From e8a8928e56ebbb33f29abcba099cef3a4152614c Mon Sep 17 00:00:00 2001
From: webtecnica <75556242+webtecnica@users.noreply.github.com>
Date: Mon, 6 Jul 2026 15:27:34 -0300
Subject: [PATCH] fix(agent): include max_tokens in
 _resolve_runtime_agent_kwargs_for_provider return dict (#59763)

---
 gateway/run.py | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

diff --git a/gateway/run.py b/gateway/run.py
index 8256c283d4cf6..d6c0dafe4f570 100644
--- a/gateway/run.py
+++ b/gateway/run.py
@@ -1872,11 +1872,32 @@ def _resolve_runtime_agent_kwargs_for_provider(provider: str) -> dict:
     from hermes_cli.runtime_provider import (
         resolve_runtime_provider,
         format_runtime_provider_error,
+        _get_model_config,
     )
     try:
         runtime = resolve_runtime_provider(requested=provider)
     except Exception as exc:
         raise RuntimeError(format_runtime_provider_error(exc)) from exc
+
+    max_tokens = None
+    _env_mt = os.environ.get("HERMES_MAX_TOKENS")
+    if _env_mt:
+        try:
+            max_tokens = int(_env_mt)
+        except (ValueError, TypeError):
+            max_tokens = None
+    elif isinstance(_get_model_config(), dict):
+        mt = _get_model_config().get("max_tokens")
+        if isinstance(mt, int):
+            max_tokens = mt
+    # Fall back to a per-provider output cap (custom_providers max_output_tokens)
+    # only when the documented global model.max_tokens isn't set, so the global
+    # key always wins.
+    if max_tokens is None:
+        _runtime_mot = runtime.get("max_output_tokens")
+        if isinstance(_runtime_mot, int) and _runtime_mot > 0:
+            max_tokens = _runtime_mot
+
     return {
         "api_key": runtime.get("api_key"),
         "base_url": runtime.get("base_url"),
@@ -1885,6 +1906,7 @@ def _resolve_runtime_agent_kwargs_for_provider(provider: str) -> dict:
         "command": runtime.get("command"),
         "args": list(runtime.get("args") or []),
         "credential_pool": runtime.get("credential_pool"),
+        "max_tokens": max_tokens,
     }