fix(cli): honor custom context lengths in model switch display

cli.py

@@ -5455,7 +5455,12 @@ def _compute_model_picker_viewport(
         scroll_offset = max(0, min(scroll_offset, n - visible))
         return scroll_offset, visible
 
-    def _apply_model_switch_result(self, result, persist_global: bool) -> None:
+    def _apply_model_switch_result(
+        self,
+        result,
+        persist_global: bool,
+        custom_providers: list | None = None,
+    ) -> None:
         if not result.success:
             _cprint(f"  ✗ {result.error_message}")
             return
@@ -5507,7 +5512,12 @@ def _apply_model_switch_result(self, result, persist_global: bool) -> None:
                 base_url=result.base_url or self.base_url or "",
                 api_key=result.api_key or self.api_key or "",
                 model_info=mi,
-                config_context_length=getattr(self.agent, "_config_context_length", None) if self.agent else None,
+                custom_providers=custom_providers,
+                config_context_length=(
+                    getattr(self.agent, "_config_context_length", None)
+                    if self.agent
+                    else None
+                ),
             )
             if ctx:
                 _cprint(f"    Context: {ctx:,} tokens")
@@ -5595,7 +5605,11 @@ def _handle_model_picker_selection(self, persist_global: bool = False) -> None:
                     custom_providers=state.get("custom_provs"),
                 )
                 self._close_model_picker()
-                self._apply_model_switch_result(result, persist_global)
+                self._apply_model_switch_result(
+                    result,
+                    persist_global,
+                    custom_providers=state.get("custom_provs"),
+                )
                 return
             self._close_model_picker()
 
@@ -5734,7 +5748,12 @@ def _handle_model_switch(self, cmd_original: str):
             base_url=result.base_url or self.base_url or "",
             api_key=result.api_key or self.api_key or "",
             model_info=mi,
-            config_context_length=getattr(self.agent, "_config_context_length", None) if self.agent else None,
+            custom_providers=custom_provs,
+            config_context_length=(
+                getattr(self.agent, "_config_context_length", None)
+                if self.agent
+                else None
+            ),
         )
         if ctx:
             _cprint(f"    Context: {ctx:,} tokens")

hermes_cli/model_switch.py

@@ -558,6 +558,26 @@ def resolve_display_context_length(
     Prefer the provider-aware value; fall back to ``model_info.context_window``
     only if the resolver returns nothing.
     """
+    if (
+        config_context_length is not None
+        and isinstance(config_context_length, int)
+        and config_context_length > 0
+    ):
+        return config_context_length
+
+    if custom_providers and base_url and model:
+        try:
+            from hermes_cli.config import get_custom_provider_context_length
+            ctx = get_custom_provider_context_length(
+                model=model,
+                base_url=base_url,
+                custom_providers=custom_providers,
+            )
+            if ctx:
+                return int(ctx)
+        except Exception:
+            pass
+
     try:
         from agent.model_metadata import get_model_context_length
         ctx = get_model_context_length(

tests/hermes_cli/test_apply_model_switch_result_context.py

@@ -43,14 +43,19 @@ class _StubCLI:
     _pending_model_switch_note = ""
 
 
-def _run_display(monkeypatch, result):
+def _run_display(monkeypatch, result, custom_providers=None):
     import cli as cli_mod
 
     captured: list[str] = []
     monkeypatch.setattr(cli_mod, "_cprint", lambda s, *a, **k: captured.append(str(s)))
     # Avoid writing to ~/.hermes/config.yaml during the test.
     monkeypatch.setattr(cli_mod, "save_config_value", lambda *a, **k: None)
-    cli_mod.HermesCLI._apply_model_switch_result(_StubCLI(), result, False)
+    cli_mod.HermesCLI._apply_model_switch_result(
+        _StubCLI(),
+        result,
+        False,
+        custom_providers=custom_providers,
+    )
     return captured
 
 
@@ -150,3 +155,35 @@ def test_picker_path_falls_back_to_model_info_when_resolver_empty(monkeypatch):
     assert "1,050,000" in ctx_line, (
         f"resolver-empty path should fall back to ModelInfo, got: {ctx_line!r}"
     )
+
+
+def test_picker_path_honors_custom_provider_context(monkeypatch):
+    """Picker confirmation must pass custom_providers into the shared resolver."""
+    result = ModelSwitchResult(
+        success=True,
+        new_model="local-model",
+        target_provider="custom:local",
+        provider_changed=True,
+        api_key="",
+        base_url="http://localhost:1234/v1",
+        api_mode="chat_completions",
+        warning_message="",
+        provider_label="Local",
+        resolved_via_alias=False,
+        capabilities=None,
+        model_info=_FakeModelInfo(),
+        is_global=False,
+    )
+    custom_providers = [
+        {
+            "name": "Local",
+            "base_url": "http://localhost:1234/v1",
+            "models": {"local-model": {"context_length": 900_000}},
+        }
+    ]
+
+    lines = _run_display(monkeypatch, result, custom_providers=custom_providers)
+
+    ctx_line = next((l for l in lines if "Context:" in l), "")
+    assert "900,000" in ctx_line
+    assert "1,050,000" not in ctx_line

tests/hermes_cli/test_model_switch_context_display.py

@@ -89,6 +89,17 @@ def test_prefers_resolver_even_when_model_info_has_larger_value(self):
             )
         assert ctx == 128_000
 
+    def test_config_context_length_overrides_resolver_and_model_info(self):
+        fake_mi = _FakeModelInfo(2_000_000)
+        ctx = resolve_display_context_length(
+            "custom-model",
+            "custom",
+            base_url="https://example.invalid/v1",
+            model_info=fake_mi,
+            config_context_length=750_000,
+        )
+        assert ctx == 750_000
+
     def test_custom_providers_override_honored(self):
         """Regression for #15779: /model switch onto a custom provider must
         surface the configured per-model context_length, not the 128K/256K
@@ -146,3 +157,23 @@ def test_custom_providers_trailing_slash_insensitive(self):
                 custom_providers=custom_provs,
             )
         assert ctx == 400_000
+
+    def test_non_matching_custom_provider_does_not_override(self):
+        fake_mi = _FakeModelInfo(1_050_000)
+        custom_provs = [
+            {
+                "base_url": "https://example.invalid/v1",
+                "models": {"m": {"context_length": 400_000}},
+            }
+        ]
+        with patch(
+            "agent.model_metadata.get_model_context_length", return_value=256_000
+        ):
+            ctx = resolve_display_context_length(
+                "m",
+                "custom",
+                base_url="https://other.invalid/v1",
+                model_info=fake_mi,
+                custom_providers=custom_provs,
+            )
+        assert ctx == 256_000