diff --git a/headroom/cli/proxy.py b/headroom/cli/proxy.py
index 88e4c0c26..229693374 100644
--- a/headroom/cli/proxy.py
+++ b/headroom/cli/proxy.py
@@ -417,6 +417,18 @@ def dashboard(port: int, no_open: bool) -> None:
         "Env: HEADROOM_ANTHROPIC_PRE_UPSTREAM_MEMORY_CONTEXT_TIMEOUT_SECONDS."
     ),
 )
+@click.option(
+    "--compression-max-workers",
+    type=int,
+    default=None,
+    envvar="HEADROOM_COMPRESSION_MAX_WORKERS",
+    help=(
+        "Bound the dedicated compression threadpool (CPU-bound Kompress work). "
+        "Default (unset): min(32, (cpu_count or 1) * 4). Lower it to reduce CPU "
+        "oversubscription under concurrent sessions; a value < 1 is clamped to 1. "
+        "Env: HEADROOM_COMPRESSION_MAX_WORKERS."
+    ),
+)
 @click.option(
     "--log-file",
     default=None,
@@ -843,6 +855,7 @@ def proxy(
     anthropic_pre_upstream_concurrency: int | None,
     anthropic_pre_upstream_acquire_timeout_seconds: float | None,
     anthropic_pre_upstream_memory_context_timeout_seconds: float | None,
+    compression_max_workers: int | None,
     log_file: str | None,
     log_messages: bool,
     codex_wire_debug: bool,
@@ -1167,6 +1180,7 @@ def proxy(
         # Precedence: CLI > env > auto-compute (click's ``envvar``
         # handles the env-var fallback).
         anthropic_pre_upstream_concurrency=anthropic_pre_upstream_concurrency,
+        compression_max_workers=compression_max_workers,
         anthropic_pre_upstream_acquire_timeout_seconds=(
             anthropic_pre_upstream_acquire_timeout_seconds
             if anthropic_pre_upstream_acquire_timeout_seconds is not None
diff --git a/tests/test_cli_proxy_improvements.py b/tests/test_cli_proxy_improvements.py
index ae1e1b6e1..3230fc2ec 100644
--- a/tests/test_cli_proxy_improvements.py
+++ b/tests/test_cli_proxy_improvements.py
@@ -448,3 +448,34 @@ class TestHelpTextCompleteness:
         result = runner.invoke(main, ["proxy", "--mode", "bogus_mode_xyz"])
         assert result.exit_code != 0
         assert "invalid" in result.output.lower() or "choice" in result.output.lower()
+
+
+class TestCompressionMaxWorkers:
+    """--compression-max-workers / HEADROOM_COMPRESSION_MAX_WORKERS must reach ProxyConfig.
+
+    Regression: the field was documented in ProxyConfig and consumed by the
+    server, but the CLI never defined the option or passed it through, so it
+    was permanently None (always resolving to the min(32, cpu*4) default).
+    """
+
+    def test_flag_reaches_config(self, runner: CliRunner, mock_run_server: dict) -> None:
+        result = runner.invoke(
+            main, ["proxy", "--compression-max-workers", "3"], catch_exceptions=False
+        )
+        assert result.exit_code == 0, result.output
+        assert mock_run_server["config"].compression_max_workers == 3
+
+    def test_env_reaches_config(self, runner: CliRunner, mock_run_server: dict) -> None:
+        result = runner.invoke(
+            main,
+            ["proxy"],
+            env={"HEADROOM_COMPRESSION_MAX_WORKERS": "5"},
+            catch_exceptions=False,
+        )
+        assert result.exit_code == 0, result.output
+        assert mock_run_server["config"].compression_max_workers == 5
+
+    def test_default_is_none(self, runner: CliRunner, mock_run_server: dict) -> None:
+        result = runner.invoke(main, ["proxy"], catch_exceptions=False)
+        assert result.exit_code == 0, result.output
+        assert mock_run_server["config"].compression_max_workers is None