@xenosystem/agent-sdk 0.10.3 → 0.10.5

package/dist/electron/index.d.cts

@@ -810,6 +810,7 @@ declare class LlmClient {
             parameters: ToolDefinition["input_schema"];
         };
     }>;
+    estimateChatCompletionsInputTokens(model: string, apiMessages: ApiMessage[], tools: ToolDefinition[]): number;
     private buildChatCompletionsRequestBody;
     private parseXenoCompletion;
     static parseRetryAfterSeconds(value: string | null): number | undefined;
@@ -818,8 +819,8 @@ declare class LlmClient {
     private static summarizeErrorBody;
     private authorizationHeaders;
     private createApiRequestSignal;
-    callChatCompletionsNonStream(model: string, apiMessages: ApiMessage[], tools: ToolDefinition[], onText?: (text: string) => void, signal?: AbortSignal): Promise<StreamResult>;
-    callChatCompletions(model: string, apiMessages: ApiMessage[], tools: ToolDefinition[], onText?: (text: string) => void, signal?: AbortSignal): Promise<StreamResult>;
+    callChatCompletionsNonStream(model: string, apiMessages: ApiMessage[], tools: ToolDefinition[], onText?: (text: string) => void, signal?: AbortSignal, maxOutputTokens?: number): Promise<StreamResult>;
+    callChatCompletions(model: string, apiMessages: ApiMessage[], tools: ToolDefinition[], onText?: (text: string) => void, signal?: AbortSignal, maxOutputTokens?: number): Promise<StreamResult>;
     callOllama(model: string, apiMessages: ApiMessage[], tools: ToolDefinition[], onText?: (text: string) => void, signal?: AbortSignal): Promise<StreamResult>;
 }
 type LLMProviderMessage = ApiMessage;
@@ -895,7 +896,7 @@ interface LLMProviderCapabilities {
 }
 interface LLMProvider {
     prepareQuotaRequest?(model: string, messages: LLMProviderMessage[], tools: ToolDefinition[], context?: LLMProviderRequestContext, signal?: AbortSignal): Promise<{
-        basis: "provider-enforced";
+        basis: "provider-enforced" | "estimated";
         inputTokensUpperBound: number;
         outputTokensUpperBound: number;
         execute(onText?: (text: string) => void, signal?: AbortSignal): Promise<LLMCompletionResult>;
@@ -1358,6 +1359,22 @@ interface ModelWorkAccounting {
     admit(request: Readonly<ModelWorkRequestIdentity>, signal: AbortSignal): void | Promise<void>;
     settle(result: Readonly<ModelWorkSettlement>, signal: AbortSignal): void | Promise<void>;
 }
+interface RunAuthorityDispatch {
+    turnId: string;
+    iteration: number;
+    model: string;
+    attempt: number;
+}
+interface RunAuthorityToolCall {
+    turnId: string;
+    iteration: number;
+    toolName: string;
+    toolCallId: string;
+}
+interface RunAuthority {
+    authorizeDispatch(request: Readonly<RunAuthorityDispatch>, signal: AbortSignal): void | Promise<void>;
+    authorizeToolCall(request: Readonly<RunAuthorityToolCall>, signal: AbortSignal): void | Promise<void>;
+}
 interface ModeSwitchRequest {
     from: ExecutionMode;
     to: ExecutionMode;
@@ -1397,6 +1414,7 @@ interface AgentLoopConfig {
     onToolEnd?: (name: string, result: ToolResult) => void;
     onIteration?: (iteration: number, totalTokens: number) => void;
     modelWorkAccounting?: ModelWorkAccounting;
+    runAuthority?: RunAuthority;
     onError?: (error: Error, context: string) => void;
     onTranscriptError?: (error: Error, context: string) => void;
     onModeSwitchRequest?: (request: ModeSwitchRequest) => Promise<boolean>;
@@ -1531,9 +1549,12 @@ declare class AgentLoop {
         meteredRequests: number;
         coverageComplete: boolean;
     };
+    createSoftBudgetCloudProvider(): LLMProvider;
+    private callCloudWithOutputCap;
     configureModelExecution(options: {
         provider?: LLMProvider;
         modelWorkAccounting?: ModelWorkAccounting;
+        runAuthority?: RunAuthority;
     }): void;
     get lastTraceId(): string | null;
     get lastRunTermination(): AgentRunTermination | null;

package/dist/electron/index.d.ts

@@ -810,6 +810,7 @@ declare class LlmClient {
             parameters: ToolDefinition["input_schema"];
         };
     }>;
+    estimateChatCompletionsInputTokens(model: string, apiMessages: ApiMessage[], tools: ToolDefinition[]): number;
     private buildChatCompletionsRequestBody;
     private parseXenoCompletion;
     static parseRetryAfterSeconds(value: string | null): number | undefined;
@@ -818,8 +819,8 @@ declare class LlmClient {
     private static summarizeErrorBody;
     private authorizationHeaders;
     private createApiRequestSignal;
-    callChatCompletionsNonStream(model: string, apiMessages: ApiMessage[], tools: ToolDefinition[], onText?: (text: string) => void, signal?: AbortSignal): Promise<StreamResult>;
-    callChatCompletions(model: string, apiMessages: ApiMessage[], tools: ToolDefinition[], onText?: (text: string) => void, signal?: AbortSignal): Promise<StreamResult>;
+    callChatCompletionsNonStream(model: string, apiMessages: ApiMessage[], tools: ToolDefinition[], onText?: (text: string) => void, signal?: AbortSignal, maxOutputTokens?: number): Promise<StreamResult>;
+    callChatCompletions(model: string, apiMessages: ApiMessage[], tools: ToolDefinition[], onText?: (text: string) => void, signal?: AbortSignal, maxOutputTokens?: number): Promise<StreamResult>;
     callOllama(model: string, apiMessages: ApiMessage[], tools: ToolDefinition[], onText?: (text: string) => void, signal?: AbortSignal): Promise<StreamResult>;
 }
 type LLMProviderMessage = ApiMessage;
@@ -895,7 +896,7 @@ interface LLMProviderCapabilities {
 }
 interface LLMProvider {
     prepareQuotaRequest?(model: string, messages: LLMProviderMessage[], tools: ToolDefinition[], context?: LLMProviderRequestContext, signal?: AbortSignal): Promise<{
-        basis: "provider-enforced";
+        basis: "provider-enforced" | "estimated";
         inputTokensUpperBound: number;
         outputTokensUpperBound: number;
         execute(onText?: (text: string) => void, signal?: AbortSignal): Promise<LLMCompletionResult>;
@@ -1358,6 +1359,22 @@ interface ModelWorkAccounting {
     admit(request: Readonly<ModelWorkRequestIdentity>, signal: AbortSignal): void | Promise<void>;
     settle(result: Readonly<ModelWorkSettlement>, signal: AbortSignal): void | Promise<void>;
 }
+interface RunAuthorityDispatch {
+    turnId: string;
+    iteration: number;
+    model: string;
+    attempt: number;
+}
+interface RunAuthorityToolCall {
+    turnId: string;
+    iteration: number;
+    toolName: string;
+    toolCallId: string;
+}
+interface RunAuthority {
+    authorizeDispatch(request: Readonly<RunAuthorityDispatch>, signal: AbortSignal): void | Promise<void>;
+    authorizeToolCall(request: Readonly<RunAuthorityToolCall>, signal: AbortSignal): void | Promise<void>;
+}
 interface ModeSwitchRequest {
     from: ExecutionMode;
     to: ExecutionMode;
@@ -1397,6 +1414,7 @@ interface AgentLoopConfig {
     onToolEnd?: (name: string, result: ToolResult) => void;
     onIteration?: (iteration: number, totalTokens: number) => void;
     modelWorkAccounting?: ModelWorkAccounting;
+    runAuthority?: RunAuthority;
     onError?: (error: Error, context: string) => void;
     onTranscriptError?: (error: Error, context: string) => void;
     onModeSwitchRequest?: (request: ModeSwitchRequest) => Promise<boolean>;
@@ -1531,9 +1549,12 @@ declare class AgentLoop {
         meteredRequests: number;
         coverageComplete: boolean;
     };
+    createSoftBudgetCloudProvider(): LLMProvider;
+    private callCloudWithOutputCap;
     configureModelExecution(options: {
         provider?: LLMProvider;
         modelWorkAccounting?: ModelWorkAccounting;
+        runAuthority?: RunAuthority;
     }): void;
     get lastTraceId(): string | null;
     get lastRunTermination(): AgentRunTermination | null;