fix: bound unset LLM request timeout (600s) instead of litellm's 6000s default

src/praisonai-agents/praisonaiagents/llm/llm.py

@@ -43,6 +43,14 @@ def mark_success(self, profile: "AuthProfile") -> None: ...
 # Bounded shrink-and-retry steps after a provider context-overflow rejection.
 _CONTEXT_OVERFLOW_SHRINK_STEPS = 3
 
+# Default per-request timeout (seconds) when ``timeout`` is not passed. Without
+# it litellm falls back to its 6000s ceiling, so a provider that accepts the
+# connection but never answers blocks the turn for ~100 minutes per attempt.
+# 600s matches the OpenAI SDK default used by the bare-model-name path and is
+# a read-gap bound for streams. Pass ``timeout=None`` (or 0) to opt out.
+_DEFAULT_LLM_TIMEOUT_S = 600
+_TIMEOUT_UNSET = object()
+
 
 # ``litellm.success_callback``/``_async_success_callback``/``callbacks`` are
 # process-global lists shared by every LLM instance. Guard the read-modify-write
@@ -493,7 +501,7 @@ def _log_llm_config(self, method_name: str, **config):
     def __init__(
         self,
         model: str,
-        timeout: Optional[int] = None,
+        timeout: Optional[int] = _TIMEOUT_UNSET,
         temperature: Optional[float] = None,
         top_p: Optional[float] = None,
         n: Optional[int] = None,
@@ -535,7 +543,7 @@ def __init__(
                 )
 
         self.model = model
-        self.timeout = timeout
+        self.timeout = _DEFAULT_LLM_TIMEOUT_S if timeout is _TIMEOUT_UNSET else timeout
         self.temperature = temperature
         self.top_p = top_p
         self.n = n

src/praisonai-agents/tests/unit/llm/test_default_request_timeout.py

@@ -0,0 +1,32 @@
+"""An unset LLM timeout is bounded; an explicit None/0 opts out (Issue #5968)."""
+
+import pytest
+
+from praisonaiagents.llm.llm import LLM, _DEFAULT_LLM_TIMEOUT_S
+
+
+def _params(llm):
+    return (
+        llm._build_completion_params(),
+        llm._build_responses_params([{'role': 'user', 'content': 'hi'}]),
+    )
+
+
+def test_unset_timeout_uses_finite_default():
+    llm = LLM(model='gpt-4o-mini')
+    assert llm.timeout == _DEFAULT_LLM_TIMEOUT_S
+    for params in _params(llm):
+        assert params['timeout'] == _DEFAULT_LLM_TIMEOUT_S
+
+
+@pytest.mark.parametrize('value', [None, 0])
+def test_explicit_opt_out_omits_timeout(value):
+    llm = LLM(model='gpt-4o-mini', timeout=value)
+    for params in _params(llm):
+        assert 'timeout' not in params
+
+
+def test_explicit_timeout_is_forwarded():
+    llm = LLM(model='gpt-4o-mini', timeout=30)
+    for params in _params(llm):
+        assert params['timeout'] == 30

src/praisonai/praisonai/_dev/parity/signatures/waivers.yaml

@@ -105,6 +105,12 @@ waivers:
       `transfer_to_<lowercased, spaces-to-underscores name>` exactly (it previously said `handoff_to_<raw
       name>`, which could contain a space and be rejected by the provider). Only the moment of derivation
       differs.
+  LLM.__init__.timeout:
+    owner: praisonai-ts
+    reason: Same effective bound, applied by a different layer. Python's sentinel resolves an unset timeout
+      to 600s in the body (litellm would otherwise wait 6000s); TS leaves `timeout` undefined and the OpenAI
+      Node SDK applies its own 600s (10 minute) default. An explicit value is forwarded on both sides.
+    issue: '#5968'
   PraisonAIError.__init__.run_id:
     owner: praisonai-ts
     reason: 'Both sides mint a UUID at construction -- Python''s `self.run_id = run_id or str(uuid.uuid4())`