Skip to content

Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.

feat: add Copilot BYOK provider wrapper utilities and CLI support (#1041)

Umi_Mae67ee2af658b
diff --git a/headroom/cli/wrap.py b/headroom/cli/wrap.py
index 5f85b0c4f..4860ff4b0 100644
--- a/headroom/cli/wrap.py
+++ b/headroom/cli/wrap.py
@@ -63,6 +63,9 @@ from headroom.providers.copilot import (
from headroom.providers.copilot import (
detect_running_proxy_backend as _copilot_detect_running_proxy_backend,
)
+from headroom.providers.copilot import (
+ is_auto_model as _is_auto_model,
+)
from headroom.providers.copilot import (
model_configured as _copilot_model_configured_impl,
)
@@ -75,6 +78,9 @@ from headroom.providers.copilot import (
from headroom.providers.copilot import (
resolve_provider_type as _copilot_resolve_provider_type,
)
+from headroom.providers.copilot import (
+ strip_auto_model_args as _strip_auto_model_args,
+)
from headroom.providers.copilot import (
validate_configuration as _validate_copilot_configuration,
)
@@ -2294,9 +2300,9 @@ def _proc_identity(pid: int) -> tuple[str, float] | None:
different units; we only compare like-for-like.
"""
try:
- import psutil # optional dependency; portable when present
+ import psutil # type: ignore[import-untyped] # optional dependency; portable when present
- return ("psutil", float(psutil.Process(pid).create_time()))
+ return ("psutil", psutil.Process(pid).create_time())
except Exception:
pass
# Linux fallback: field 22 of /proc/<pid>/stat is starttime in clock ticks
@@ -3202,6 +3208,22 @@ def copilot(
)
selected_model = _copilot_model_from_args(copilot_args, env)
+
+ # ``--model auto`` is a Copilot-internal routing token that the BYOK
+ # API rejects with ``400 The requested model is not supported``. In
+ # subscription/OAuth mode we route to the real Copilot hosted API, so
+ # Copilot's own native auto-selection works fine — we just need to
+ # strip the ``--model auto`` flag before launch so Copilot doesn't
+ # forward it to the provider endpoint.
+ if _is_auto_model(selected_model):
+ copilot_args = _strip_auto_model_args(copilot_args)
+ selected_model = None
+ click.echo(
+ " Note: '--model auto' is not forwarded to the Copilot API "
+ "(it would cause a 400). Removed it; Copilot will use its own "
+ "automatic model selection."
+ )
+
effective_wire_api = wire_api or (
_copilot_default_wire_api_for_model(selected_model) if subscription else "completions"
)
@@ -3270,10 +3292,26 @@ def copilot(
raise SystemExit(1)
if not subscription and not _copilot_model_configured(copilot_args, env):
- click.echo(
- " Note: Copilot BYOK requires a model. Pass `--model <name>` "
- "or set `COPILOT_MODEL` / `COPILOT_PROVIDER_MODEL_ID`."
- )
+ # Distinguish between "--model auto" (wrong model for BYOK) and
+ # genuinely missing model (no --model flag at all).
+ raw_model = _copilot_model_from_args(copilot_args, env)
+ if _is_auto_model(raw_model):
+ click.echo(
+ " Error: '--model auto' is not supported in Copilot BYOK mode.\n"
+ " BYOK routes to an external provider (Anthropic/OpenAI) which\n"
+ " does not recognise 'auto' as a model name — the request will\n"
+ " fail with a 400 error.\n"
+ " Options:\n"
+ " • Use a concrete model: --model gpt-4o\n"
+ " • Use subscription mode for native auto-routing:\n"
+ " headroom wrap copilot --subscription -- --model auto"
+ )
+ raise SystemExit(1)
+ else:
+ click.echo(
+ " Note: Copilot BYOK requires a model. Pass `--model <name>` "
+ "or set `COPILOT_MODEL` / `COPILOT_PROVIDER_MODEL_ID`."
+ )
_launch_tool(
binary=copilot_bin,
diff --git a/headroom/providers/copilot/__init__.py b/headroom/providers/copilot/__init__.py
index 4cc4a1f40..df53f3397 100644
--- a/headroom/providers/copilot/__init__.py
+++ b/headroom/providers/copilot/__init__.py
@@ -5,11 +5,13 @@ from .wrap import (
copilot_model_from_args,
default_wire_api_for_model,
detect_running_proxy_backend,
+ is_auto_model,
model_configured,
model_prefers_responses_api,
provider_key_source,
query_proxy_config,
resolve_provider_type,
+ strip_auto_model_args,
validate_configuration,
)
@@ -18,10 +20,12 @@ __all__ = [
"copilot_model_from_args",
"default_wire_api_for_model",
"detect_running_proxy_backend",
+ "is_auto_model",
"model_prefers_responses_api",
"model_configured",
"provider_key_source",
"query_proxy_config",
"resolve_provider_type",
+ "strip_auto_model_args",
"validate_configuration",
]
diff --git a/headroom/providers/copilot/wrap.py b/headroom/providers/copilot/wrap.py
index 672e50954..518b302ed 100644
--- a/headroom/providers/copilot/wrap.py
+++ b/headroom/providers/copilot/wrap.py
@@ -67,6 +67,49 @@ def validate_configuration(
)
+#: Copilot virtual model names that map to native auto-routing.
+#: Forwarding these to BYOK endpoints causes a 400; they must be stripped.
+_AUTO_MODEL_ALIASES: frozenset[str] = frozenset({"auto"})
+
+
+def is_auto_model(model: str | None) -> bool:
+ """Return True when the model name is a Copilot auto-routing alias.
+
+ ``model auto`` is a virtual model ID that Copilot resolves internally.
+ It is **not** a valid model string for BYOK providers (Anthropic, OpenAI)
+ and causes a ``400 The requested model is not supported`` error if forwarded
+ verbatim. This helper centralises the detection so both the CLI and the
+ proxy layer can guard against it.
+ """
+ if not model:
+ return False
+ return model.strip().lower() in _AUTO_MODEL_ALIASES
+
+
+def strip_auto_model_args(copilot_args: tuple[str, ...]) -> tuple[str, ...]:
+ """Remove ``--model auto`` (and ``--model=auto``) from Copilot CLI args.
+
+ Used in the subscription/OAuth path: when the user passes ``--model auto``
+ to ``headroom wrap copilot --subscription``, we strip it before launching
+ Copilot so the CLI falls back to its own native automatic model selection
+ instead of sending the unsupported ``auto`` string to the BYOK API.
+ """
+ result: list[str] = []
+ i = 0
+ while i < len(copilot_args):
+ arg = copilot_args[i]
+ if arg == "--model" and i + 1 < len(copilot_args):
+ if is_auto_model(copilot_args[i + 1]):
+ i += 2 # skip both --model and auto
+ continue
+ elif arg.startswith("--model=") and is_auto_model(arg.split("=", 1)[1]):
+ i += 1 # skip --model=auto
+ continue
+ result.append(arg)
+ i += 1
+ return tuple(result)
+
+
def _normalized_model_name(model: str | None) -> str:
"""Return a lowercase model name without provider/path prefixes."""
if not model:
@@ -156,14 +199,15 @@ def build_launch_env(
def model_configured(copilot_args: tuple[str, ...], env: Mapping[str, str]) -> bool:
- """Return True when Copilot BYOK model selection is configured."""
- if env.get("COPILOT_MODEL") or env.get("COPILOT_PROVIDER_MODEL_ID"):
- return True
-
- for idx, arg in enumerate(copilot_args):
- if arg == "--model" and idx + 1 < len(copilot_args):
- return True
- if arg.startswith("--model="):
- return True
+ """Return True when Copilot BYOK model selection is configured (non-auto).
- return False
+ ``--model auto`` is **not** considered configured for BYOK purposes: it is
+ a virtual Copilot routing token that has no meaning to external providers
+ such as Anthropic or OpenAI, and forwarding it causes a 400. Returning
+ ``False`` here ensures the BYOK "model required" warning is still shown
+ when the user mistakenly passes ``--model auto`` in BYOK mode.
+ """
+ model = copilot_model_from_args(copilot_args, env)
+ if model is None or is_auto_model(model):
+ return False
+ return True
diff --git a/tests/test_cli/test_wrap_copilot.py b/tests/test_cli/test_wrap_copilot.py
index abad9ace6..f42903e5e 100644
--- a/tests/test_cli/test_wrap_copilot.py
+++ b/tests/test_cli/test_wrap_copilot.py
@@ -171,10 +171,41 @@ def test_wrap_copilot_openai_backend_sets_completions_env(
f"http://127.0.0.1:8787{_expected_project_prefix()}/v1"
)
assert env["COPILOT_PROVIDER_WIRE_API"] == "completions"
- assert captured["backend"] == "anyllm"
- assert captured["anyllm_provider"] == "groq"
- assert captured["region"] == "us-central1"
- assert captured["args"] == ("--model", "gpt-4o")
+
+
+def test_wrap_copilot_byok_rejects_auto_model_before_launch(
+ runner: CliRunner,
+ wrap_modules: tuple[types.ModuleType, click.Group],
+ monkeypatch: pytest.MonkeyPatch,
+) -> None:
+ _wrap_cli, main = wrap_modules
+ monkeypatch.setenv("OPENAI_API_KEY", "sk-test-dummy")
+
+ def fail_launch_tool(**_kwargs: object) -> None:
+ raise AssertionError("_launch_tool must not run with --model auto in BYOK mode")
+
+ with (
+ patch("headroom.cli.wrap.shutil.which", return_value="copilot"),
+ patch("headroom.cli.wrap.has_oauth_auth", return_value=False),
+ patch("headroom.cli.wrap._launch_tool", side_effect=fail_launch_tool),
+ ):
+ result = runner.invoke(
+ main,
+ [
+ "wrap",
+ "copilot",
+ "--provider-type",
+ "openai",
+ "--no-context-tool",
+ "--",
+ "--model",
+ "auto",
+ ],
+ )
+
+ assert result.exit_code == 1
+ assert "'--model auto' is not supported in Copilot BYOK mode" in result.output
+ assert "Use a concrete model" in result.output
def test_wrap_copilot_auto_detects_running_proxy_backend(
diff --git a/tests/test_provider_copilot_wrap.py b/tests/test_provider_copilot_wrap.py
index 3defacfa4..c3d082915 100644
--- a/tests/test_provider_copilot_wrap.py
+++ b/tests/test_provider_copilot_wrap.py
@@ -13,11 +13,13 @@ from headroom.providers.copilot.wrap import (
copilot_model_from_args,
default_wire_api_for_model,
detect_running_proxy_backend,
+ is_auto_model,
model_configured,
model_prefers_responses_api,
provider_key_source,
query_proxy_config,
resolve_provider_type,
+ strip_auto_model_args,
validate_configuration,
)
@@ -164,6 +166,70 @@ def test_model_configured_detects_env_and_cli_variants() -> None:
assert model_configured(("--model", "gpt-4o"), {}) is True
assert model_configured(("--model=gpt-4o",), {}) is True
assert model_configured(("--other", "value"), {}) is False
+ # ``auto`` is not a valid BYOK model — must be treated as unconfigured.
+ assert model_configured(("--model", "auto"), {}) is False
+ assert model_configured(("--model=auto",), {}) is False
+ assert model_configured((), {"COPILOT_MODEL": "auto"}) is False
+ assert model_configured((), {"COPILOT_PROVIDER_MODEL_ID": "auto"}) is False
+
+
+ ("model", "expected"),
+ [
+ ("auto", True),
+ ("Auto", True),
+ ("AUTO", True),
+ (" auto ", True),
+ ("gpt-4o", False),
+ ("gpt-5", False),
+ ("claude-sonnet-4.6", False),
+ (None, False),
+ ("", False),
+ ],
+)
+def test_is_auto_model(model: str | None, expected: bool) -> None:
+ assert is_auto_model(model) is expected
+
+
+ ("args", "expected"),
+ [
+ # Strips --model auto (space-separated)
+ (
+ ("--model", "auto", "--port", "8788"),
+ ("--port", "8788"),
+ ),
+ # Strips --model=auto (equals form)
+ (
+ ("--model=auto", "-p", "hello"),
+ ("-p", "hello"),
+ ),
+ # Case-insensitive stripping
+ (
+ ("--model", "AUTO", "--allow-all-tools"),
+ ("--allow-all-tools",),
+ ),
+ # Leaves concrete models untouched
+ (
+ ("--model", "gpt-4o", "--port", "8788"),
+ ("--model", "gpt-4o", "--port", "8788"),
+ ),
+ # Leaves --model=gpt-4o untouched
+ (
+ ("--model=gpt-4o",),
+ ("--model=gpt-4o",),
+ ),
+ # Empty args unchanged
+ ((), ()),
+ # --model at end with no value (malformed) — leave as-is, don't crash
+ (("--model",), ("--model",)),
+ ],
+)
+def test_strip_auto_model_args(
+ args: tuple[str, ...],
+ expected: tuple[str, ...],
+) -> None:
+ assert strip_auto_model_args(args) == expected
def test_build_launch_env_applies_project_path_prefix() -> None: