Skip rate-limiter token counting without limits

apply_rate_limiter tokenized the full serialized prompt on every call
even when no request, input or output limit was configured. Return
early in that case; callers already skip output accounting when no
limiter is returned.

Add a regression for missing config, disabled limits, and each enabled
limit; document the accounting contract.

Validation: 115 rate-limiter, streaming, Responses and embedding tests
passed with this change alone on the original base in the Docker
framework runtime. The disabled-limits regression fails on the old
code. Earlier live Codex measurements with a 760 KB prompt showed
about 70 ms dropping to 0 ms in apply_rate_limiter.

AGENTS.md

@@ -21,6 +21,7 @@
 - Native function calls use canonical call content for history and repeat comparison, even when accompanied by commentary. The canonical content passes through the normal history template/masking hook; provider output metadata remains intact.
 - `initialize.py` owns framework initialization.
 - `models.py` owns model-provider configuration and LiteLLM integration. `get_api_key_raw()` reads the stored key or key list for settings/editing; `get_api_key()` selects a runtime key and runs provider extension hooks.
+- `models.apply_rate_limiter` skips token counting and returns no limiter when all request, input, and output limits are unset; any configured limit retains normal accounting.
 - `requirements.txt` includes HTTPX's SOCKS extra so framework clients support SOCKS proxy environment variables before plugin setup.
 - Native agent-turn response callbacks may receive a display-only commentary preview as their full text; legacy callbacks, accumulated model output, usage and native dispatch retain the original text and result metadata.
 - `run_ui.py` is the WebUI entry point.

models.py

@@ -336,7 +336,10 @@ async def apply_rate_limiter(
         Callable[[str, str, int, int], Awaitable[bool]] | None
     ) = None,
 ):
-    if not model_config:
+    # skip token counting of the whole prompt when no limit is set
+    if not model_config or not (
+        model_config.limit_requests or model_config.limit_input or model_config.limit_output
+    ):
         return
     limiter = get_rate_limiter(
         model_config.provider,

tests/test_model_rate_limiter.py

@@ -0,0 +1,32 @@
+import sys
+from pathlib import Path
+
+import pytest
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+import models
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("limit", [None, "none", "requests", "input", "output"])
+async def test_rate_limiter_counts_tokens_only_when_limits_are_configured(monkeypatch, limit):
+    monkeypatch.setattr(models, "rate_limiters", {})
+    counted = []
+    monkeypatch.setattr(models, "approximate_tokens", lambda text: counted.append(text) or 3)
+    config = models.ModelConfig(
+        type=models.ModelType.CHAT, provider="openai", name="test",
+    ) if limit is not None else None
+    if limit in ("requests", "input", "output"):
+        setattr(config, f"limit_{limit}", 10)
+
+    limiter = await models.apply_rate_limiter(config, "prompt")
+
+    if limit in (None, "none"):
+        assert limiter is None
+        assert counted == []
+        assert models.rate_limiters == {}
+    else:
+        assert counted == ["prompt"]
+        assert limiter.limits[limit] == 10
+        assert await limiter.get_total("input") == 3
+        assert await limiter.get_total("requests") == 1