@otto-code/protocol 0.8.1 → 0.8.3

package/dist/messages.js

@@ -1544,6 +1544,21 @@ export const BrainCapabilitiesSchema = z
     load: z.boolean().default(false),
     resources: z.boolean().default(false),
     inventory: z.boolean().default(false),
+    /**
+     * GET /__host/events: a live SSE stream of complete status snapshots.
+     *
+     * The daemon reads this before deciding how to watch the brain. False (the
+     * default, and what every brain built before the stream reports) keeps the
+     * daemon on the ordinary status poll, which is why nothing else about the
+     * management API had to change for pushed status to ship.
+     */
+    events: z.boolean().default(false),
+    /**
+     * Status events include bounded live inference stages, token counts and
+     * throughput. False for the first event-stream generation, whose snapshots
+     * only moved at phase boundaries.
+     */
+    liveInference: z.boolean().default(false),
     /** Whether writes are permitted right now (the brain's allowRemoteConfig). */
     writable: z.boolean().default(false),
 })
@@ -1556,6 +1571,14 @@ export const BrainHostStatusSchema = z
     running: z.boolean(),
     pid: z.number().nullable().optional(),
     version: z.string().nullable().optional(),
+    /**
+     * Which generation of the brain's management contract the far side speaks.
+     *
+     * Additive and separate from `version`, which is the package build. A daemon
+     * reads this plus `capabilities` instead of pinning an exact brain version;
+     * absent means a brain from before the field existed.
+     */
+    apiVersion: z.number().nullable().optional(),
     host: z.string().nullable().optional(),
     port: z.number().nullable().optional(),
     displayHost: z.string().nullable().optional(),
@@ -1577,7 +1600,7 @@ export const BrainHostStatusSchema = z
      * configuration. Null from a brain that predates the management API.
      */
     capabilities: BrainCapabilitiesSchema.nullable().optional(),
-    /** Live CPU/RAM/GPU/slot telemetry, only when the request asked for it. */
+    /** Live CPU/RAM/GPU telemetry, only when the request asked for it. */
     resources: z.record(z.string(), z.unknown()).nullable().optional(),
     /** How many log lines the brain currently holds, for the Logs tab. */
     logLineCount: z.number().nullable().optional(),
@@ -1607,6 +1630,20 @@ export const BrainHostStatusSchema = z
      * because it is one boolean the router already knows.
      */
     reasoning: z.boolean().nullable().optional(),
+    /**
+     * Exact aggregate lifecycle counts for requests currently dispatched to
+     * llama-server. Additive in host API v2; absent on older brains.
+     */
+    inference: z
+        .object({
+        activeRequests: z.number().int().nonnegative().optional(),
+        processing: z.number().int().nonnegative().optional(),
+        thinking: z.number().int().nonnegative().optional(),
+        generating: z.number().int().nonnegative().optional(),
+    })
+        .passthrough()
+        .nullable()
+        .optional(),
     /**
      * Slot occupancy split by phase, so a client can tell a prompt being
      * ingested from a model that has started answering. Rides here rather than
@@ -1620,6 +1657,19 @@ export const BrainHostStatusSchema = z
         idle: z.number().nullable().optional(),
         prefill: z.number().nullable().optional(),
         decode: z.number().nullable().optional(),
+        /** Bounded-rate per-slot performance samples; host API v2 and newer. */
+        threads: z
+            .array(z
+            .object({
+            slot: z.number().optional(),
+            phase: z.enum(["prefill", "decode"]).optional(),
+            promptTokens: z.number().nullable().optional(),
+            generatedTokens: z.number().nullable().optional(),
+            promptTokensPerSecond: z.number().nullable().optional(),
+            tokensPerSecond: z.number().nullable().optional(),
+        })
+            .passthrough())
+            .optional(),
     })
         .passthrough()
         .nullable()
@@ -1638,9 +1688,9 @@ const BrainHostStatusResultSchema = z.object({
 export const BrainHostStatusRequestSchema = z.object({
     type: z.literal("brain.host.status.request"),
     /**
-     * Ask for live CPU/RAM/GPU/slot telemetry alongside the status. Off by default
-     * on purpose: it costs an `nvidia-smi` spawn plus a /slots round trip on the
-     * brain, and this RPC is also the liveness poll. Only a surface actually
+     * Ask for live CPU/RAM/GPU telemetry alongside the status. Off by default on
+     * purpose: it costs an `nvidia-smi` spawn on the brain, and this RPC is also
+     * the liveness poll. Only a surface actually
      * rendering the numbers should turn it on.
      */
     resources: z.boolean().default(false),
@@ -2321,6 +2371,39 @@ export const BrainModelDeleteResponseSchema = z.object({
         requestId: z.string(),
     }),
 });
+// Rename a model's display name. The brain rejects a collision with another
+// model's current id/displayName: /v1/models keys its id on displayName, and
+// both the completion path and defaultModel/switchTo resolve a model by
+// displayName === name || id === name - a duplicate would strand one model
+// unreachable by name with no error anywhere else in the chain.
+export const BrainModelRenameRequestSchema = z.object({
+    type: z.literal("brain.model.rename.request"),
+    modelId: z.string(),
+    displayName: z.string(),
+    requestId: z.string(),
+});
+export const BrainModelRenameResponseSchema = z.object({
+    type: z.literal("brain.model.rename.response"),
+    payload: z.object({
+        displayName: z.string().nullable().default(null),
+        error: z.string().nullable(),
+        requestId: z.string(),
+    }),
+});
+// Reset a model's display name back to its scan-derived default.
+export const BrainModelRenameResetRequestSchema = z.object({
+    type: z.literal("brain.model.rename.reset.request"),
+    modelId: z.string(),
+    requestId: z.string(),
+});
+export const BrainModelRenameResetResponseSchema = z.object({
+    type: z.literal("brain.model.rename.reset.response"),
+    payload: z.object({
+        displayName: z.string().nullable().default(null),
+        error: z.string().nullable(),
+        requestId: z.string(),
+    }),
+});
 // Tail the brain's llama-server log.
 export const BrainLogsTailRequestSchema = z.object({
     type: z.literal("brain.logs.tail.request"),
@@ -5857,6 +5940,8 @@ export const SessionInboundMessageSchema = z.discriminatedUnion("type", [
     BrainModelLoadRequestSchema,
     BrainModelUnloadRequestSchema,
     BrainModelDeleteRequestSchema,
+    BrainModelRenameRequestSchema,
+    BrainModelRenameResetRequestSchema,
     BrainLogsTailRequestSchema,
     UpdateAgentRequestMessageSchema,
     ProjectRenameRequestSchema,
@@ -6340,6 +6425,14 @@ export const ServerInfoStatusPayloadSchema = z
         // brain version independently.
         // COMPAT(brainConsole): added in v0.7.7, drop the gate when daemon floor >= v0.7.7.
         brainConsole: z.boolean().optional(),
+        // COMPAT(brainStatusPush): added in v0.8.3, drop the gate when daemon
+        // floor >= v0.8.3. The daemon subscribes to the brain's own SSE status
+        // stream and broadcasts `brain_status_changed`. Unlike the flags above
+        // this is not a fixed daemon capability: it is true only while the
+        // SELECTED brain also advertises `capabilities.events`, and the daemon
+        // re-broadcasts server_info when that changes. A client that sees it
+        // false keeps the explicit status poll.
+        brainStatusPush: z.boolean().optional(),
         // COMPAT(agentForkContext): added in v0.1.102, remove gate after 2026-12-28.
         agentForkContext: z.boolean().optional(),
         // COMPAT(providerRemove): added in v0.1.105, drop the gate when daemon floor >= v0.1.105.
@@ -6805,7 +6898,27 @@ export const LspDiagnosticsChangedStatusPayloadSchema = z
     diagnostics: z.array(CodeDiagnosticSchema),
 })
     .passthrough();
+/**
+ * The brain's own state, pushed the moment it changes.
+ *
+ * A complete cheap `BrainHostStatus` snapshot, never a delta - which is what
+ * makes a missed message and a reconnect the same, idempotent, recovery. It
+ * excludes `resources`, whose collection spawns `nvidia-smi` on the brain host
+ * and stays an opt-in pull for the Overview tab.
+ *
+ * Scoped by the daemon connection that delivers it: the client writes it under
+ * that runtime's `serverId`, so two connected hosts cannot overwrite each
+ * other's brain state. Gated by `server_info.features.brainStatusPush`, which
+ * is true only when both the daemon and the brain it reaches support the stream.
+ */
+export const BrainStatusChangedStatusPayloadSchema = z
+    .object({
+    status: z.literal("brain_status_changed"),
+    brain: BrainHostStatusSchema,
+})
+    .passthrough();
 export const KnownStatusPayloadSchema = z.discriminatedUnion("status", [
+    BrainStatusChangedStatusPayloadSchema,
     LspActivityChangedStatusPayloadSchema,
     LspDiagnosticsChangedStatusPayloadSchema,
     AgentCreatedStatusPayloadSchema,
@@ -10417,6 +10530,8 @@ export const SessionOutboundMessageSchema = z.discriminatedUnion("type", [
     BrainModelLoadResponseSchema,
     BrainModelUnloadResponseSchema,
     BrainModelDeleteResponseSchema,
+    BrainModelRenameResponseSchema,
+    BrainModelRenameResetResponseSchema,
     BrainLogsTailResponseSchema,
     AgentArchivedMessageSchema,
     CloseItemsResponseSchema,

package/dist/validation/ws-outbound-schema-metadata.d.ts

@@ -6488,6 +6488,7 @@ export declare const WSOutboundMessageSchema: {
                     running: import("zod").ZodBoolean;
                     pid: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     version: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
+                    apiVersion: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     host: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
                     port: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     displayHost: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
@@ -6510,6 +6511,8 @@ export declare const WSOutboundMessageSchema: {
                         load: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         resources: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         inventory: import("zod").ZodDefault<import("zod").ZodBoolean>;
+                        events: import("zod").ZodDefault<import("zod").ZodBoolean>;
+                        liveInference: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         writable: import("zod").ZodDefault<import("zod").ZodBoolean>;
                     }, import("zod/v4/core").$loose>>>;
                     resources: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodRecord<import("zod").ZodString, import("zod").ZodUnknown>>>;
@@ -6521,12 +6524,29 @@ export declare const WSOutboundMessageSchema: {
                         startedAt: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
                     }, import("zod/v4/core").$loose>>>;
                     reasoning: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodBoolean>>;
+                    inference: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodObject<{
+                        activeRequests: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        processing: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        thinking: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        generating: import("zod").ZodOptional<import("zod").ZodNumber>;
+                    }, import("zod/v4/core").$loose>>>;
                     slots: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodObject<{
                         total: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         busy: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         idle: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         prefill: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         decode: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                        threads: import("zod").ZodOptional<import("zod").ZodArray<import("zod").ZodObject<{
+                            slot: import("zod").ZodOptional<import("zod").ZodNumber>;
+                            phase: import("zod").ZodOptional<import("zod").ZodEnum<{
+                                prefill: "prefill";
+                                decode: "decode";
+                            }>>;
+                            promptTokens: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            generatedTokens: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            promptTokensPerSecond: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            tokensPerSecond: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                        }, import("zod/v4/core").$loose>>>;
                     }, import("zod/v4/core").$loose>>>;
                     queued: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     reachable: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodBoolean>>;
@@ -6541,6 +6561,7 @@ export declare const WSOutboundMessageSchema: {
                     running: import("zod").ZodBoolean;
                     pid: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     version: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
+                    apiVersion: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     host: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
                     port: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     displayHost: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
@@ -6563,6 +6584,8 @@ export declare const WSOutboundMessageSchema: {
                         load: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         resources: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         inventory: import("zod").ZodDefault<import("zod").ZodBoolean>;
+                        events: import("zod").ZodDefault<import("zod").ZodBoolean>;
+                        liveInference: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         writable: import("zod").ZodDefault<import("zod").ZodBoolean>;
                     }, import("zod/v4/core").$loose>>>;
                     resources: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodRecord<import("zod").ZodString, import("zod").ZodUnknown>>>;
@@ -6574,12 +6597,29 @@ export declare const WSOutboundMessageSchema: {
                         startedAt: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
                     }, import("zod/v4/core").$loose>>>;
                     reasoning: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodBoolean>>;
+                    inference: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodObject<{
+                        activeRequests: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        processing: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        thinking: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        generating: import("zod").ZodOptional<import("zod").ZodNumber>;
+                    }, import("zod/v4/core").$loose>>>;
                     slots: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodObject<{
                         total: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         busy: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         idle: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         prefill: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         decode: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                        threads: import("zod").ZodOptional<import("zod").ZodArray<import("zod").ZodObject<{
+                            slot: import("zod").ZodOptional<import("zod").ZodNumber>;
+                            phase: import("zod").ZodOptional<import("zod").ZodEnum<{
+                                prefill: "prefill";
+                                decode: "decode";
+                            }>>;
+                            promptTokens: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            generatedTokens: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            promptTokensPerSecond: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            tokensPerSecond: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                        }, import("zod/v4/core").$loose>>>;
                     }, import("zod/v4/core").$loose>>>;
                     queued: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     reachable: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodBoolean>>;
@@ -6594,6 +6634,7 @@ export declare const WSOutboundMessageSchema: {
                     running: import("zod").ZodBoolean;
                     pid: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     version: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
+                    apiVersion: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     host: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
                     port: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     displayHost: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
@@ -6616,6 +6657,8 @@ export declare const WSOutboundMessageSchema: {
                         load: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         resources: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         inventory: import("zod").ZodDefault<import("zod").ZodBoolean>;
+                        events: import("zod").ZodDefault<import("zod").ZodBoolean>;
+                        liveInference: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         writable: import("zod").ZodDefault<import("zod").ZodBoolean>;
                     }, import("zod/v4/core").$loose>>>;
                     resources: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodRecord<import("zod").ZodString, import("zod").ZodUnknown>>>;
@@ -6627,12 +6670,29 @@ export declare const WSOutboundMessageSchema: {
                         startedAt: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
                     }, import("zod/v4/core").$loose>>>;
                     reasoning: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodBoolean>>;
+                    inference: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodObject<{
+                        activeRequests: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        processing: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        thinking: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        generating: import("zod").ZodOptional<import("zod").ZodNumber>;
+                    }, import("zod/v4/core").$loose>>>;
                     slots: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodObject<{
                         total: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         busy: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         idle: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         prefill: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         decode: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                        threads: import("zod").ZodOptional<import("zod").ZodArray<import("zod").ZodObject<{
+                            slot: import("zod").ZodOptional<import("zod").ZodNumber>;
+                            phase: import("zod").ZodOptional<import("zod").ZodEnum<{
+                                prefill: "prefill";
+                                decode: "decode";
+                            }>>;
+                            promptTokens: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            generatedTokens: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            promptTokensPerSecond: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            tokensPerSecond: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                        }, import("zod/v4/core").$loose>>>;
                     }, import("zod/v4/core").$loose>>>;
                     queued: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     reachable: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodBoolean>>;
@@ -6647,6 +6707,7 @@ export declare const WSOutboundMessageSchema: {
                     running: import("zod").ZodBoolean;
                     pid: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     version: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
+                    apiVersion: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     host: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
                     port: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     displayHost: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
@@ -6669,6 +6730,8 @@ export declare const WSOutboundMessageSchema: {
                         load: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         resources: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         inventory: import("zod").ZodDefault<import("zod").ZodBoolean>;
+                        events: import("zod").ZodDefault<import("zod").ZodBoolean>;
+                        liveInference: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         writable: import("zod").ZodDefault<import("zod").ZodBoolean>;
                     }, import("zod/v4/core").$loose>>>;
                     resources: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodRecord<import("zod").ZodString, import("zod").ZodUnknown>>>;
@@ -6680,12 +6743,29 @@ export declare const WSOutboundMessageSchema: {
                         startedAt: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
                     }, import("zod/v4/core").$loose>>>;
                     reasoning: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodBoolean>>;
+                    inference: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodObject<{
+                        activeRequests: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        processing: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        thinking: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        generating: import("zod").ZodOptional<import("zod").ZodNumber>;
+                    }, import("zod/v4/core").$loose>>>;
                     slots: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodObject<{
                         total: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         busy: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         idle: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         prefill: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         decode: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                        threads: import("zod").ZodOptional<import("zod").ZodArray<import("zod").ZodObject<{
+                            slot: import("zod").ZodOptional<import("zod").ZodNumber>;
+                            phase: import("zod").ZodOptional<import("zod").ZodEnum<{
+                                prefill: "prefill";
+                                decode: "decode";
+                            }>>;
+                            promptTokens: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            generatedTokens: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            promptTokensPerSecond: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            tokensPerSecond: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                        }, import("zod/v4/core").$loose>>>;
                     }, import("zod/v4/core").$loose>>>;
                     queued: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     reachable: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodBoolean>>;
@@ -7273,6 +7353,7 @@ export declare const WSOutboundMessageSchema: {
                     running: import("zod").ZodBoolean;
                     pid: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     version: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
+                    apiVersion: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     host: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
                     port: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     displayHost: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
@@ -7295,6 +7376,8 @@ export declare const WSOutboundMessageSchema: {
                         load: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         resources: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         inventory: import("zod").ZodDefault<import("zod").ZodBoolean>;
+                        events: import("zod").ZodDefault<import("zod").ZodBoolean>;
+                        liveInference: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         writable: import("zod").ZodDefault<import("zod").ZodBoolean>;
                     }, import("zod/v4/core").$loose>>>;
                     resources: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodRecord<import("zod").ZodString, import("zod").ZodUnknown>>>;
@@ -7306,12 +7389,29 @@ export declare const WSOutboundMessageSchema: {
                         startedAt: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
                     }, import("zod/v4/core").$loose>>>;
                     reasoning: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodBoolean>>;
+                    inference: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodObject<{
+                        activeRequests: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        processing: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        thinking: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        generating: import("zod").ZodOptional<import("zod").ZodNumber>;
+                    }, import("zod/v4/core").$loose>>>;
                     slots: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodObject<{
                         total: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         busy: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         idle: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         prefill: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         decode: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                        threads: import("zod").ZodOptional<import("zod").ZodArray<import("zod").ZodObject<{
+                            slot: import("zod").ZodOptional<import("zod").ZodNumber>;
+                            phase: import("zod").ZodOptional<import("zod").ZodEnum<{
+                                prefill: "prefill";
+                                decode: "decode";
+                            }>>;
+                            promptTokens: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            generatedTokens: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            promptTokensPerSecond: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            tokensPerSecond: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                        }, import("zod/v4/core").$loose>>>;
                     }, import("zod/v4/core").$loose>>>;
                     queued: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     reachable: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodBoolean>>;
@@ -7336,6 +7436,7 @@ export declare const WSOutboundMessageSchema: {
                     running: import("zod").ZodBoolean;
                     pid: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     version: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
+                    apiVersion: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     host: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
                     port: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     displayHost: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
@@ -7358,6 +7459,8 @@ export declare const WSOutboundMessageSchema: {
                         load: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         resources: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         inventory: import("zod").ZodDefault<import("zod").ZodBoolean>;
+                        events: import("zod").ZodDefault<import("zod").ZodBoolean>;
+                        liveInference: import("zod").ZodDefault<import("zod").ZodBoolean>;
                         writable: import("zod").ZodDefault<import("zod").ZodBoolean>;
                     }, import("zod/v4/core").$loose>>>;
                     resources: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodRecord<import("zod").ZodString, import("zod").ZodUnknown>>>;
@@ -7369,12 +7472,29 @@ export declare const WSOutboundMessageSchema: {
                         startedAt: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodString>>;
                     }, import("zod/v4/core").$loose>>>;
                     reasoning: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodBoolean>>;
+                    inference: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodObject<{
+                        activeRequests: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        processing: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        thinking: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        generating: import("zod").ZodOptional<import("zod").ZodNumber>;
+                    }, import("zod/v4/core").$loose>>>;
                     slots: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodObject<{
                         total: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         busy: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         idle: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         prefill: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                         decode: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                        threads: import("zod").ZodOptional<import("zod").ZodArray<import("zod").ZodObject<{
+                            slot: import("zod").ZodOptional<import("zod").ZodNumber>;
+                            phase: import("zod").ZodOptional<import("zod").ZodEnum<{
+                                prefill: "prefill";
+                                decode: "decode";
+                            }>>;
+                            promptTokens: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            generatedTokens: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            promptTokensPerSecond: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                            tokensPerSecond: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
+                        }, import("zod/v4/core").$loose>>>;
                     }, import("zod/v4/core").$loose>>>;
                     queued: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodNumber>>;
                     reachable: import("zod").ZodOptional<import("zod").ZodNullable<import("zod").ZodBoolean>>;
@@ -7392,6 +7512,20 @@ export declare const WSOutboundMessageSchema: {
                 error: import("zod").ZodNullable<import("zod").ZodString>;
                 requestId: import("zod").ZodString;
             }, import("zod/v4/core").$strip>;
+        }, import("zod/v4/core").$strip>, import("zod").ZodObject<{
+            type: import("zod").ZodLiteral<"brain.model.rename.response">;
+            payload: import("zod").ZodObject<{
+                displayName: import("zod").ZodDefault<import("zod").ZodNullable<import("zod").ZodString>>;
+                error: import("zod").ZodNullable<import("zod").ZodString>;
+                requestId: import("zod").ZodString;
+            }, import("zod/v4/core").$strip>;
+        }, import("zod/v4/core").$strip>, import("zod").ZodObject<{
+            type: import("zod").ZodLiteral<"brain.model.rename.reset.response">;
+            payload: import("zod").ZodObject<{
+                displayName: import("zod").ZodDefault<import("zod").ZodNullable<import("zod").ZodString>>;
+                error: import("zod").ZodNullable<import("zod").ZodString>;
+                requestId: import("zod").ZodString;
+            }, import("zod/v4/core").$strip>;
         }, import("zod/v4/core").$strip>, import("zod").ZodObject<{
             type: import("zod").ZodLiteral<"brain.logs.tail.response">;
             payload: import("zod").ZodObject<{

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@otto-code/protocol",
-  "version": "0.8.1",
+  "version": "0.8.3",
   "description": "Otto shared protocol schemas and wire types",
   "license": "AGPL-3.0-or-later",
   "files": [