Skip to content

Free, open-source AI video generator for Mac. Run LTX-2.5 and MiniMax H3 (Hailuo) locally on Apple Silicon: text-to-video and image-to-video with sound, lip-sync, character LoRA training, songs from lyrics (YuE2) and image editing (Qwen-Image-Edit). One-click install via Pinokio. Offline: no cloud, no API key, no NVIDIA GPU.

fix(agent/smol): bump LocalPythonExecutor per-step timeout to 30 min

mrbizarro303cd64fb125
diff --git a/agent/runtime_smol.py b/agent/runtime_smol.py
index 38c4169..30aa989 100644
--- a/agent/runtime_smol.py
+++ b/agent/runtime_smol.py
@@ -453,6 +453,13 @@ def run_turn(session: Session, user_message: str | None,
instructions += "\n\n# Prior conversation in this session\n\n" + prior
# 4. Construct the CodeAgent.
+ #
+ # `executor_kwargs={"timeout_seconds": ...}` bumps LocalPythonExecutor's
+ # per-step Python execution cap from its 30 s default. Our tools are
+ # SLOW (generate_shot_images ~2-4 min for an mflux Qwen-Edit batch,
+ # wait_for_shot can poll for tens of minutes). 30 s would kill nearly
+ # every meaningful turn. The cap is on the executor, not on the LLM
+ # request — the LLM still has cfg.max_tokens room.
try:
smol_model = _build_smol_model(session.engine_config)
smol_tools = _build_smol_tools(panel_ops, session.tool_state, on_event=emit)
@@ -463,6 +470,7 @@ def run_turn(session: Session, user_message: str | None,
additional_authorized_imports=[], # NO imports — only our tools
max_steps=max_steps,
verbosity_level=0,
+ executor_kwargs={"timeout_seconds": 1800}, # 30 min per step
)
except Exception as e: # noqa: BLE001
err_msg = f"smolagents init failed: {type(e).__name__}: {e}"