@ai-matrx/desktop-protocol 0.21.42 → 0.21.44

package/dist/index.cjs

@@ -194,6 +194,8 @@ __export(src_exports, {
   CalendarListParams: () => CalendarListParams,
   CalendarListResult: () => CalendarListResult,
   CancelMsg: () => CancelMsg,
+  CatalogFile: () => CatalogFile,
+  CatalogModelEntry: () => CatalogModelEntry,
   ChatAppProject: () => ChatAppProject,
   ChatFamily: () => ChatFamily,
   ChatHidden: () => ChatHidden,
@@ -479,6 +481,7 @@ __export(src_exports, {
   LEGACY_ACTIONS: () => LEGACY_ACTIONS,
   LEGACY_DISPATCHERS: () => LEGACY_DISPATCHERS,
   Limits: () => Limits,
+  LlamaServerBuild: () => LlamaServerBuild,
   LlmRunning: () => LlmRunning,
   LlmServerState: () => LlmServerState,
   LocationGetParams: () => LocationGetParams,
@@ -500,6 +503,8 @@ __export(src_exports, {
   MessagesMessage: () => MessagesMessage,
   MessagesSendParams: () => MessagesSendParams,
   MessagesSendResult: () => MessagesSendResult,
+  MlCatalogSetParams: () => MlCatalogSetParams,
+  MlCatalogSetResult: () => MlCatalogSetResult,
   MlInstallPhase: () => MlInstallPhase,
   MlInstallProgress: () => MlInstallProgress,
   MlProgress: () => MlProgress,
@@ -508,6 +513,7 @@ __export(src_exports, {
   MlServiceStatus: () => MlServiceStatus,
   MlStatusParams: () => MlStatusParams,
   MlStatusResult: () => MlStatusResult,
+  ModelKey: () => ModelKey,
   ModelsDownloadsCancelParams: () => ModelsDownloadsCancelParams,
   ModelsDownloadsCancelResult: () => ModelsDownloadsCancelResult,
   ModelsDownloadsListParams: () => ModelsDownloadsListParams,
@@ -697,6 +703,7 @@ __export(src_exports, {
   SessionListResult: () => SessionListResult,
   SessionReleaseParams: () => SessionReleaseParams,
   SessionReleaseResult: () => SessionReleaseResult,
+  Sha256Hex: () => Sha256Hex,
   ShownSection: () => ShownSection,
   Signal: () => Signal,
   SpendAccount: () => SpendAccount,
@@ -876,7 +883,7 @@ module.exports = __toCommonJS(src_exports);
 var import_zod = require("zod");
 var import_uuid = require("@ai-matrx/kit/uuid");
 var PROTOCOL_MAJOR = 1;
-var PROTOCOL_MINOR = 47;
+var PROTOCOL_MINOR = 49;
 var PROTOCOL_VERSION = `${PROTOCOL_MAJOR}.${PROTOCOL_MINOR}`;
 var ProtocolVersion = import_zod.z.string().regex(/^[1-9][0-9]{0,2}\.[0-9]{1,3}$/);
 var RequestId = import_zod.z.string().min(1).max(64).regex(/^[A-Za-z0-9_-]+$/);
@@ -1689,7 +1696,8 @@ var RelayStatusResponse = RelayDeviceStatusEvent;
 
 // src/schema/models.ts
 var import_zod10 = require("zod");
-var TranscriptionModelId = import_zod10.z.enum(["tiny", "base", "small", "large-v3-turbo"]);
+var ModelKey = import_zod10.z.string().min(1).max(128).regex(/^[a-z0-9][a-z0-9._-]*(\/[a-z0-9][a-z0-9._-]*)?$/);
+var TranscriptionModelId = ModelKey;
 var MlRuntimeState = import_zod10.z.enum(["not_installed", "installing", "installed", "failed"]);
 var MlInstallPhase = import_zod10.z.enum(["packages", "model", "verifying"]);
 var MlInstallProgress = import_zod10.z.strictObject({
@@ -1749,18 +1757,7 @@ var MlProgress = import_zod10.z.strictObject({
   /** The segment just transcribed (transcribe only): text appears as it is recognised. */
   segment: TranscriptSegment.nullable()
 });
-var TextModelId = import_zod10.z.enum([
-  "smollm2-135m",
-  "llama-3.2-3b",
-  "phi-4-mini",
-  "gemma-3-4b",
-  "llama-3.1-8b",
-  "qwen-2.5-7b",
-  "mistral-small-3.1",
-  "gpt-oss-20b",
-  "qwen-2.5-coder-7b",
-  "qwen-2.5-coder-14b"
-]);
+var TextModelId = ModelKey;
 var TextModelInfo = import_zod10.z.strictObject({
   id: TextModelId,
   /** Download size of the weights. */
@@ -1768,7 +1765,14 @@ var TextModelInfo = import_zod10.z.strictObject({
   installed: import_zod10.z.boolean()
 });
 var LlmServerState = import_zod10.z.enum(["stopped", "starting", "running", "failed"]);
-var LlmRunning = import_zod10.z.strictObject({ id: TextModelId, port: import_zod10.z.int().min(1024).max(65535) });
+var LlmRunning = import_zod10.z.strictObject({
+  id: TextModelId,
+  port: import_zod10.z.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: import_zod10.z.int().min(512).max(131072).optional(),
+  gpu_layers: import_zod10.z.int().min(-1).max(999).optional(),
+  threads: import_zod10.z.int().min(1).max(512).optional()
+});
 var ModelsLlmStatusParams = import_zod10.z.strictObject({});
 var ModelsLlmStatusResult = import_zod10.z.strictObject({
   state: LlmServerState,
@@ -1786,12 +1790,57 @@ var ModelsLlmInstallParams = import_zod10.z.strictObject({ model: TextModelId })
 var ModelsLlmInstallResult = import_zod10.z.strictObject({ model: TextModelId, elapsed_ms: nonNegInt });
 var ModelsLlmStartParams = import_zod10.z.strictObject({
   model: TextModelId,
-  /** Tokens of context; the service picks a size that fits this computer when omitted. */
-  context_length: import_zod10.z.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: import_zod10.z.int().min(512).max(131072).optional(),
+  /** (1.48) Layers offloaded to the GPU; -1 = all. From the knob desktop.models/gpu_layers. */
+  gpu_layers: import_zod10.z.int().min(-1).max(999).optional(),
+  /** (1.48) CPU threads, computed by the app from the knobs. */
+  threads: import_zod10.z.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: import_zod10.z.int().min(10).max(3600).optional()
 });
 var ModelsLlmStartResult = import_zod10.z.strictObject({ running: LlmRunning, elapsed_ms: nonNegInt });
 var ModelsLlmStopParams = import_zod10.z.strictObject({});
 var ModelsLlmStopResult = import_zod10.z.strictObject({ stopped: import_zod10.z.boolean() });
+var Sha256Hex = import_zod10.z.string().regex(/^[0-9a-f]{64}$/);
+var CatalogFile = import_zod10.z.strictObject({
+  /** Path inside the model's folder (the artifact's path_in_repo). */
+  path: import_zod10.z.string().min(1).max(512),
+  /** The pinned https URL it is downloaded from. */
+  url: import_zod10.z.string().url().max(2048),
+  size_bytes: nonNegInt,
+  sha256: Sha256Hex,
+  /** weights, weights_shard, mmproj, config, tokenizer, … (ai.model_artifact.role). */
+  role: import_zod10.z.string().max(32)
+});
+var CatalogModelEntry = import_zod10.z.strictObject({
+  /** The key callers name this model by (the 1.48 id for a seeded model). */
+  key: ModelKey,
+  /** The selected variant's key (`q4_k_m`, `ct2`). */
+  variant: import_zod10.z.string().min(1).max(64),
+  /** The upstream revision (or the variant id when there is none): a change re-downloads. */
+  revision: import_zod10.z.string().min(1).max(64),
+  multilingual: import_zod10.z.boolean(),
+  /** Exactly the files this computer fetches for this model: never more, never fewer. */
+  files: import_zod10.z.array(CatalogFile).min(1).max(256)
+});
+var LlamaServerBuild = import_zod10.z.strictObject({
+  version: import_zod10.z.string().min(1).max(32),
+  url: import_zod10.z.string().url().max(2048),
+  size_bytes: nonNegInt,
+  sha256: Sha256Hex
+});
+var MlCatalogSetParams = import_zod10.z.strictObject({
+  text: import_zod10.z.array(CatalogModelEntry).max(512),
+  transcription: import_zod10.z.array(CatalogModelEntry).max(64),
+  /** Null where the catalog has no build for this os/arch (installs then say so). */
+  llama_server: LlamaServerBuild.nullable(),
+  /** The knob desktop.models/default_transcription_model. */
+  default_transcription: TranscriptionModelId,
+  /** When the catalog was last changed (catalog_updated_at), for the service's log. */
+  catalog_updated_at: import_zod10.z.string().max(40)
+});
+var MlCatalogSetResult = import_zod10.z.strictObject({ text: nonNegInt, transcription: nonNegInt });
 var DownloadKind = import_zod10.z.enum(["speech_model", "text_model", "runtime"]);
 var DownloadRow = import_zod10.z.strictObject({
   /** The job's key in the service: "model:base", "llm:smollm2-135m", "llm-server", "runtime". */
@@ -5463,7 +5512,8 @@ var SERVICES = {
       "ml.llm_start": { params: ModelsLlmStartParams, result: ModelsLlmStartResult },
       "ml.llm_stop": { params: ModelsLlmStopParams, result: ModelsLlmStopResult },
       "ml.downloads": { params: ModelsDownloadsListParams, result: ModelsDownloadsListResult },
-      "ml.download_cancel": { params: ModelsDownloadsCancelParams, result: ModelsDownloadsCancelResult }
+      "ml.download_cancel": { params: ModelsDownloadsCancelParams, result: ModelsDownloadsCancelResult },
+      "ml.catalog_set": { params: MlCatalogSetParams, result: MlCatalogSetResult }
     }
   },
   coding: {
@@ -5615,6 +5665,13 @@ var NAMED_SCHEMAS = {
   ModelsProgressEvent,
   TextModelId,
   TextModelInfo,
+  ModelKey,
+  Sha256Hex,
+  CatalogFile,
+  CatalogModelEntry,
+  LlamaServerBuild,
+  MlCatalogSetParams,
+  MlCatalogSetResult,
   LlmServerState,
   LlmRunning,
   ModelsLlmStatusParams,
@@ -6262,6 +6319,8 @@ var NAMED_SCHEMAS = {
 
 // src/versioning.ts
 var FIELDS_SINCE = {
+  LlmRunning: { context_length: "1.48", gpu_layers: "1.48", threads: "1.48" },
+  ModelsLlmStartParams: { gpu_layers: "1.48", threads: "1.48", load_timeout_s: "1.48" },
   FsPatchResult: { edits_applied: "1.1", edits_failed: "1.1" },
   "ProtocolErrorBody.data": { edits_failed: "1.1" },
   "RelayCallResponse|0": { protocol_version: "1.1" },
@@ -6458,7 +6517,7 @@ function forPeer(op, part, value, version) {
 var VERSIONED_RESULT_OPS = [.../* @__PURE__ */ new Set([...RESULT_PATH.keys(), ...WINDOW_RESULTS.keys(), ...NESTED_RESULTS.keys()])];
 
 // src/contract.ts
-var CONTRACT_SHA256 = "502c8718b1d2fd0fde43daed047da5a7b158f81ffa3846c7d73cda879777fe37";
+var CONTRACT_SHA256 = "783487638685c921a761d27c5c6dca3103d1d6d5ae0ec38ffd59086b68a2207a";
 
 // src/close-policy.ts
 var CLOSE_RULES = {