diff --git a/desktop/src/main/services/ffmpeg.ts b/desktop/src/main/services/ffmpeg.ts
index 6b1171459..2b55ac366 100644
--- a/desktop/src/main/services/ffmpeg.ts
+++ b/desktop/src/main/services/ffmpeg.ts
@@ -112,3 +112,56 @@ const ffmpegBinaryPath = () => {
     // https://github.com/eugeneware/ffmpeg-static/issues/16
     return ensure(pathToFfmpeg).replace("app.asar", "app.asar.unpacked");
 };
+
+/**
+ * A variant of {@link ffmpegExec} adapted to work with streams so that it can
+ * handle the MP4 conversion of large video files.
+ *
+ * See: [Note: Convert to MP4]
+ *
+ * @param command
+ * @param dataOrPathOrZipItem
+ * @param outputFileExtension
+ * @param timeoutMS
+ * @returns
+ */
+export const ffmpegConvertToMP4 = async (
+    command: string[],
+    dataOrPathOrZipItem: Uint8Array | string | ZipItem,
+    outputFileExtension: string,
+    timeoutMS: number,
+): Promise<Uint8Array> => {
+    // TODO (MR): This currently copies files for both input (when
+    // dataOrPathOrZipItem is data) and output. This needs to be tested
+    // extremely large video files when invoked downstream of `convertToMP4` in
+    // the web code.
+
+    const {
+        path: inputFilePath,
+        isFileTemporary: isInputFileTemporary,
+        writeToTemporaryFile: writeToTemporaryInputFile,
+    } = await makeFileForDataOrPathOrZipItem(dataOrPathOrZipItem);
+
+    const outputFilePath = await makeTempFilePath(outputFileExtension);
+    try {
+        await writeToTemporaryInputFile();
+
+        const cmd = substitutePlaceholders(
+            command,
+            inputFilePath,
+            outputFilePath,
+        );
+
+        if (timeoutMS) await withTimeout(execAsync(cmd), timeoutMS);
+        else await execAsync(cmd);
+
+        return fs.readFile(outputFilePath);
+    } finally {
+        try {
+            if (isInputFileTemporary) await deleteTempFile(inputFilePath);
+            await deleteTempFile(outputFilePath);
+        } catch (e) {
+            log.error("Could not clean up temp files", e);
+        }
+    }
+};
diff --git a/desktop/src/main/stream.ts b/desktop/src/main/stream.ts
index bae13aa12..2c8a8f44b 100644
--- a/desktop/src/main/stream.ts
+++ b/desktop/src/main/stream.ts
@@ -181,3 +181,32 @@ const writeNodeStream = async (filePath: string, fileStream: Readable) => {
         });
     });
 };
+
+/**
+ * [Note: Convert to MP4]
+ *
+ * When we want to convert a video to MP4, if we were to send the entire
+ * contents of the video from the renderer to the main process over IPC, it just
+ * causes the renderer to run out of memory and restart when the videos are very
+ * large. So we need to stream the original video renderer → main and then
+ * stream back the converted video renderer ← main.
+ *
+ * Currently Chromium does not support bi-directional streaming ("full" duplex
+ * mode for the Web fetch API). So we need to simulate that using two different
+ * streaming requests.
+ *
+ *     renderer → main  stream://convert-to-mp4
+ *                      → request.body is the original video
+ *                      ← response is a token
+ *
+ *     renderer → main  stream://convert-to-mp4?token=<token>
+ *                      ← response.body is the converted video
+ *
+ * Note that the conversion itself is not streaming. The conversion still
+ * happens in a single shot, we are just streaming the data across the IPC
+ * boundary to allow us to pass large amounts of data without running out of
+ * memory.
+ *
+ * See also: [Note: IPC streams]
+ */
+const convertToMP4 = (token: string | undefined) => {};