fix: read max_tokens from custom_providers per-model config

agent/agent_init.py

@@ -1255,6 +1255,20 @@ def init_agent(
     # AFTER the custom_providers branch so per-model overrides aren't lost.
     agent._config_context_length = _config_context_length
 
+    # Check custom_providers per-model max_tokens (symmetric to context_length above)
+    if agent.max_tokens is None and _custom_providers:
+        try:
+            from hermes_cli.config import get_custom_provider_max_tokens
+            _cp_max_tokens = get_custom_provider_max_tokens(
+                model=agent.model,
+                base_url=agent.base_url,
+                custom_providers=_custom_providers,
+            )
+            if _cp_max_tokens:
+                agent.max_tokens = int(_cp_max_tokens)
+        except Exception:
+            pass
+
     agent._ensure_lmstudio_runtime_loaded(_config_context_length)
 
 

hermes_cli/config.py

@@ -3169,30 +3169,23 @@ def _append_if_new(entry: Optional[Dict[str, Any]]) -> None:
 
     return compatible
 
-
-def get_custom_provider_context_length(
-    model: str,
-    base_url: str,
-    custom_providers: Optional[List[Dict[str, Any]]] = None,
-    config: Optional[Dict[str, Any]] = None,
-) -> Optional[int]:
-    """Look up a per-model ``context_length`` override from ``custom_providers``.
+def get_custom_provider_model_field(
+    field: str,
+    model: str | None = None,
+    base_url: str | None = None,
+    *,
+    custom_providers: list | None = None,
+    config: dict | None = None,
+    value_type: type = int,
+    positive_only: bool = True,
+) -> int | None:
+    """Look up a per-model field override from ``custom_providers``.
 
     Matches any entry whose ``base_url`` equals ``base_url`` (trailing-slash
-    insensitive) and returns ``custom_providers[i].models.<model>.context_length``
+    insensitive) and returns ``custom_providers[i].models.<model>.<field>``
     if present and valid.  Returns ``None`` when no override applies.
 
-    This is the single source of truth for custom-provider context overrides,
-    used by:
-      * ``AIAgent.__init__`` (startup resolution)
-      * ``AIAgent.switch_model`` (mid-session ``/model`` switch)
-      * ``hermes_cli.model_switch.resolve_display_context_length`` (``/model`` confirmation display)
-      * ``gateway.run._format_session_info`` (``/info`` display)
-      * ``agent.model_metadata.get_model_context_length`` (when custom_providers is threaded through)
-
-    Before this helper existed, the lookup was duplicated in ``run_agent.py``'s
-    startup path only; every other path (notably ``/model`` switch) fell back
-    to the 128K default.  See #15779.
+    Generic helper used for both ``context_length`` and ``max_tokens`` lookups.
     """
     if not model or not base_url:
         return None
@@ -3223,18 +3216,66 @@ def get_custom_provider_context_length(
         model_cfg = models.get(model)
         if not isinstance(model_cfg, dict):
             continue
-        raw_ctx = model_cfg.get("context_length")
-        if raw_ctx is None:
+        raw_val = model_cfg.get(field)
+        if raw_val is None:
             continue
         try:
-            ctx = int(raw_ctx)
+            val = value_type(raw_val)
         except (TypeError, ValueError):
             continue
-        if ctx > 0:
-            return ctx
+        if positive_only and val <= 0:
+            continue
+        return val
     return None
 
 
+def get_custom_provider_context_length(
+    model: str | None = None,
+    base_url: str | None = None,
+    custom_providers: list | None = None,
+    config: dict | None = None,
+) -> int | None:
+    """Look up a per-model ``context_length`` override from ``custom_providers``.
+
+    Matches any entry whose ``base_url`` equals ``base_url`` (trailing-slash
+    insensitive) and returns ``custom_providers[i].models.<model>.context_length``
+    if present and valid.  Returns ``None`` when no override applies.
+
+    This is the single source of truth for custom-provider context overrides,
+    used by:
+      * ``AIAgent.__init__`` (startup resolution)
+      * ``AIAgent.switch_model`` (mid-session ``/model`` switch)
+      * ``hermes_cli.model_switch.resolve_display_context_length`` (``/model`` confirmation display)
+      * ``gateway.run._format_session_info`` (``/info`` display)
+      * ``agent.model_metadata.get_model_context_length`` (when custom_providers is threaded through)
+
+    Before this helper existed, the lookup was duplicated in ``run_agent.py``'s
+    startup path only; every other path (notably ``/model`` switch) fell back
+    to the 128K default.  See #15779.
+    """
+    return get_custom_provider_model_field(
+        "context_length", model, base_url,
+        custom_providers=custom_providers, config=config,
+    )
+
+
+def get_custom_provider_max_tokens(
+    model: str | None = None,
+    base_url: str | None = None,
+    custom_providers: list | None = None,
+    config: dict | None = None,
+) -> int | None:
+    """Look up a per-model ``max_tokens`` override from ``custom_providers``.
+
+    Symmetric to ``get_custom_provider_context_length`` but for the output
+    token limit.  Returns ``None`` when no override applies.
+    """
+    return get_custom_provider_model_field(
+        "max_tokens", model, base_url,
+        custom_providers=custom_providers, config=config,
+    )
+
+
 def check_config_version() -> Tuple[int, int]:
     """
     Check config version.

tests/hermes_cli/test_custom_provider_max_tokens.py

@@ -0,0 +1,158 @@
+"""Regression tests for custom_providers per-model max_tokens resolution.
+
+Covers the fix for #28046 — custom_providers per-model max_tokens was silently
+ignored, always defaulting to 4096.  The fix adds a symmetric lookup to the
+existing context_length mechanism via get_custom_provider_model_field.
+"""
+from __future__ import annotations
+
+from hermes_cli.config import get_custom_provider_max_tokens
+
+
+class TestGetCustomProviderMaxTokens:
+    def test_returns_override_for_matching_entry(self):
+        custom = [
+            {
+                "name": "my-endpoint",
+                "base_url": "https://example.invalid/v1",
+                "models": {"gpt-5.5": {"max_tokens": 16384}},
+            }
+        ]
+        assert (
+            get_custom_provider_max_tokens(
+                "gpt-5.5", "https://example.invalid/v1", custom
+            )
+            == 16384
+        )
+
+    def test_trailing_slash_insensitive(self):
+        custom = [
+            {
+                "base_url": "https://example.invalid/v1/",
+                "models": {"m": {"max_tokens": 8192}},
+            }
+        ]
+        assert (
+            get_custom_provider_max_tokens(
+                "m", "https://example.invalid/v1", custom
+            )
+            == 8192
+        )
+
+    def test_returns_none_when_no_override(self):
+        custom = [
+            {
+                "base_url": "https://other.invalid/v1",
+                "models": {"m": {"max_tokens": 4096}},
+            }
+        ]
+        assert (
+            get_custom_provider_max_tokens(
+                "m", "https://example.invalid/v1", custom
+            )
+            is None
+        )
+
+    def test_returns_none_for_zero_value(self):
+        custom = [
+            {
+                "base_url": "https://example.invalid/v1",
+                "models": {"m": {"max_tokens": 0}},
+            }
+        ]
+        assert (
+            get_custom_provider_max_tokens(
+                "m", "https://example.invalid/v1", custom
+            )
+            is None
+        )
+
+    def test_returns_none_for_negative_value(self):
+        custom = [
+            {
+                "base_url": "https://example.invalid/v1",
+                "models": {"m": {"max_tokens": -100}},
+            }
+        ]
+        assert (
+            get_custom_provider_max_tokens(
+                "m", "https://example.invalid/v1", custom
+            )
+            is None
+        )
+
+    def test_returns_none_for_string_value(self):
+        custom = [
+            {
+                "base_url": "https://example.invalid/v1",
+                "models": {"m": {"max_tokens": "16384"}},
+            }
+        ]
+        # int("16384") succeeds so this should work
+        assert (
+            get_custom_provider_max_tokens(
+                "m", "https://example.invalid/v1", custom
+            )
+            == 16384
+        )
+
+    def test_returns_none_for_non_int_string(self):
+        custom = [
+            {
+                "base_url": "https://example.invalid/v1",
+                "models": {"m": {"max_tokens": "16K"}},
+            }
+        ]
+        assert (
+            get_custom_provider_max_tokens(
+                "m", "https://example.invalid/v1", custom
+            )
+            is None
+        )
+
+    def test_coexists_with_context_length(self):
+        """Both fields can be present in the same model config."""
+        custom = [
+            {
+                "base_url": "https://example.invalid/v1",
+                "models": {
+                    "m": {
+                        "context_length": 256_000,
+                        "max_tokens": 8192,
+                    }
+                },
+            }
+        ]
+        from hermes_cli.config import get_custom_provider_context_length
+
+        assert (
+            get_custom_provider_context_length("m", "https://example.invalid/v1", custom)
+            == 256_000
+        )
+        assert (
+            get_custom_provider_max_tokens("m", "https://example.invalid/v1", custom)
+            == 8192
+        )
+
+    def test_first_matching_entry_wins(self):
+        custom = [
+            {
+                "base_url": "https://example.invalid/v1",
+                "models": {"m": {"max_tokens": 4096}},
+            },
+            {
+                "base_url": "https://example.invalid/v1",
+                "models": {"m": {"max_tokens": 8192}},
+            },
+        ]
+        assert (
+            get_custom_provider_max_tokens(
+                "m", "https://example.invalid/v1", custom
+            )
+            == 4096
+        )
+
+    def test_none_inputs(self):
+        assert get_custom_provider_max_tokens(None, "url", []) is None
+        assert get_custom_provider_max_tokens("m", None, []) is None
+        assert get_custom_provider_max_tokens(None, None, []) is None