feat: stop delivery on client disconnect across providers, and opt-in non-streaming timing for all providers

CHANGELOG.md

@@ -10,11 +10,15 @@
 - Malformed fixture `citations` or `files` (not an array, a missing or mistyped field, a non-integer offset, an offset outside the text, or citations on a turn with no text: tool calls only, `blocks` with no text block, or empty content) answer a 500 before any content is written: `aimock_invalid_fixture_citation` or `aimock_invalid_fixture_file`. A non-`image/*` file on the OpenAI Responses API answers a 500 with `aimock_unsupported_generated_file`. This is the same on every wire that reads these keys, with or without a misbehavior fault: the fixture is checked on its original turn before any fault applies (only an invalid misbehavior config answers its own 400 or 501 first). A valid fixture under a fault that changes the reply text is served without its `citations`. Only Chat Completions, Anthropic Messages, Gemini and OpenAI Responses read these keys; other wires ignore them.
 - Fixture load validation warns about malformed `citations` and `files`, with the same problem description the 500 carries (the 500 message adds an `aimock: invalid fixture` prefix; the warning adds which 500 the requests answer). The fixture still loads.
 - `InvalidFixtureOutputError` (the error behind `aimock_invalid_fixture_citation` and `aimock_invalid_fixture_file`) is exported from `@copilotkit/aimock`.
+- Opt-in fixture `nonStreamingTimingAllProviders: true` delays non-streaming replies on OpenAI chat, Anthropic, Responses, Gemini/Vertex, Gemini Interactions, Cohere, Bedrock invoke/converse and Ollama; `nonStreamingTiming` is unchanged (OpenAI chat only); error replies are never delayed. On the OpenAI Responses API the wait also counts the stream frames that fixture `citations` and generated `files` add
 
 ### Fixed
 
 - Misbehavior faults on Chat Completions, Anthropic Messages, Gemini and the OpenAI Responses API (HTTP and WebSocket): a fixture the wire rejects (generated `files` on Chat Completions or Anthropic, a non-`image/*` file on Responses, malformed `citations` or `files`) is never faulted. It takes the normal path, is journaled as a 500 that keeps its fixture, and consumes no fault `times` budget or seed ordinal; `aimock_misbehavior_total` does not count it. A non-stream Chat Completions fault is journaled before its timing wait.
 - Journal accuracy: a client that hangs up during the timing wait of any non-stream Chat Completions reply (faulted or not) is journaled as interrupted, not as served.
+- Client disconnects stop delivery (no more chunks or waits) on OpenAI chat, Anthropic, Responses, Gemini/Vertex, Gemini Interactions, Cohere, Bedrock, Ollama, transcription streams and the Responses, Realtime and Gemini Live WebSockets
+- Client disconnects on those surfaces are journaled as `interrupted` with `interruptReason: "client aborted"` (HTTP `status: 0` when nothing was sent); aimock-caused teardowns record `"disconnect"`. Journal snapshots of hung-up requests now carry these fields
+- WebSocket: a client TCP FIN without a close frame now closes the connection at once (1006) and ends the turn
 - Docker image: the base image is pinned to `node:22.23.4-alpine` by digest. Node 22.23.4 fixes CVE-2026-111497 (Brotli decompression memory bug); the 1.46.0 image shipped Node 22.23.3.
 
 ## [1.46.0] - 2026-10-20

docs/aws-bedrock/index.html

@@ -107,6 +107,16 @@ <h2>URL Patterns</h2>
           </tbody>
         </table>
 
+        <p>
+          If the client hangs up before the reply is complete, aimock stops delivery and journals
+          the request with <code>interruptReason: "client aborted"</code> (see
+          <a href="/error-injection#client-disconnects">Client Disconnects</a>). To delay
+          InvokeModel and Converse replies, set
+          <code>nonStreamingTimingAllProviders: true</code> with a <code>streamingProfile</code> or
+          <code>recordedTimings</code> (see
+          <a href="/streaming-physics#non-streaming-all-providers">All Providers</a>).
+        </p>
+
         <h2>Request Format</h2>
         <p>
           Bedrock Claude requests use the Anthropic Messages format. The

docs/chat-completions/index.html

@@ -73,6 +73,16 @@ <h2>Endpoint</h2>
           </tbody>
         </table>
 
+        <p>
+          If the client hangs up before the reply is complete, aimock stops delivery and journals
+          the request with <code>interruptReason: "client aborted"</code> (see
+          <a href="/error-injection#client-disconnects">Client Disconnects</a>). To delay
+          <code>stream: false</code> replies, set <code>nonStreamingTiming: true</code> or
+          <code>nonStreamingTimingAllProviders: true</code> with a <code>streamingProfile</code> or
+          <code>recordedTimings</code> (see
+          <a href="/streaming-physics#non-streaming">Non-Streaming Responses</a>).
+        </p>
+
         <h2>Unit Test: Text Response</h2>
         <p>
           Using the programmatic API with vitest, register a fixture and assert on the response.

docs/cohere/index.html

@@ -73,6 +73,16 @@ <h2>Endpoint</h2>
           </tbody>
         </table>
 
+        <p>
+          If the client hangs up before the reply is complete, aimock stops delivery and journals
+          the request with <code>interruptReason: "client aborted"</code> (see
+          <a href="/error-injection#client-disconnects">Client Disconnects</a>). To delay
+          <code>stream: false</code> replies, set
+          <code>nonStreamingTimingAllProviders: true</code> with a <code>streamingProfile</code> or
+          <code>recordedTimings</code> (see
+          <a href="/streaming-physics#non-streaming-all-providers">All Providers</a>).
+        </p>
+
         <h2>Key Features</h2>
         <ul>
           <li>

docs/error-injection/index.html

@@ -188,8 +188,50 @@ <h2>Interruption Behavior</h2>
             <code>response.interrupted: true</code> and <code>response.interruptReason</code>
           </li>
           <li>
-            A client that disconnects from a Claude Messages stream is recorded the same way, with
-            <code>response.interruptReason: "client aborted"</code>
+            A client that disconnects is recorded the same way, with
+            <code>response.interruptReason: "client aborted"</code> (see
+            <a href="#client-disconnects">Client Disconnects</a>)
+          </li>
+        </ul>
+
+        <h2 id="client-disconnects">Client Disconnects</h2>
+        <p>
+          When the client hangs up before the reply is complete, aimock stops delivery. It sends no
+          more chunks and skips the remaining waits. This applies to these endpoints, streaming and
+          non-streaming:
+        </p>
+        <ul>
+          <li>OpenAI Chat Completions, Anthropic Messages and OpenAI Responses</li>
+          <li>Gemini and Vertex AI, and Gemini Interactions</li>
+          <li>Cohere v2 chat</li>
+          <li>Bedrock InvokeModel and Converse, and their streaming variants</li>
+          <li>Ollama <code>/api/chat</code> and <code>/api/generate</code></li>
+          <li>Audio transcription streams</li>
+          <li>WebSocket: OpenAI Responses, OpenAI Realtime and Gemini Live</li>
+        </ul>
+        <p>
+          On WebSocket, a client that closes its TCP connection without sending a close frame is
+          also a disconnect. aimock closes the connection at once (code 1006, "Connection lost"),
+          stops the turn and journals it.
+        </p>
+        <p>The journal entry records how the reply ended:</p>
+        <ul>
+          <li>
+            <code>response.interrupted: true</code> and
+            <code>response.interruptReason: "client aborted"</code> when the client left.
+          </li>
+          <li>
+            On HTTP, <code>response.status: 0</code> when no byte of the reply reached the client.
+            Otherwise the status stays the one aimock sent.
+          </li>
+          <li>
+            <code>response.interruptReason: "disconnect"</code> when aimock tore the response down
+            itself, for example with the chaos <code>disconnect</code> action or a server-side
+            destroy. This is not blamed on the client.
+          </li>
+          <li>
+            Fixture interruptions (<code>truncateAfterChunks</code>,
+            <code>disconnectAfterMs</code>) keep their own reasons.
           </li>
         </ul>
 

docs/fixtures/index.html

@@ -1049,10 +1049,23 @@ <h2>Fixture Options</h2>
               <td>
                 Also delay <code>stream: false</code> OpenAI Chat Completions replies by the
                 fixture's <code>streamingProfile</code> or <code>recordedTimings</code> (default
-                <code>false</code>). See
+                <code>false</code>). Other providers ignore it (use
+                <code>nonStreamingTimingAllProviders</code>). See
                 <a href="/streaming-physics#non-streaming">Non-Streaming Responses</a>
               </td>
             </tr>
+            <tr>
+              <td>nonStreamingTimingAllProviders</td>
+              <td>boolean</td>
+              <td>
+                Delay <code>stream: false</code> replies the same way on OpenAI Chat Completions,
+                Anthropic Messages, OpenAI Responses, Gemini, Vertex AI, Gemini Interactions,
+                Cohere, Bedrock InvokeModel and Converse, and Ollama (default <code>false</code>).
+                Needs <code>streamingProfile</code> or <code>recordedTimings</code>. Error replies
+                and audio transcription are not delayed. See
+                <a href="/streaming-physics#non-streaming-all-providers">All Providers</a>
+              </td>
+            </tr>
             <tr>
               <td>chaos</td>
               <td>object</td>

docs/gemini-interactions/index.html

@@ -75,6 +75,16 @@ <h2>Endpoint</h2>
           </tbody>
         </table>
 
+        <p>
+          If the client hangs up before the reply is complete, aimock stops delivery and journals
+          the request with <code>interruptReason: "client aborted"</code> (see
+          <a href="/error-injection#client-disconnects">Client Disconnects</a>). To delay
+          <code>stream: false</code> replies, set
+          <code>nonStreamingTimingAllProviders: true</code> with a <code>streamingProfile</code> or
+          <code>recordedTimings</code> (see
+          <a href="/streaming-physics#non-streaming-all-providers">All Providers</a>).
+        </p>
+
         <h2>Quick Start</h2>
         <p>Set up a minimal mock that responds to a Gemini Interactions request:</p>
 

docs/gemini/index.html

@@ -98,6 +98,16 @@ <h2>Endpoints</h2>
           </tbody>
         </table>
 
+        <p>
+          If the client hangs up before the reply is complete, aimock stops delivery and journals
+          the request with <code>interruptReason: "client aborted"</code> (see
+          <a href="/error-injection#client-disconnects">Client Disconnects</a>). Gemini Live records
+          the turn the same way. To delay <code>generateContent</code> replies, set
+          <code>nonStreamingTimingAllProviders: true</code> with a <code>streamingProfile</code> or
+          <code>recordedTimings</code> (see
+          <a href="/streaming-physics#non-streaming-all-providers">All Providers</a>).
+        </p>
+
         <h2 id="model-catalog">Model Catalog</h2>
         <p>
           The native catalog contains Google IDs and unrecognized IDs from fixtures, in first-seen

docs/ollama/index.html

@@ -93,6 +93,16 @@ <h2>Endpoints</h2>
           </tbody>
         </table>
 
+        <p>
+          If the client hangs up before the reply is complete, aimock stops delivery and journals
+          the request with <code>interruptReason: "client aborted"</code> (see
+          <a href="/error-injection#client-disconnects">Client Disconnects</a>). To delay
+          <code>"stream": false</code> replies, set
+          <code>nonStreamingTimingAllProviders: true</code> with a <code>streamingProfile</code> or
+          <code>recordedTimings</code> (see
+          <a href="/streaming-physics#non-streaming-all-providers">All Providers</a>).
+        </p>
+
         <h2>Embeddings</h2>
         <p>
           Use the Ollama SDK's <code>embed({ model, input })</code> with a string or array of

docs/openrouter-chat/index.html

@@ -107,6 +107,16 @@ <h2>Endpoints</h2>
           </tbody>
         </table>
 
+        <p>
+          If the client hangs up before the reply is complete, aimock stops delivery and journals
+          the request with <code>interruptReason: "client aborted"</code> (see
+          <a href="/error-injection#client-disconnects">Client Disconnects</a>). To delay
+          <code>stream: false</code> replies, set <code>nonStreamingTiming: true</code> or
+          <code>nonStreamingTimingAllProviders: true</code> with a <code>streamingProfile</code> or
+          <code>recordedTimings</code> (see
+          <a href="/streaming-physics#non-streaming">Non-Streaming Responses</a>).
+        </p>
+
         <h2>Point the OpenAI SDK at aimock</h2>
         <p>
           The dominant real-world integration path is the OpenAI SDK with its

docs/responses-api/index.html

@@ -77,6 +77,16 @@ <h2>Endpoints</h2>
           </tbody>
         </table>
 
+        <p>
+          If the client hangs up before the reply is complete, aimock stops delivery and journals
+          the request with <code>interruptReason: "client aborted"</code> (see
+          <a href="/error-injection#client-disconnects">Client Disconnects</a>). The WebSocket
+          transport records the turn the same way. To delay HTTP <code>stream: false</code> replies,
+          set <code>nonStreamingTimingAllProviders: true</code> with a
+          <code>streamingProfile</code> or <code>recordedTimings</code> (see
+          <a href="/streaming-physics#non-streaming-all-providers">All Providers</a>).
+        </p>
+
         <h2>Unit Test: HTTP SSE Text Response</h2>
 
         <div class="code-block">

docs/streaming-physics/index.html

@@ -195,9 +195,65 @@ <h2 id="non-streaming">Non-Streaming Responses</h2>
         <p>
           <code>nonStreamingTiming</code> has an effect only on OpenAI Chat Completions, and only
           when the fixture also sets <code>streamingProfile</code> or <code>recordedTimings</code>.
-          Non-streaming responses from other providers are not delayed.
+          Non-streaming responses from other providers are not delayed by it.
         </p>
 
+        <h3 id="non-streaming-all-providers">All Providers</h3>
+        <p>
+          To delay <code>stream: false</code> replies on other providers too, set
+          <code>nonStreamingTimingAllProviders: true</code> on the fixture. It uses the same formula
+          as <code>nonStreamingTiming: true</code>, and it also needs a
+          <code>streamingProfile</code> or <code>recordedTimings</code>. It applies to these
+          non-streaming endpoints:
+        </p>
+        <ul>
+          <li>OpenAI Chat Completions (the same delay as <code>nonStreamingTiming: true</code>)</li>
+          <li>Anthropic Messages</li>
+          <li>OpenAI Responses</li>
+          <li>Gemini and Vertex AI <code>generateContent</code></li>
+          <li>Gemini Interactions (with <code>stream: false</code>)</li>
+          <li>Cohere v2 chat</li>
+          <li>Bedrock InvokeModel and Converse</li>
+          <li>Ollama <code>/api/chat</code> and <code>/api/generate</code></li>
+        </ul>
+        <p>
+          Audio transcription replies are not delayed. <code>nonStreamingTiming</code> is unchanged:
+          it still applies to OpenAI Chat Completions only. Setting
+          <code>nonStreamingTimingAllProviders: false</code> does not turn off
+          <code>nonStreamingTiming: true</code>.
+        </p>
+
+        <div class="code-block">
+          <div class="code-block-header">
+            fixtures/slow-non-streaming-all.json <span class="lang-tag">json</span>
+          </div>
+          <pre><code>{
+  <span class="key">"fixtures"</span>: [
+    {
+      <span class="key">"match"</span>: { <span class="key">"userMessage"</span>: <span class="str">"think carefully"</span> },
+      <span class="key">"response"</span>: { <span class="key">"content"</span>: <span class="str">"Let me think about this..."</span> },
+      <span class="key">"streamingProfile"</span>: { <span class="key">"ttft"</span>: <span class="num">800</span>, <span class="key">"tps"</span>: <span class="num">50</span> },
+      <span class="key">"nonStreamingTimingAllProviders"</span>: <span class="kw">true</span>
+    }
+  ]
+}</code></pre>
+        </div>
+
+        <h3 id="non-streaming-errors-and-disconnects">Errors and Disconnects</h3>
+        <ul>
+          <li>
+            Error replies (a fixture <code>error</code> response) are never delayed. aimock sends
+            them at once, with the fixture status.
+          </li>
+          <li>
+            If the client disconnects during the wait, aimock writes nothing. The journal entry
+            records <code>response.interrupted: true</code>,
+            <code>response.interruptReason: "client aborted"</code> and
+            <code>response.status: 0</code>. See
+            <a href="/error-injection#client-disconnects">Client Disconnects</a>.
+          </li>
+        </ul>
+
         <h2>Interaction with <code>latency</code></h2>
         <ul>
           <li>
@@ -216,7 +272,8 @@ <h2>Interaction with <code>latency</code></h2>
             <code>latency</code> and <code>--latency</code> are gaps between SSE chunks only. They
             never delay a non-streaming response, including the fallback case above. To delay
             <code>stream: false</code> replies, set <code>streamingProfile</code> and
-            <code>nonStreamingTiming: true</code> (see
+            <code>nonStreamingTiming: true</code> (OpenAI Chat Completions) or
+            <code>nonStreamingTimingAllProviders: true</code> (see
             <a href="#non-streaming">Non-Streaming Responses</a>).
           </li>
         </ul>

docs/transcription/index.html

@@ -73,6 +73,13 @@ <h2>Endpoint</h2>
           </tbody>
         </table>
 
+        <p>
+          If the client hangs up during a streamed transcription (<code>stream=true</code>), aimock
+          stops the stream and journals the request with
+          <code>interruptReason: "client aborted"</code> (see
+          <a href="/error-injection#client-disconnects">Client Disconnects</a>).
+        </p>
+
         <h2>Unit Test: Simple Transcription</h2>
         <p>
           Using the programmatic API with vitest, register a fixture and assert on the response.

docs/vertex-ai/index.html

@@ -82,6 +82,16 @@ <h2>Endpoints</h2>
           </tbody>
         </table>
 
+        <p>
+          If the client hangs up before the reply is complete, aimock stops delivery and journals
+          the request with <code>interruptReason: "client aborted"</code> (see
+          <a href="/error-injection#client-disconnects">Client Disconnects</a>). To delay
+          <code>generateContent</code> replies, set
+          <code>nonStreamingTimingAllProviders: true</code> with a <code>streamingProfile</code> or
+          <code>recordedTimings</code> (see
+          <a href="/streaming-physics#non-streaming-all-providers">All Providers</a>).
+        </p>
+
         <h2>URL Pattern Difference</h2>
         <p>
           The key difference between consumer Gemini and Vertex AI is the URL routing. Consumer

docs/websocket/index.html

@@ -88,6 +88,13 @@ <h2>Endpoints</h2>
           </tbody>
         </table>
 
+        <p>
+          On OpenAI Responses, OpenAI Realtime and Gemini Live, if the client closes the connection
+          during a reply, aimock stops sending and journals the turn with
+          <code>interruptReason: "client aborted"</code> (see
+          <a href="/error-injection#client-disconnects">Client Disconnects</a>).
+        </p>
+
         <h2 id="openai-live">OpenAI GPT-Live</h2>
         <p>
           GPT-Live uses a primary WebSocket upgrade: <code>GET /v1/live/sessions</code>. Connect to