Improve in-flight cancel latency via run abort signal propagation

src/models/anthropic.ts

@@ -100,7 +100,10 @@ export class AnthropicClient implements ModelClient {
       params.thinking = { type: 'enabled', budget_tokens: 4096 };
     }
 
-    const response = await this.client.messages.create(params) as AnthropicMessage;
+    const response = await this.client.messages.create(
+      params,
+      request.signal ? { signal: request.signal } : undefined,
+    ) as AnthropicMessage;
 
     const textContent = response.content.find((c) => c.type === 'text');
     const content = textContent?.type === 'text' ? textContent.text : '';

src/models/bedrock.ts

@@ -65,7 +65,10 @@ export class BedrockClient implements ModelClient {
     }
 
     const command = new ConverseCommand(params);
-    const response = await this.client.send(command);
+    const response = await this.client.send(
+      command,
+      request.signal ? { abortSignal: request.signal } : undefined,
+    );
 
     // Extract text and tool_use content from the response
     const outputContent = response.output?.message?.content ?? [];
@@ -126,7 +129,10 @@ export class BedrockClient implements ModelClient {
 
     try {
       const command = new ConverseStreamCommand(params);
-      const response = await this.client.send(command);
+      const response = await this.client.send(
+        command,
+        request.signal ? { abortSignal: request.signal } : undefined,
+      );
 
       let inputTokens = 0;
       let outputTokens = 0;

src/models/github.ts

@@ -163,7 +163,10 @@ export class GitHubModelsClient implements ModelClient {
       (params as OpenAI.ChatCompletionCreateParamsNonStreaming & { reasoning_effort?: 'low' | 'medium' | 'high' }).reasoning_effort = 'medium';
     }
 
-    const response = await this.client.chat.completions.create(params);
+    const response = await this.client.chat.completions.create(
+      params,
+      request.signal ? { signal: request.signal } : undefined,
+    );
 
     const choice = response.choices[0];
     const content = choice?.message?.content ?? '';
@@ -237,7 +240,10 @@ export class GitHubModelsClient implements ModelClient {
     }
 
     try {
-      const stream = await this.client.chat.completions.create(params);
+      const stream = await this.client.chat.completions.create(
+        params,
+        request.signal ? { signal: request.signal } : undefined,
+      );
 
       let totalInputTokens = 0;
       let totalOutputTokens = 0;

src/models/local/llamacpp.ts

@@ -247,13 +247,16 @@ export class LlamaCppClient implements ModelClient {
       }
 
       const controller = new AbortController();
+      const signal = request.signal
+        ? AbortSignal.any([request.signal, controller.signal])
+        : controller.signal;
       const timer = setTimeout(() => controller.abort(), this.requestTimeout);
       try {
         response = await fetch(`${this.endpoint}/v1/chat/completions`, {
           method: 'POST',
           headers,
           body: JSON.stringify(body),
-          signal: controller.signal,
+          signal,
         });
       } finally {
         clearTimeout(timer);
@@ -331,6 +334,7 @@ export class LlamaCppClient implements ModelClient {
         method: 'POST',
         headers,
         body: JSON.stringify(body),
+        signal: request.signal,
       });
 
       if (!response.ok) {

src/models/openai.ts

@@ -140,6 +140,7 @@ export class OpenAIClient implements ModelClient {
       method: 'POST',
       headers,
       body: JSON.stringify(body),
+      signal: request.signal,
     });
 
     if (!res.ok) {
@@ -277,7 +278,10 @@ export class OpenAIClient implements ModelClient {
 
     let response: OpenAI.ChatCompletion;
     try {
-      response = await this.client.chat.completions.create(params);
+      response = await this.client.chat.completions.create(
+        params,
+        request.signal ? { signal: request.signal } : undefined,
+      );
     } catch (error) {
       const status = typeof (error as { status?: unknown })?.status === 'number'
         ? (error as { status: number }).status

src/models/types.ts

@@ -80,6 +80,8 @@ export interface ChatRequest {
   tools?: ToolDefinition[];
   /** Enable extended thinking/reasoning mode for this request. */
   thinking?: boolean;
+  /** Optional abort signal for cancelling in-flight provider requests. */
+  signal?: AbortSignal;
 }
 
 export interface ChatResponse {