@ai-matrx/desktop-protocol 0.21.39 → 0.21.43

package/dist/client.d.cts

@@ -1,8 +1,8 @@
 import { BackoffOptions } from '@ai-matrx/realtime';
 export { RECONNECT_ALARM_ATTEMPTS } from '@ai-matrx/realtime';
 import { z } from 'zod';
-import { W as Welcome, D as DesktopProtocolError, C as CloseOutcome, O as OpName, a as OpParams, b as OpResult, c as OPS, S as StreamMeta, E as EndReason, d as EVENTS, e as ClientType, M as ModelsTranscribeResult, T as TranscriptionModelId, f as MlProgress } from './models-Ut2nuRzV.cjs';
-export { i as isDesktopProtocolError } from './models-Ut2nuRzV.cjs';
+import { W as Welcome, D as DesktopProtocolError, C as CloseOutcome, O as OpName, a as OpParams, b as OpResult, c as OPS, S as StreamMeta, E as EndReason, d as EVENTS, e as ClientType, M as ModelsTranscribeResult, T as TranscriptionModelId, f as MlProgress } from './models-HrUx_KP2.cjs';
+export { i as isDesktopProtocolError } from './models-HrUx_KP2.cjs';
 import { FrameHeader } from './frame.cjs';
 
 /** The subset of the WHATWG WebSocket the client uses — browser, React Native, `ws` and Workers all fit. */

package/dist/client.d.ts

@@ -1,8 +1,8 @@
 import { BackoffOptions } from '@ai-matrx/realtime';
 export { RECONNECT_ALARM_ATTEMPTS } from '@ai-matrx/realtime';
 import { z } from 'zod';
-import { W as Welcome, D as DesktopProtocolError, C as CloseOutcome, O as OpName, a as OpParams, b as OpResult, c as OPS, S as StreamMeta, E as EndReason, d as EVENTS, e as ClientType, M as ModelsTranscribeResult, T as TranscriptionModelId, f as MlProgress } from './models-Ut2nuRzV.js';
-export { i as isDesktopProtocolError } from './models-Ut2nuRzV.js';
+import { W as Welcome, D as DesktopProtocolError, C as CloseOutcome, O as OpName, a as OpParams, b as OpResult, c as OPS, S as StreamMeta, E as EndReason, d as EVENTS, e as ClientType, M as ModelsTranscribeResult, T as TranscriptionModelId, f as MlProgress } from './models-HrUx_KP2.js';
+export { i as isDesktopProtocolError } from './models-HrUx_KP2.js';
 import { FrameHeader } from './frame.js';
 
 /** The subset of the WHATWG WebSocket the client uses — browser, React Native, `ws` and Workers all fit. */

package/dist/client.js

@@ -112,7 +112,7 @@ import { z as z2 } from "zod";
 import { z } from "zod";
 import { UUID_SHAPE_PATTERN } from "@ai-matrx/kit/uuid";
 var PROTOCOL_MAJOR = 1;
-var PROTOCOL_MINOR = 47;
+var PROTOCOL_MINOR = 48;
 var PROTOCOL_VERSION = `${PROTOCOL_MAJOR}.${PROTOCOL_MINOR}`;
 var ProtocolVersion = z.string().regex(/^[1-9][0-9]{0,2}\.[0-9]{1,3}$/);
 var RequestId = z.string().min(1).max(64).regex(/^[A-Za-z0-9_-]+$/);
@@ -843,7 +843,14 @@ var TextModelInfo = z9.strictObject({
   installed: z9.boolean()
 });
 var LlmServerState = z9.enum(["stopped", "starting", "running", "failed"]);
-var LlmRunning = z9.strictObject({ id: TextModelId, port: z9.int().min(1024).max(65535) });
+var LlmRunning = z9.strictObject({
+  id: TextModelId,
+  port: z9.int().min(1024).max(65535),
+  /** (1.48) What the service actually applied, echoed so the app can attest "read and applied". Optional until the service sends them. */
+  context_length: z9.int().min(512).max(131072).optional(),
+  gpu_layers: z9.int().min(-1).max(999).optional(),
+  threads: z9.int().min(1).max(512).optional()
+});
 var ModelsLlmStatusParams = z9.strictObject({});
 var ModelsLlmStatusResult = z9.strictObject({
   state: LlmServerState,
@@ -861,8 +868,14 @@ var ModelsLlmInstallParams = z9.strictObject({ model: TextModelId });
 var ModelsLlmInstallResult = z9.strictObject({ model: TextModelId, elapsed_ms: nonNegInt });
 var ModelsLlmStartParams = z9.strictObject({
   model: TextModelId,
-  /** Tokens of context; the service picks a size that fits this computer when omitted. */
-  context_length: z9.int().min(512).max(131072).optional()
+  /** Tokens of context, from the knob desktop.models/context_length. Omitted today means the service's fixed 4096 (optional until callers send it). */
+  context_length: z9.int().min(512).max(131072).optional(),
+  /** (1.48) Layers offloaded to the GPU; -1 = all. From the knob desktop.models/gpu_layers. */
+  gpu_layers: z9.int().min(-1).max(999).optional(),
+  /** (1.48) CPU threads, computed by the app from the knobs. */
+  threads: z9.int().min(1).max(512).optional(),
+  /** (1.48) Seconds to wait for the model to load, computed by the app from the knobs. */
+  load_timeout_s: z9.int().min(10).max(3600).optional()
 });
 var ModelsLlmStartResult = z9.strictObject({ running: LlmRunning, elapsed_ms: nonNegInt });
 var ModelsLlmStopParams = z9.strictObject({});