feat: per-provider max_tokens via custom_providers models.<model>.max_tokens

agent/agent_init.py

@@ -1165,6 +1165,23 @@ def init_agent(
                 )
     agent._session_init_model_config["max_tokens"] = agent.max_tokens
 
+    # If max_tokens is still unset, check custom_providers for a per-provider
+    # override (e.g. custom_providers[].models.<model>.max_tokens).
+    # This allows provider-scoped output-token caps without the global
+    # model.max_tokens affecting fallback providers.
+    if agent.max_tokens is None:
+        try:
+            from hermes_cli.config import get_compatible_custom_providers, get_custom_provider_max_tokens
+            _cp_max_tokens = get_custom_provider_max_tokens(
+                model=agent.model,
+                base_url=agent.base_url,
+                custom_providers=get_compatible_custom_providers(_agent_cfg),
+            )
+            if _cp_max_tokens is not None:
+                agent.max_tokens = int(_cp_max_tokens)
+        except Exception:
+            pass
+
     # Read explicit context_length override from model config
     if isinstance(_model_cfg, dict):
         _config_context_length = _model_cfg.get("context_length")

hermes_cli/config.py

@@ -3000,7 +3000,7 @@ def _normalize_custom_provider_entry(
     _KNOWN_KEYS = {
         "name", "api", "url", "base_url", "api_key", "key_env", "api_key_env",
         "api_mode", "transport", "model", "default_model", "models",
-        "context_length", "rate_limit_delay",
+        "context_length", "max_tokens", "rate_limit_delay",
         "request_timeout_seconds", "stale_timeout_seconds",
         "discover_models",
     }
@@ -3229,6 +3229,66 @@ def get_custom_provider_context_length(
     return None
 
 
+def get_custom_provider_max_tokens(
+    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 ``max_tokens`` override from ``custom_providers``.
+
+    Matches any entry whose ``base_url`` equals ``base_url`` (trailing-slash
+    insensitive) and returns ``custom_providers[i].models.<model>.max_tokens``
+    if present and valid.  Returns ``None`` when no override applies.
+
+    Mirrors ``get_custom_provider_context_length`` — the same pattern for
+    output-token caps that already exists for context windows.
+
+    Used by:
+      * ``AIAgent.__init__`` (startup resolution, after the global
+        ``model.max_tokens`` fallback)
+    """
+    if not model or not base_url:
+        return None
+    if custom_providers is None:
+        try:
+            custom_providers = get_compatible_custom_providers(config)
+        except Exception:
+            if config is None:
+                return None
+            raw = config.get("custom_providers")
+            custom_providers = raw if isinstance(raw, list) else []
+    if not isinstance(custom_providers, list):
+        return None
+
+    target_url = (base_url or "").rstrip("/")
+    if not target_url:
+        return None
+
+    for entry in custom_providers:
+        if not isinstance(entry, dict):
+            continue
+        entry_url = (entry.get("base_url") or "").rstrip("/")
+        if not entry_url or entry_url != target_url:
+            continue
+        models = entry.get("models")
+        if not isinstance(models, dict):
+            continue
+        model_cfg = models.get(model)
+        if not isinstance(model_cfg, dict):
+            continue
+        raw_mt = model_cfg.get("max_tokens")
+        if raw_mt is None:
+            continue
+        try:
+            mt = int(raw_mt)
+        except (TypeError, ValueError):
+            continue
+        if mt > 0:
+            return mt
+    return None
+
+
 def check_config_version() -> Tuple[int, int]:
     """
     Check config version.