, not a size: the byte budget it turns
+         * into (`--cache-ram`) depends on the measured KV bytes/token and the
+         * per-slot context, both of which move when other fields are edited.
+         * 0 means "leave llama.cpp's own default alone" - the flag is not emitted.
+         */
+        cachedChats: z.ZodDefault<z.ZodNumber>;
         /** RoPE extension factor; 1 keeps the GGUF-native context window. */
         contextMultiplier: z.ZodDefault<z.ZodNumber>;
         /** Cleared only by a successful calibration of this saved model profile. */
@@ -883,7 +1175,31 @@ export declare const ProfilesStoreSchema: z.ZodObject<{
         vision: z.ZodDefault<z.ZodBoolean>;
         reasoningBudget: z.ZodDefault<z.ZodNumber>;
         reasoningBudgetMessage: z.ZodDefault<z.ZodString>;
+        /**
+         * Tri-state, matching llama-server's own `--reasoning-preserve` /
+         * `--no-reasoning-preserve` / unset: `true` keeps the reasoning trace in the
+         * whole history, `false` forces it to the last assistant message only, and
+         * null/undefined leaves the template's own default alone. Null is the
+         * default precisely so an untouched profile emits no flag and launches
+         * exactly as it did before this field existed.
+         */
+        preserveReasoning: z.ZodOptional<z.ZodNullable<z.ZodBoolean>>;
+        temperature: z.ZodDefault<z.ZodNumber>;
+        topP: z.ZodDefault<z.ZodNumber>;
+        topK: z.ZodDefault<z.ZodNumber>;
+        minP: z.ZodDefault<z.ZodNumber>;
+        presencePenalty: z.ZodDefault<z.ZodNumber>;
+        repeatPenalty: z.ZodDefault<z.ZodNumber>;
         parallelSlots: z.ZodDefault<z.ZodNumber>;
+        /**
+         * How many chats' KV state llama-server may park in system RAM when they
+         * lose their GPU slot, so returning to one costs a bulk copy instead of a
+         * full re-prefill. Stored as a count, not a size: the byte budget it turns
+         * into (`--cache-ram`) depends on the measured KV bytes/token and the
+         * per-slot context, both of which move when other fields are edited.
+         * 0 means "leave llama.cpp's own default alone" - the flag is not emitted.
+         */
+        cachedChats: z.ZodDefault<z.ZodNumber>;
         /** RoPE extension factor; 1 keeps the GGUF-native context window. */
         contextMultiplier: z.ZodDefault<z.ZodNumber>;
         /** Cleared only by a successful calibration of this saved model profile. */
@@ -1096,12 +1412,16 @@ export type Tls = z.infer<typeof TlsSchema>;
 export declare const RuntimeConfigSchema: z.ZodObject<{
     source: z.ZodDefault<z.ZodEnum<["auto", "managed", "lmstudio"]>>;
     path: z.ZodDefault<z.ZodNullable<z.ZodString>>;
+    /** llama.cpp's `--log-verbosity`: 0 generic output through 5 debug. */
+    logVerbosity: z.ZodDefault<z.ZodNumber>;
 }, "strict", z.ZodTypeAny, {
     path: string | null;
     source: "auto" | "managed" | "lmstudio";
+    logVerbosity: number;
 }, {
     path?: string | null | undefined;
     source?: "auto" | "managed" | "lmstudio" | undefined;
+    logVerbosity?: number | undefined;
 }>;
 export type RuntimeConfig = z.infer<typeof RuntimeConfigSchema>;
 export declare const ProfileDefaultsSchema: z.ZodObject<{
@@ -1182,16 +1502,24 @@ export declare const BrainConfigSchema: z.ZodObject<{
     runtime: z.ZodDefault<z.ZodObject<{
         source: z.ZodDefault<z.ZodEnum<["auto", "managed", "lmstudio"]>>;
         path: z.ZodDefault<z.ZodNullable<z.ZodString>>;
+        /** llama.cpp's `--log-verbosity`: 0 generic output through 5 debug. */
+        logVerbosity: z.ZodDefault<z.ZodNumber>;
     }, "strict", z.ZodTypeAny, {
         path: string | null;
         source: "auto" | "managed" | "lmstudio";
+        logVerbosity: number;
     }, {
         path?: string | null | undefined;
         source?: "auto" | "managed" | "lmstudio" | undefined;
+        logVerbosity?: number | undefined;
     }>>;
     modelsDir: z.ZodDefault<z.ZodNullable<z.ZodString>>;
     hfToken: z.ZodDefault<z.ZodNullable<z.ZodString>>;
     defaultModel: z.ZodDefault<z.ZodNullable<z.ZodString>>;
+    /** Maximum independently hosted llama-server model processes. */
+    maxLoadedModels: z.ZodDefault<z.ZodNumber>;
+    /** Stable ids selected for residency while model locking is enabled. */
+    lockedModels: z.ZodDefault<z.ZodArray<z.ZodString, "many">>;
     lockModel: z.ZodDefault<z.ZodBoolean>;
     allowRemoteConfig: z.ZodDefault<z.ZodBoolean>;
     allowInsecureBind: z.ZodDefault<z.ZodBoolean>;
@@ -1242,10 +1570,13 @@ export declare const BrainConfigSchema: z.ZodObject<{
     runtime: {
         path: string | null;
         source: "auto" | "managed" | "lmstudio";
+        logVerbosity: number;
     };
     modelsDir: string | null;
     hfToken: string | null;
     defaultModel: string | null;
+    maxLoadedModels: number;
+    lockedModels: string[];
     lockModel: boolean;
     allowRemoteConfig: boolean;
     allowInsecureBind: boolean;
@@ -1282,10 +1613,13 @@ export declare const BrainConfigSchema: z.ZodObject<{
     runtime?: {
         path?: string | null | undefined;
         source?: "auto" | "managed" | "lmstudio" | undefined;
+        logVerbosity?: number | undefined;
     } | undefined;
     modelsDir?: string | null | undefined;
     hfToken?: string | null | undefined;
     defaultModel?: string | null | undefined;
+    maxLoadedModels?: number | undefined;
+    lockedModels?: string[] | undefined;
     lockModel?: boolean | undefined;
     allowRemoteConfig?: boolean | undefined;
     allowInsecureBind?: boolean | undefined;
@@ -1307,6 +1641,8 @@ export declare const CatalogModelSchema: z.ZodObject<{
     name: z.ZodString;
     /** Stable UI family identity. Otto clients resolve this to a monochrome glyph. */
     family: z.ZodOptional<z.ZodString>;
+    /** Otto-curated favorite shown as a gold premium badge in the Brain Library. */
+    favorite: z.ZodDefault<z.ZodBoolean>;
     publisher: z.ZodOptional<z.ZodString>;
     hfRepo: z.ZodString;
     quant: z.ZodString;
@@ -1317,6 +1653,27 @@ export declare const CatalogModelSchema: z.ZodObject<{
     vision: z.ZodOptional<z.ZodBoolean>;
     thinking: z.ZodOptional<z.ZodBoolean>;
     reasoningEfforts: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+    reasoningEffortDefault: z.ZodOptional<z.ZodString>;
+    reasoningTemplate: z.ZodOptional<z.ZodObject<{
+        enableThinkingArgument: z.ZodString;
+        effortArgument: z.ZodString;
+    }, "strict", z.ZodTypeAny, {
+        enableThinkingArgument: string;
+        effortArgument: string;
+    }, {
+        enableThinkingArgument: string;
+        effortArgument: string;
+    }>>;
+    reasoningPreservation: z.ZodOptional<z.ZodObject<{
+        templateArgument: z.ZodString;
+        default: z.ZodOptional<z.ZodBoolean>;
+    }, "strict", z.ZodTypeAny, {
+        templateArgument: string;
+        default?: boolean | undefined;
+    }, {
+        templateArgument: string;
+        default?: boolean | undefined;
+    }>>;
     contextMax: z.ZodOptional<z.ZodNumber>;
     useCases: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
     tier: z.ZodOptional<z.ZodString>;
@@ -1367,6 +1724,8 @@ export declare const CatalogModelSchema: z.ZodObject<{
     name: z.ZodString;
     /** Stable UI family identity. Otto clients resolve this to a monochrome glyph. */
     family: z.ZodOptional<z.ZodString>;
+    /** Otto-curated favorite shown as a gold premium badge in the Brain Library. */
+    favorite: z.ZodDefault<z.ZodBoolean>;
     publisher: z.ZodOptional<z.ZodString>;
     hfRepo: z.ZodString;
     quant: z.ZodString;
@@ -1377,6 +1736,27 @@ export declare const CatalogModelSchema: z.ZodObject<{
     vision: z.ZodOptional<z.ZodBoolean>;
     thinking: z.ZodOptional<z.ZodBoolean>;
     reasoningEfforts: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+    reasoningEffortDefault: z.ZodOptional<z.ZodString>;
+    reasoningTemplate: z.ZodOptional<z.ZodObject<{
+        enableThinkingArgument: z.ZodString;
+        effortArgument: z.ZodString;
+    }, "strict", z.ZodTypeAny, {
+        enableThinkingArgument: string;
+        effortArgument: string;
+    }, {
+        enableThinkingArgument: string;
+        effortArgument: string;
+    }>>;
+    reasoningPreservation: z.ZodOptional<z.ZodObject<{
+        templateArgument: z.ZodString;
+        default: z.ZodOptional<z.ZodBoolean>;
+    }, "strict", z.ZodTypeAny, {
+        templateArgument: string;
+        default?: boolean | undefined;
+    }, {
+        templateArgument: string;
+        default?: boolean | undefined;
+    }>>;
     contextMax: z.ZodOptional<z.ZodNumber>;
     useCases: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
     tier: z.ZodOptional<z.ZodString>;
@@ -1427,6 +1807,8 @@ export declare const CatalogModelSchema: z.ZodObject<{
     name: z.ZodString;
     /** Stable UI family identity. Otto clients resolve this to a monochrome glyph. */
     family: z.ZodOptional<z.ZodString>;
+    /** Otto-curated favorite shown as a gold premium badge in the Brain Library. */
+    favorite: z.ZodDefault<z.ZodBoolean>;
     publisher: z.ZodOptional<z.ZodString>;
     hfRepo: z.ZodString;
     quant: z.ZodString;
@@ -1437,6 +1819,27 @@ export declare const CatalogModelSchema: z.ZodObject<{
     vision: z.ZodOptional<z.ZodBoolean>;
     thinking: z.ZodOptional<z.ZodBoolean>;
     reasoningEfforts: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+    reasoningEffortDefault: z.ZodOptional<z.ZodString>;
+    reasoningTemplate: z.ZodOptional<z.ZodObject<{
+        enableThinkingArgument: z.ZodString;
+        effortArgument: z.ZodString;
+    }, "strict", z.ZodTypeAny, {
+        enableThinkingArgument: string;
+        effortArgument: string;
+    }, {
+        enableThinkingArgument: string;
+        effortArgument: string;
+    }>>;
+    reasoningPreservation: z.ZodOptional<z.ZodObject<{
+        templateArgument: z.ZodString;
+        default: z.ZodOptional<z.ZodBoolean>;
+    }, "strict", z.ZodTypeAny, {
+        templateArgument: string;
+        default?: boolean | undefined;
+    }, {
+        templateArgument: string;
+        default?: boolean | undefined;
+    }>>;
     contextMax: z.ZodOptional<z.ZodNumber>;
     useCases: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
     tier: z.ZodOptional<z.ZodString>;
@@ -1487,6 +1890,8 @@ export declare const CatalogSchema: z.ZodObject<{
     note: z.ZodOptional<z.ZodString>;
     vramBudgetBytes: z.ZodOptional<z.ZodNumber>;
     systemRamBytes: z.ZodOptional<z.ZodNumber>;
+    /** Retired curated ids that no longer belong to any canonical catalog entry. */
+    retiredModelIds: z.ZodDefault<z.ZodArray<z.ZodString, "many">>;
     models: z.ZodDefault<z.ZodArray<z.ZodObject<{
         id: z.ZodString;
         /** Retired Otto-curated ids this canonical catalog entry replaces. */
@@ -1494,6 +1899,8 @@ export declare const CatalogSchema: z.ZodObject<{
         name: z.ZodString;
         /** Stable UI family identity. Otto clients resolve this to a monochrome glyph. */
         family: z.ZodOptional<z.ZodString>;
+        /** Otto-curated favorite shown as a gold premium badge in the Brain Library. */
+        favorite: z.ZodDefault<z.ZodBoolean>;
         publisher: z.ZodOptional<z.ZodString>;
         hfRepo: z.ZodString;
         quant: z.ZodString;
@@ -1504,6 +1911,27 @@ export declare const CatalogSchema: z.ZodObject<{
         vision: z.ZodOptional<z.ZodBoolean>;
         thinking: z.ZodOptional<z.ZodBoolean>;
         reasoningEfforts: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+        reasoningEffortDefault: z.ZodOptional<z.ZodString>;
+        reasoningTemplate: z.ZodOptional<z.ZodObject<{
+            enableThinkingArgument: z.ZodString;
+            effortArgument: z.ZodString;
+        }, "strict", z.ZodTypeAny, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }>>;
+        reasoningPreservation: z.ZodOptional<z.ZodObject<{
+            templateArgument: z.ZodString;
+            default: z.ZodOptional<z.ZodBoolean>;
+        }, "strict", z.ZodTypeAny, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }>>;
         contextMax: z.ZodOptional<z.ZodNumber>;
         useCases: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
         tier: z.ZodOptional<z.ZodString>;
@@ -1554,6 +1982,8 @@ export declare const CatalogSchema: z.ZodObject<{
         name: z.ZodString;
         /** Stable UI family identity. Otto clients resolve this to a monochrome glyph. */
         family: z.ZodOptional<z.ZodString>;
+        /** Otto-curated favorite shown as a gold premium badge in the Brain Library. */
+        favorite: z.ZodDefault<z.ZodBoolean>;
         publisher: z.ZodOptional<z.ZodString>;
         hfRepo: z.ZodString;
         quant: z.ZodString;
@@ -1564,6 +1994,27 @@ export declare const CatalogSchema: z.ZodObject<{
         vision: z.ZodOptional<z.ZodBoolean>;
         thinking: z.ZodOptional<z.ZodBoolean>;
         reasoningEfforts: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+        reasoningEffortDefault: z.ZodOptional<z.ZodString>;
+        reasoningTemplate: z.ZodOptional<z.ZodObject<{
+            enableThinkingArgument: z.ZodString;
+            effortArgument: z.ZodString;
+        }, "strict", z.ZodTypeAny, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }>>;
+        reasoningPreservation: z.ZodOptional<z.ZodObject<{
+            templateArgument: z.ZodString;
+            default: z.ZodOptional<z.ZodBoolean>;
+        }, "strict", z.ZodTypeAny, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }>>;
         contextMax: z.ZodOptional<z.ZodNumber>;
         useCases: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
         tier: z.ZodOptional<z.ZodString>;
@@ -1614,6 +2065,8 @@ export declare const CatalogSchema: z.ZodObject<{
         name: z.ZodString;
         /** Stable UI family identity. Otto clients resolve this to a monochrome glyph. */
         family: z.ZodOptional<z.ZodString>;
+        /** Otto-curated favorite shown as a gold premium badge in the Brain Library. */
+        favorite: z.ZodDefault<z.ZodBoolean>;
         publisher: z.ZodOptional<z.ZodString>;
         hfRepo: z.ZodString;
         quant: z.ZodString;
@@ -1624,6 +2077,27 @@ export declare const CatalogSchema: z.ZodObject<{
         vision: z.ZodOptional<z.ZodBoolean>;
         thinking: z.ZodOptional<z.ZodBoolean>;
         reasoningEfforts: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+        reasoningEffortDefault: z.ZodOptional<z.ZodString>;
+        reasoningTemplate: z.ZodOptional<z.ZodObject<{
+            enableThinkingArgument: z.ZodString;
+            effortArgument: z.ZodString;
+        }, "strict", z.ZodTypeAny, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }>>;
+        reasoningPreservation: z.ZodOptional<z.ZodObject<{
+            templateArgument: z.ZodString;
+            default: z.ZodOptional<z.ZodBoolean>;
+        }, "strict", z.ZodTypeAny, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }>>;
         contextMax: z.ZodOptional<z.ZodNumber>;
         useCases: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
         tier: z.ZodOptional<z.ZodString>;
@@ -1673,6 +2147,8 @@ export declare const CatalogSchema: z.ZodObject<{
     note: z.ZodOptional<z.ZodString>;
     vramBudgetBytes: z.ZodOptional<z.ZodNumber>;
     systemRamBytes: z.ZodOptional<z.ZodNumber>;
+    /** Retired curated ids that no longer belong to any canonical catalog entry. */
+    retiredModelIds: z.ZodDefault<z.ZodArray<z.ZodString, "many">>;
     models: z.ZodDefault<z.ZodArray<z.ZodObject<{
         id: z.ZodString;
         /** Retired Otto-curated ids this canonical catalog entry replaces. */
@@ -1680,6 +2156,8 @@ export declare const CatalogSchema: z.ZodObject<{
         name: z.ZodString;
         /** Stable UI family identity. Otto clients resolve this to a monochrome glyph. */
         family: z.ZodOptional<z.ZodString>;
+        /** Otto-curated favorite shown as a gold premium badge in the Brain Library. */
+        favorite: z.ZodDefault<z.ZodBoolean>;
         publisher: z.ZodOptional<z.ZodString>;
         hfRepo: z.ZodString;
         quant: z.ZodString;
@@ -1690,6 +2168,27 @@ export declare const CatalogSchema: z.ZodObject<{
         vision: z.ZodOptional<z.ZodBoolean>;
         thinking: z.ZodOptional<z.ZodBoolean>;
         reasoningEfforts: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+        reasoningEffortDefault: z.ZodOptional<z.ZodString>;
+        reasoningTemplate: z.ZodOptional<z.ZodObject<{
+            enableThinkingArgument: z.ZodString;
+            effortArgument: z.ZodString;
+        }, "strict", z.ZodTypeAny, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }>>;
+        reasoningPreservation: z.ZodOptional<z.ZodObject<{
+            templateArgument: z.ZodString;
+            default: z.ZodOptional<z.ZodBoolean>;
+        }, "strict", z.ZodTypeAny, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }>>;
         contextMax: z.ZodOptional<z.ZodNumber>;
         useCases: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
         tier: z.ZodOptional<z.ZodString>;
@@ -1740,6 +2239,8 @@ export declare const CatalogSchema: z.ZodObject<{
         name: z.ZodString;
         /** Stable UI family identity. Otto clients resolve this to a monochrome glyph. */
         family: z.ZodOptional<z.ZodString>;
+        /** Otto-curated favorite shown as a gold premium badge in the Brain Library. */
+        favorite: z.ZodDefault<z.ZodBoolean>;
         publisher: z.ZodOptional<z.ZodString>;
         hfRepo: z.ZodString;
         quant: z.ZodString;
@@ -1750,6 +2251,27 @@ export declare const CatalogSchema: z.ZodObject<{
         vision: z.ZodOptional<z.ZodBoolean>;
         thinking: z.ZodOptional<z.ZodBoolean>;
         reasoningEfforts: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+        reasoningEffortDefault: z.ZodOptional<z.ZodString>;
+        reasoningTemplate: z.ZodOptional<z.ZodObject<{
+            enableThinkingArgument: z.ZodString;
+            effortArgument: z.ZodString;
+        }, "strict", z.ZodTypeAny, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }>>;
+        reasoningPreservation: z.ZodOptional<z.ZodObject<{
+            templateArgument: z.ZodString;
+            default: z.ZodOptional<z.ZodBoolean>;
+        }, "strict", z.ZodTypeAny, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }>>;
         contextMax: z.ZodOptional<z.ZodNumber>;
         useCases: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
         tier: z.ZodOptional<z.ZodString>;
@@ -1800,6 +2322,8 @@ export declare const CatalogSchema: z.ZodObject<{
         name: z.ZodString;
         /** Stable UI family identity. Otto clients resolve this to a monochrome glyph. */
         family: z.ZodOptional<z.ZodString>;
+        /** Otto-curated favorite shown as a gold premium badge in the Brain Library. */
+        favorite: z.ZodDefault<z.ZodBoolean>;
         publisher: z.ZodOptional<z.ZodString>;
         hfRepo: z.ZodString;
         quant: z.ZodString;
@@ -1810,6 +2334,27 @@ export declare const CatalogSchema: z.ZodObject<{
         vision: z.ZodOptional<z.ZodBoolean>;
         thinking: z.ZodOptional<z.ZodBoolean>;
         reasoningEfforts: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+        reasoningEffortDefault: z.ZodOptional<z.ZodString>;
+        reasoningTemplate: z.ZodOptional<z.ZodObject<{
+            enableThinkingArgument: z.ZodString;
+            effortArgument: z.ZodString;
+        }, "strict", z.ZodTypeAny, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }>>;
+        reasoningPreservation: z.ZodOptional<z.ZodObject<{
+            templateArgument: z.ZodString;
+            default: z.ZodOptional<z.ZodBoolean>;
+        }, "strict", z.ZodTypeAny, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }>>;
         contextMax: z.ZodOptional<z.ZodNumber>;
         useCases: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
         tier: z.ZodOptional<z.ZodString>;
@@ -1859,6 +2404,8 @@ export declare const CatalogSchema: z.ZodObject<{
     note: z.ZodOptional<z.ZodString>;
     vramBudgetBytes: z.ZodOptional<z.ZodNumber>;
     systemRamBytes: z.ZodOptional<z.ZodNumber>;
+    /** Retired curated ids that no longer belong to any canonical catalog entry. */
+    retiredModelIds: z.ZodDefault<z.ZodArray<z.ZodString, "many">>;
     models: z.ZodDefault<z.ZodArray<z.ZodObject<{
         id: z.ZodString;
         /** Retired Otto-curated ids this canonical catalog entry replaces. */
@@ -1866,6 +2413,8 @@ export declare const CatalogSchema: z.ZodObject<{
         name: z.ZodString;
         /** Stable UI family identity. Otto clients resolve this to a monochrome glyph. */
         family: z.ZodOptional<z.ZodString>;
+        /** Otto-curated favorite shown as a gold premium badge in the Brain Library. */
+        favorite: z.ZodDefault<z.ZodBoolean>;
         publisher: z.ZodOptional<z.ZodString>;
         hfRepo: z.ZodString;
         quant: z.ZodString;
@@ -1876,6 +2425,27 @@ export declare const CatalogSchema: z.ZodObject<{
         vision: z.ZodOptional<z.ZodBoolean>;
         thinking: z.ZodOptional<z.ZodBoolean>;
         reasoningEfforts: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+        reasoningEffortDefault: z.ZodOptional<z.ZodString>;
+        reasoningTemplate: z.ZodOptional<z.ZodObject<{
+            enableThinkingArgument: z.ZodString;
+            effortArgument: z.ZodString;
+        }, "strict", z.ZodTypeAny, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }>>;
+        reasoningPreservation: z.ZodOptional<z.ZodObject<{
+            templateArgument: z.ZodString;
+            default: z.ZodOptional<z.ZodBoolean>;
+        }, "strict", z.ZodTypeAny, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }>>;
         contextMax: z.ZodOptional<z.ZodNumber>;
         useCases: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
         tier: z.ZodOptional<z.ZodString>;
@@ -1926,6 +2496,8 @@ export declare const CatalogSchema: z.ZodObject<{
         name: z.ZodString;
         /** Stable UI family identity. Otto clients resolve this to a monochrome glyph. */
         family: z.ZodOptional<z.ZodString>;
+        /** Otto-curated favorite shown as a gold premium badge in the Brain Library. */
+        favorite: z.ZodDefault<z.ZodBoolean>;
         publisher: z.ZodOptional<z.ZodString>;
         hfRepo: z.ZodString;
         quant: z.ZodString;
@@ -1936,6 +2508,27 @@ export declare const CatalogSchema: z.ZodObject<{
         vision: z.ZodOptional<z.ZodBoolean>;
         thinking: z.ZodOptional<z.ZodBoolean>;
         reasoningEfforts: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+        reasoningEffortDefault: z.ZodOptional<z.ZodString>;
+        reasoningTemplate: z.ZodOptional<z.ZodObject<{
+            enableThinkingArgument: z.ZodString;
+            effortArgument: z.ZodString;
+        }, "strict", z.ZodTypeAny, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }>>;
+        reasoningPreservation: z.ZodOptional<z.ZodObject<{
+            templateArgument: z.ZodString;
+            default: z.ZodOptional<z.ZodBoolean>;
+        }, "strict", z.ZodTypeAny, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }>>;
         contextMax: z.ZodOptional<z.ZodNumber>;
         useCases: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
         tier: z.ZodOptional<z.ZodString>;
@@ -1986,6 +2579,8 @@ export declare const CatalogSchema: z.ZodObject<{
         name: z.ZodString;
         /** Stable UI family identity. Otto clients resolve this to a monochrome glyph. */
         family: z.ZodOptional<z.ZodString>;
+        /** Otto-curated favorite shown as a gold premium badge in the Brain Library. */
+        favorite: z.ZodDefault<z.ZodBoolean>;
         publisher: z.ZodOptional<z.ZodString>;
         hfRepo: z.ZodString;
         quant: z.ZodString;
@@ -1996,6 +2591,27 @@ export declare const CatalogSchema: z.ZodObject<{
         vision: z.ZodOptional<z.ZodBoolean>;
         thinking: z.ZodOptional<z.ZodBoolean>;
         reasoningEfforts: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+        reasoningEffortDefault: z.ZodOptional<z.ZodString>;
+        reasoningTemplate: z.ZodOptional<z.ZodObject<{
+            enableThinkingArgument: z.ZodString;
+            effortArgument: z.ZodString;
+        }, "strict", z.ZodTypeAny, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }, {
+            enableThinkingArgument: string;
+            effortArgument: string;
+        }>>;
+        reasoningPreservation: z.ZodOptional<z.ZodObject<{
+            templateArgument: z.ZodString;
+            default: z.ZodOptional<z.ZodBoolean>;
+        }, "strict", z.ZodTypeAny, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }, {
+            templateArgument: string;
+            default?: boolean | undefined;
+        }>>;
         contextMax: z.ZodOptional<z.ZodNumber>;
         useCases: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
         tier: z.ZodOptional<z.ZodString>;