diff --git a/openhands/resolver/send_pull_request.py b/openhands/resolver/send_pull_request.py
index 8857602ec1..047592c9cc 100644
--- a/openhands/resolver/send_pull_request.py
+++ b/openhands/resolver/send_pull_request.py
@@ -23,6 +23,7 @@ from openhands.resolver.patching import apply_diff, parse_patch
 from openhands.resolver.resolver_output import ResolverOutput
 from openhands.resolver.utils import identify_token
 from openhands.utils.async_utils import GENERAL_TIMEOUT, call_async_from_sync
+from openhands.utils.environment import get_effective_llm_base_url
 
 
 def apply_patch(repo_dir: str, patch: str) -> None:
@@ -707,10 +708,16 @@ def main() -> None:
     )
 
     api_key = my_args.llm_api_key or os.environ['LLM_API_KEY']
+    model_name = my_args.llm_model or os.environ['LLM_MODEL']
+    base_url = my_args.llm_base_url or os.environ.get('LLM_BASE_URL')
+    resolved_base_url = get_effective_llm_base_url(
+        model_name,
+        base_url,
+    )
     llm_config = LLMConfig(
-        model=my_args.llm_model or os.environ['LLM_MODEL'],
+        model=model_name,
         api_key=SecretStr(api_key) if api_key else None,
-        base_url=my_args.llm_base_url or os.environ.get('LLM_BASE_URL', None),
+        base_url=resolved_base_url,
     )
 
     if not os.path.exists(my_args.output_dir):
diff --git a/openhands/server/routes/manage_conversations.py b/openhands/server/routes/manage_conversations.py
index e3fe4e3c38..140eefa0d5 100644
--- a/openhands/server/routes/manage_conversations.py
+++ b/openhands/server/routes/manage_conversations.py
@@ -91,6 +91,7 @@ from openhands.storage.locations import get_experiment_config_filename
 from openhands.storage.settings.settings_store import SettingsStore
 from openhands.utils.async_utils import wait_all
 from openhands.utils.conversation_summary import get_default_conversation_title
+from openhands.utils.environment import get_effective_llm_base_url
 
 app = APIRouter(prefix='/api', dependencies=get_dependencies())
 app_conversation_service_dependency = depends_app_conversation_service()
@@ -545,10 +546,15 @@ async def get_prompt(
         # placeholder for error handling
         raise ValueError('Settings not found')
 
+    settings_base_url = settings.llm_base_url
+    effective_base_url = get_effective_llm_base_url(
+        settings.llm_model,
+        settings_base_url,
+    )
     llm_config = LLMConfig(
         model=settings.llm_model or '',
         api_key=settings.llm_api_key,
-        base_url=settings.llm_base_url,
+        base_url=effective_base_url,
     )
 
     prompt_template = generate_prompt_template(stringified_events)
diff --git a/openhands/utils/conversation_summary.py b/openhands/utils/conversation_summary.py
index f281fe1f5f..999a5eabea 100644
--- a/openhands/utils/conversation_summary.py
+++ b/openhands/utils/conversation_summary.py
@@ -10,6 +10,7 @@ from openhands.events.event_store import EventStore
 from openhands.llm.llm_registry import LLMRegistry
 from openhands.storage.data_models.settings import Settings
 from openhands.storage.files import FileStore
+from openhands.utils.environment import get_effective_llm_base_url
 
 
 async def generate_conversation_title(
@@ -114,10 +115,15 @@ async def auto_generate_title(
             try:
                 if settings and settings.llm_model:
                     # Create LLM config from settings
+                    settings_base_url = settings.llm_base_url
+                    effective_base_url = get_effective_llm_base_url(
+                        settings.llm_model,
+                        settings_base_url,
+                    )
                     llm_config = LLMConfig(
                         model=settings.llm_model,
                         api_key=settings.llm_api_key,
-                        base_url=settings.llm_base_url,
+                        base_url=effective_base_url,
                     )
 
                     # Try to generate title using LLM
diff --git a/openhands/utils/environment.py b/openhands/utils/environment.py
new file mode 100644
index 0000000000..140c1d385d
--- /dev/null
+++ b/openhands/utils/environment.py
@@ -0,0 +1,58 @@
+from __future__ import annotations
+
+import os
+from functools import lru_cache
+from pathlib import Path
+
+LEMONADE_DOCKER_BASE_URL = 'http://host.docker.internal:8000/api/v1/'
+_LEMONADE_PROVIDER_NAME = 'lemonade'
+_LEMONADE_MODEL_PREFIX = 'lemonade/'
+
+
+@lru_cache(maxsize=1)
+def is_running_in_docker() -> bool:
+    """Best-effort detection for Docker containers."""
+    docker_env_markers = (
+        Path('/.dockerenv'),
+        Path('/run/.containerenv'),
+    )
+    if any(marker.exists() for marker in docker_env_markers):
+        return True
+
+    if os.environ.get('DOCKER_CONTAINER') == 'true':
+        return True
+
+    try:
+        with Path('/proc/self/cgroup').open('r', encoding='utf-8') as cgroup_file:
+            for line in cgroup_file:
+                if any(token in line for token in ('docker', 'containerd', 'kubepods')):
+                    return True
+    except FileNotFoundError:
+        pass
+
+    return False
+
+
+def is_lemonade_provider(
+    model: str | None,
+    custom_provider: str | None = None,
+) -> bool:
+    provider = (custom_provider or '').strip().lower()
+    if provider == _LEMONADE_PROVIDER_NAME:
+        return True
+    return (model or '').startswith(_LEMONADE_MODEL_PREFIX)
+
+
+def get_effective_llm_base_url(
+    model: str | None,
+    base_url: str | None,
+    custom_provider: str | None = None,
+) -> str | None:
+    """Return the runtime LLM base URL with provider-specific overrides."""
+    if (
+        base_url in (None, '')
+        and is_lemonade_provider(model, custom_provider)
+        and is_running_in_docker()
+    ):
+        return LEMONADE_DOCKER_BASE_URL
+    return base_url
diff --git a/openhands/utils/utils.py b/openhands/utils/utils.py
index 8ea3b96b54..77ae3bae10 100644
--- a/openhands/utils/utils.py
+++ b/openhands/utils/utils.py
@@ -1,3 +1,4 @@
+import os
 from copy import deepcopy
 
 from openhands.core.config.openhands_config import OpenHandsConfig
@@ -5,6 +6,7 @@ from openhands.llm.llm_registry import LLMRegistry
 from openhands.server.services.conversation_stats import ConversationStats
 from openhands.storage import get_file_store
 from openhands.storage.data_models.settings import Settings
+from openhands.utils.environment import get_effective_llm_base_url
 
 
 def setup_llm_config(config: OpenHandsConfig, settings: Settings) -> OpenHandsConfig:
@@ -14,7 +16,19 @@ def setup_llm_config(config: OpenHandsConfig, settings: Settings) -> OpenHandsCo
     llm_config = config.get_llm_config()
     llm_config.model = settings.llm_model or ''
     llm_config.api_key = settings.llm_api_key
-    llm_config.base_url = settings.llm_base_url
+    env_base_url = os.environ.get('LLM_BASE_URL')
+    settings_base_url = settings.llm_base_url
+
+    # Use env_base_url if available, otherwise fall back to settings_base_url
+    base_url_to_use = (
+        env_base_url if env_base_url not in (None, '') else settings_base_url
+    )
+
+    llm_config.base_url = get_effective_llm_base_url(
+        llm_config.model,
+        base_url_to_use,
+        llm_config.custom_llm_provider,
+    )
     config.set_llm_config(llm_config)
     return config
 
diff --git a/tests/unit/utils/test_environment.py b/tests/unit/utils/test_environment.py
new file mode 100644
index 0000000000..5c9bdc4111
--- /dev/null
+++ b/tests/unit/utils/test_environment.py
@@ -0,0 +1,32 @@
+import pytest
+
+from openhands.utils import environment
+
+
+@pytest.fixture(autouse=True)
+def clear_docker_cache():
+    if hasattr(environment.is_running_in_docker, 'cache_clear'):
+        environment.is_running_in_docker.cache_clear()
+    yield
+    if hasattr(environment.is_running_in_docker, 'cache_clear'):
+        environment.is_running_in_docker.cache_clear()
+
+
+def test_get_effective_base_url_lemonade_in_docker(monkeypatch):
+    monkeypatch.setattr(environment, 'is_running_in_docker', lambda: True)
+    result = environment.get_effective_llm_base_url('lemonade/example', None)
+    assert result == environment.LEMONADE_DOCKER_BASE_URL
+
+
+def test_get_effective_base_url_lemonade_outside_docker(monkeypatch):
+    monkeypatch.setattr(environment, 'is_running_in_docker', lambda: False)
+    base_url = 'http://localhost:8000/api/v1/'
+    result = environment.get_effective_llm_base_url('lemonade/example', base_url)
+    assert result == base_url
+
+
+def test_get_effective_base_url_non_lemonade(monkeypatch):
+    monkeypatch.setattr(environment, 'is_running_in_docker', lambda: True)
+    base_url = 'https://api.example.com'
+    result = environment.get_effective_llm_base_url('openai/gpt-4', base_url)
+    assert result == base_url