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.
Python agentaianthropicclaude-codecompressioncontext-engineeringcontext-windowcursorfastapilangchainllmmcpopenaiprompt-engineeringproxypythonragtoken-optimizationtokenstypescript
fix(doctor): detect root openai_base_url Codex routing and agent-configured proxy ports (#4490)
gglucass222f2d1df3c6
diff --git a/headroom/cli/doctor.py b/headroom/cli/doctor.pyindex aa98ec6da..eee6dc20f 100644--- a/headroom/cli/doctor.py+++ b/headroom/cli/doctor.py@@ -46,7 +46,7 @@ from headroom.providers.claude import ()from .main import get_version, main-from .port_discovery import find_live_proxy_elsewhere, is_headroom_livez+from .port_discovery import find_live_proxy_elsewhere, is_headroom_livez, loopback_portfrom .wrap import _read_wrap_marker, _wrap_marker_is_stalePASS = "pass"@@ -59,6 +59,7 @@ DEFAULT_PROXY_PORT = 8787_LOOPBACK_URL_RE = re.compile(r"https?://(?:127\.0\.0\.1|localhost):(\d+)")_CODEX_BASE_URL_RE = re.compile(r'(?m)^[ \t]*base_url\s*=\s*"([^"\r\n]+)"')_CODEX_MODEL_PROVIDER_RE = re.compile(r'(?m)^[ \t]*model_provider\s*=\s*"([^"\r\n]+)"')+_CODEX_OPENAI_BASE_URL_RE = re.compile(r'(?m)^[ \t]*openai_base_url\s*=\s*"([^"\r\n]+)"')# Ollama's fixed default port. `ollama launch claude` writes# ``ANTHROPIC_BASE_URL=http://127.0.0.1:11434`` into the launched Claude Code@@ -573,7 +574,19 @@ def check_wrap_marker_staleness(settings_path: Path) -> CheckResult:def codex_active_base_url(text: str) -> tuple[str, str | None]:"""Return ``(provider_id, base_url)`` for the active Codex provider in ``text``."""- active_match = _CODEX_MODEL_PROVIDER_RE.search(text)+ first_table = re.search(r"(?m)^[ \t]*\[", text)+ root = text[: first_table.start()] if first_table is not None else text+ # Only a root `model_provider` picks the default session's provider; one+ # under `[profiles.*]` applies only when that profile is selected.+ active_match = _CODEX_MODEL_PROVIDER_RE.search(root)+ if active_match is None or active_match.group(1) == "openai":+ # A root `openai_base_url` re-points Codex's built-in `openai` provider+ # (the default when `model_provider` is unset), so it beats an inactive+ # `[model_providers.headroom]` table. Headroom Desktop routes this way+ # because the ChatGPT app hides features for any other provider.+ root_url = _CODEX_OPENAI_BASE_URL_RE.search(root)+ if root_url is not None:+ return "openai", root_url.group(1)provider_id = active_match.group(1) if active_match else "headroom"return provider_id, _codex_provider_base_url(text, provider_id)@@ -661,6 +674,24 @@ def _codex_provider_base_url(text: str, provider_id: str) -> str | None:return base_url_match.group(1) if base_url_match else None+def _configured_client_ports(+ claude_paths: Sequence[Path], codex_paths: Sequence[Path]+) -> list[int]:+ """Loopback ports the Claude Code and Codex configs (project and user) send traffic to.++ Headroom Desktop (port 6767) and ``wrap --port N`` leave no deployment+ manifest, so without these a bare ``doctor`` only ever finds 8787.+ """+ urls = [_claude_base_url_in(path)[0] for path in claude_paths]+ for path in codex_paths:+ try:+ text = path.read_text(encoding="utf-8", errors="replace")+ except OSError:+ continue+ urls.append(codex_active_base_url(text)[1] or "")+ return [port for port in map(loopback_port, urls) if port is not None]++def _codex_block_missing_openai_auth(text: str, config_path: Path, provider_id: str = "headroom") -> bool:@@ -1211,20 +1242,25 @@ def doctor(health = probe_json(f"{base_url}/health", timeout=5.0) if livez else Nonestats = probe_json(f"{base_url}/stats", timeout=5.0) if livez else Noneinstalled = get_version()+ project_claude_settings = Path.cwd() / ".claude" / "settings.json"+ project_local_claude_settings = Path.cwd() / ".claude" / "settings.local.json"live_elsewhere = Noneif livez is None:# Cheap, bounded: only ports Headroom recorded (HEADROOM_PORT, 8787,- # deployment manifests, the project's wrap marker), probed in parallel.+ # deployment manifests, the project's wrap marker) and the ports the+ # agents are configured for, probed in parallel.live_elsewhere = find_live_proxy_elsewhere(port,manifests=manifests,+ extra=_configured_client_ports(+ [project_local_claude_settings, project_claude_settings, claude_settings_path()],+ [codex_project_config_path(), codex_config_path()],+ ),probe=lambda p: is_headroom_livez(probe_json(f"http://127.0.0.1:{p}/livez", timeout=0.5)),)- project_claude_settings = Path.cwd() / ".claude" / "settings.json"- project_local_claude_settings = Path.cwd() / ".claude" / "settings.local.json"checks = [note_live_proxy_elsewhere(check_proxy_liveness(livez, base_url), live_elsewhere),check_proxy_readiness(livez, readyz, base_url),diff --git a/tests/test_cli/test_port_alignment.py b/tests/test_cli/test_port_alignment.pyindex 908c0bb1b..b14382a95 100644--- a/tests/test_cli/test_port_alignment.py+++ b/tests/test_cli/test_port_alignment.py@@ -434,6 +434,21 @@ class TestHookAlignment:message = init_mod.alignment_warning("codex", "init-user", probe=lambda p: False)assert message is not None and "--port 8787 codex" in message+ def test_codex_root_openai_base_url_beats_an_inactive_headroom_table(+ self, home: Path, tmp_path: Path, monkeypatch+ ) -> None:+ custom = tmp_path / "custom-codex"+ custom.mkdir()+ (custom / "config.toml").write_text(+ 'openai_base_url = "http://127.0.0.1:9200/v1"\n[model_providers.headroom]\n'+ 'base_url = "http://127.0.0.1:8787/v1"\n',+ encoding="utf-8",+ )+ monkeypatch.setenv("CODEX_HOME", str(custom))+ monkeypatch.setattr(init_mod, "load_manifest", lambda profile: _manifest(profile, 8787))+ message = init_mod.alignment_warning("codex", "init-user", probe=lambda p: False)+ assert message is not None and "127.0.0.1:9200" in message+def test_session_start_emits_hook_json(self, claude_settings: Path) -> None:self._route_claude(claude_settings, 9200)result = CliRunner().invoke(diff --git a/tests/test_cli_doctor.py b/tests/test_cli_doctor.pyindex 0b37c9975..665d5e1a5 100644--- a/tests/test_cli_doctor.py+++ b/tests/test_cli_doctor.py@@ -587,6 +587,57 @@ class TestCodexRouting:assert result.status == WARNassert "9999" in result.summary+ def test_root_openai_base_url_on_builtin_provider_passes(self, tmp_path):+ """Headroom Desktop's layout: built-in `openai` provider, root base URL."""+ path = tmp_path / "config.toml"+ path.write_text(+ 'model_provider = "openai"\n'+ 'openai_base_url = "http://127.0.0.1:6767/v1"\n'+ '[projects."/work/app"]\n'+ 'trust_level = "trusted"\n',+ encoding="utf-8",+ )+ result = check_codex_routing(path, 6767)+ assert result.status == PASS, result.summary+ assert check_codex_routing(path, 8787).status == WARN++ def test_root_openai_base_url_without_model_provider_passes(self, tmp_path):+ path = tmp_path / "config.toml"+ path.write_text('openai_base_url = "http://127.0.0.1:8787/v1"\n', encoding="utf-8")+ assert check_codex_routing(path, 8787).status == PASS++ def test_openai_base_url_inside_a_table_is_not_root(self, tmp_path):+ path = tmp_path / "config.toml"+ path.write_text(+ 'model_provider = "openai"\n'+ "[profiles.fast]\n"+ 'openai_base_url = "http://127.0.0.1:8787/v1"\n',+ encoding="utf-8",+ )+ assert check_codex_routing(path, 8787).status == WARN++ def test_root_openai_base_url_beats_an_inactive_headroom_table(self, tmp_path):+ """No root `model_provider`: Codex uses its built-in `openai`, not the table."""+ path = tmp_path / "config.toml"+ path.write_text(+ 'openai_base_url = "http://127.0.0.1:6767/v1"\n'+ "[model_providers.headroom]\n"+ 'base_url = "http://127.0.0.1:8787/v1"\n',+ encoding="utf-8",+ )+ assert check_codex_routing(path, 6767).status == PASS+ assert check_codex_routing(path, 8787).status == WARN++ def test_profile_model_provider_does_not_hide_the_default_route(self, tmp_path):+ path = tmp_path / "config.toml"+ path.write_text(+ 'openai_base_url = "http://127.0.0.1:6767/v1"\n'+ "[profiles.work]\n"+ 'model_provider = "azure"\n',+ encoding="utf-8",+ )+ assert check_codex_routing(path, 6767).status == PASS+def test_no_marker_warns(self, tmp_path):path = tmp_path / "config.toml"path.write_text('model = "gpt-5"\n', encoding="utf-8")@@ -986,6 +1037,49 @@ class TestDoctorCommand:assert "http://127.0.0.1:8789/livez" in seenassert not any("127.0.0.1:8787" in url for url in seen)+ def test_names_the_live_port_the_agents_are_configured_for(self, runner, isolated, monkeypatch):+ """Headroom Desktop runs on 6767 and records no deployment manifest."""+ monkeypatch.setenv("HEADROOM_PORT_DISCOVERY", "1")+ (isolated / "settings.json").write_text(+ json.dumps({"env": {"ANTHROPIC_BASE_URL": "http://127.0.0.1:6767"}}),+ encoding="utf-8",+ )++ def desktop_probe(url, timeout=2.0):+ return LIVEZ_OK if url == "http://127.0.0.1:6767/livez" else None++ monkeypatch.setattr(doctor_mod, "probe_json", desktop_probe)+ result = runner.invoke(main, ["doctor", "--json"])+ proxy = next(c for c in json.loads(result.output)["checks"] if c["name"] == "proxy")+ assert "a Headroom proxy IS running on port 6767" in proxy["summary"]++ @pytest.mark.parametrize(+ ("relative", "content"),+ [+ ("project-codex.toml", 'openai_base_url = "http://127.0.0.1:6767/v1"\n'),+ (+ ".claude/settings.local.json",+ json.dumps({"env": {"ANTHROPIC_BASE_URL": "http://127.0.0.1:6767"}}),+ ),+ ],+ )+ def test_names_the_live_port_only_a_project_config_routes_to(+ self, runner, isolated, monkeypatch, relative, content+ ):+ monkeypatch.setenv("HEADROOM_PORT_DISCOVERY", "1")+ monkeypatch.chdir(isolated)+ path = isolated / relative+ path.parent.mkdir(exist_ok=True)+ path.write_text(content, encoding="utf-8")++ def desktop_probe(url, timeout=2.0):+ return LIVEZ_OK if url == "http://127.0.0.1:6767/livez" else None++ monkeypatch.setattr(doctor_mod, "probe_json", desktop_probe)+ result = runner.invoke(main, ["doctor", "--json"])+ proxy = next(c for c in json.loads(result.output)["checks"] if c["name"] == "proxy")+ assert "a Headroom proxy IS running on port 6767" in proxy["summary"]+class TestCostTrackerBudgetKeys:def test_stats_exposes_budget_config(self):