tsugite-codex-cli 0.23.1__tar.gz → 0.24.0__tar.gz

{tsugite_codex_cli-0.23.1 → tsugite_codex_cli-0.24.0}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.5
 Name: tsugite-codex-cli
-Version: 0.23.1
+Version: 0.24.0
 Summary: Tsugite plugin: ChatGPT subscription auth via Codex CLI token store
 Author: Justyn Shull
 Requires-Python: >=3.11
-Requires-Dist: tsugite-cli==0.23.1
+Requires-Dist: tsugite-cli==0.24.0

{tsugite_codex_cli-0.23.1 → tsugite_codex_cli-0.24.0}/pyproject.toml

@@ -1,16 +1,19 @@
 [project]
 name = "tsugite-codex-cli"
-version = "0.23.1"
+version = "0.24.0"
 description = "Tsugite plugin: ChatGPT subscription auth via Codex CLI token store"
 authors = [{ name = "Justyn Shull" }]
 requires-python = ">=3.11"
 dependencies = [
-  "tsugite-cli==0.23.1",
+  "tsugite-cli==0.24.0",
 ]
 
 [project.entry-points."tsugite.providers"]
 codex_cli = "tsugite_codex_cli:create_provider"
 
+[project.entry-points."tsugite.usage_providers"]
+codex_cli = "tsugite_codex_cli.usage"
+
 [build-system]
 requires = ["hatchling"]
 build-backend = "hatchling.build"

{tsugite_codex_cli-0.23.1 → tsugite_codex_cli-0.24.0}/tsugite_codex_cli/auth.py

@@ -64,9 +64,9 @@ class CodexAuthStore:
 
     async def get_access_token(self) -> tuple[str, str]:
         """Return (access_token, account_id), refreshing if near expiry."""
-        return await asyncio.to_thread(self._sync_get_or_refresh)
+        return await asyncio.to_thread(self.sync_get_or_refresh)
 
-    def _sync_get_or_refresh(self) -> tuple[str, str]:
+    def sync_get_or_refresh(self) -> tuple[str, str]:
         if not self._auth_path.exists():
             raise CodexAuthError(f"No Codex auth file at {self._auth_path}. Run `codex login` first.")
 

{tsugite_codex_cli-0.23.1 → tsugite_codex_cli-0.24.0}/tsugite_codex_cli/provider.py

@@ -18,15 +18,9 @@ from tsugite_codex_cli.auth import CodexAuthError, CodexAuthStore
 logger = logging.getLogger(__name__)
 
 API_BASE = "https://chatgpt.com/backend-api/codex"
-_REASONING_LEVELS = ["low", "medium", "high"]
-# gpt-5.5 added xhigh; older codex models don't accept it.
-_REASONING_LEVELS_V5_5 = ["low", "medium", "high", "xhigh"]
-# gpt-5.6 added max; its sol/terra tiers and gpt-6-astra also accept ultra (delegating).
-_REASONING_LEVELS_V5_6 = [*_REASONING_LEVELS_V5_5, "max"]
-_REASONING_LEVELS_V5_6_AGENTIC = [*_REASONING_LEVELS_V5_6, "ultra"]
 
 
-def _codex_model_info(max_input_tokens: int, effort_levels: list[str] = _REASONING_LEVELS) -> ModelInfo:
+def _codex_model_info(max_input_tokens: int, effort_levels: list[str]) -> ModelInfo:
     # Subscription billing means per-token cost is not surfaced here.
     return ModelInfo(
         max_input_tokens=max_input_tokens,
@@ -39,18 +33,22 @@ def _codex_model_info(max_input_tokens: int, effort_levels: list[str] = _REASONI
     )
 
 
-# Context windows are the `context_window` the backend reports for the
-# subscription path, which is smaller than the same model's API tier. Not
-# `max_context_window`, which some slugs report as 1M.
+# --- BEGIN GENERATED MODELS (codex_cli) --- #
+# Auto-generated by scripts/update_model_registry.py, do not edit manually
+# fmt: off
 _CODEX_CLI_MODELS: dict[str, ModelInfo] = {
-    "codex_cli/gpt-5.6-sol": _codex_model_info(272_000, effort_levels=_REASONING_LEVELS_V5_6_AGENTIC),
-    "codex_cli/gpt-5.6-terra": _codex_model_info(272_000, effort_levels=_REASONING_LEVELS_V5_6_AGENTIC),
-    "codex_cli/gpt-5.6-luna": _codex_model_info(272_000, effort_levels=_REASONING_LEVELS_V5_6),
-    "codex_cli/gpt-5.5": _codex_model_info(272_000, effort_levels=_REASONING_LEVELS_V5_5),
-    "codex_cli/gpt-6-astra": _codex_model_info(272_000, effort_levels=_REASONING_LEVELS_V5_6_AGENTIC),
+    "codex_cli/gpt-5.5": _codex_model_info(272_000, ["low", "medium", "high", "xhigh"]),
+    "codex_cli/gpt-5.6-luna": _codex_model_info(272_000, ["low", "medium", "high", "xhigh", "max"]),
+    "codex_cli/gpt-5.6-sol": _codex_model_info(272_000, ["low", "medium", "high", "xhigh", "max", "ultra"]),
+    "codex_cli/gpt-5.6-terra": _codex_model_info(272_000, ["low", "medium", "high", "xhigh", "max", "ultra"]),
+    "codex_cli/gpt-6-astra": _codex_model_info(272_000, ["low", "medium", "high", "xhigh", "max", "ultra"]),
+    "codex_cli/gpt-6-luna": _codex_model_info(272_000, ["low", "medium", "high", "xhigh", "max"]),
+    "codex_cli/gpt-6-sol": _codex_model_info(272_000, ["low", "medium", "high", "xhigh", "max", "ultra"]),
 }
+# fmt: on
+# --- END GENERATED MODELS (codex_cli) --- #
 
-# Used when /models is unreachable; kept in sync with the registry above.
+# Used when /models is unreachable.
 _FALLBACK_MODELS = [k.split("/", 1)[1] for k in _CODEX_CLI_MODELS]
 
 
@@ -335,22 +333,25 @@ class CodexCliProvider:
         # on parallel sessions. The client is released when the process exits.
         pass
 
+    async def fetch_models(self) -> list[dict]:
+        access_token, account_id = await self._auth.get_access_token()
+        client = self._get_client()
+        resp = await client.get(
+            f"{API_BASE}/models?client_version=1.0.0",
+            headers=self._build_headers(access_token, account_id),
+        )
+        resp.raise_for_status()
+        # The Codex backend answers {"models":[{"slug":...}]}, not the OpenAI
+        # {"data":[{"id":...}]} shape. `visibility: hide` marks special-purpose
+        # entries (auto-review, watermark variants) the CLI keeps out of its
+        # own picker, so they stay out of ours too.
+        return [m for m in resp.json()["models"] if isinstance(m, dict) and m.get("visibility") != "hide"]
+
     async def list_models(self) -> list[str]:
         # Auth lookup is inside the try: model-picker UIs should always get a
         # usable list even before `codex login` has been run.
         try:
-            access_token, account_id = await self._auth.get_access_token()
-            client = self._get_client()
-            resp = await client.get(
-                f"{API_BASE}/models?client_version=1.0.0",
-                headers=self._build_headers(access_token, account_id),
-            )
-            resp.raise_for_status()
-            # The Codex backend answers {"models":[{"slug":...}]}, not the OpenAI
-            # {"data":[{"id":...}]} shape. `visibility: hide` marks special-purpose
-            # entries (auto-review, watermark variants) the CLI keeps out of its
-            # own picker, so they stay out of ours too.
-            return [m["slug"] for m in resp.json()["models"] if isinstance(m, dict) and m.get("visibility") != "hide"]
+            return [m["slug"] for m in await self.fetch_models()]
         except (httpx.HTTPError, KeyError, ValueError, CodexAuthError):
             return list(_FALLBACK_MODELS)
 

tsugite_codex_cli-0.24.0/tsugite_codex_cli/usage.py

@@ -0,0 +1,81 @@
+"""ChatGPT subscription limits for Codex as a tsugite quota source.
+
+Read from ``rate_limit`` in the ChatGPT usage endpoint, with the Codex CLI's
+OAuth token. Each window is labeled by its duration, so a window length
+tsugite has never seen still shows up.
+"""
+
+from __future__ import annotations
+
+import logging
+import time
+from datetime import datetime, timezone
+
+import httpx
+
+from tsugite.usage.quota import QuotaReport, QuotaWindow, register_quota_source
+from tsugite.user_agent import set_user_agent_header
+from tsugite_codex_cli.auth import CodexAuthStore
+
+logger = logging.getLogger(__name__)
+
+PROVIDER = "codex_cli"
+LABEL = "Codex"
+USAGE_URL = "https://chatgpt.com/backend-api/wham/usage"
+LIVE_TTL_S = 60
+
+_last_live: tuple[float, dict] | None = None
+
+
+def _live_rate_limit() -> dict:
+    global _last_live
+    if _last_live and time.monotonic() - _last_live[0] < LIVE_TTL_S:
+        return _last_live[1]
+    access_token, account_id = CodexAuthStore().sync_get_or_refresh()
+    headers = {"Authorization": f"Bearer {access_token}", "ChatGPT-Account-ID": account_id}
+    set_user_agent_header(headers)
+    response = httpx.get(USAGE_URL, headers=headers, timeout=10)
+    response.raise_for_status()
+    rate_limit = response.json()["rate_limit"]
+    _last_live = (time.monotonic(), rate_limit)
+    return rate_limit
+
+
+def _label(seconds: int) -> str:
+    if seconds == 7 * 24 * 3600:
+        return "week"
+    if seconds % 3600:
+        return f"{seconds // 60}m"
+    return f"{seconds // 3600}h"
+
+
+def _error(message: str) -> QuotaReport:
+    return QuotaReport(provider=PROVIDER, label=LABEL, error=message)
+
+
+def fetch() -> QuotaReport:
+    try:
+        rate_limit = _live_rate_limit()
+    except Exception as e:
+        # A failed token refresh can quote the auth server's response body.
+        logger.warning("Codex usage unavailable: %s", type(e).__name__)
+        return _error("Codex's usage limits could not be fetched")
+
+    try:
+        windows = tuple(
+            QuotaWindow(
+                key=key,
+                label=_label(window["limit_window_seconds"]),
+                used_pct=float(window["used_percent"]),
+                resets_at=datetime.fromtimestamp(window["reset_at"], timezone.utc).isoformat(),
+            )
+            for key in ("primary", "secondary")
+            if (window := rate_limit.get(f"{key}_window"))
+        )
+    except Exception:
+        return _error("Codex's usage limits are not in a shape tsugite understands")
+
+    return QuotaReport(provider=PROVIDER, label=LABEL, windows=windows)
+
+
+register_quota_source(PROVIDER, LABEL, fetch)