[vitalii-dynamiq]

Draft. Companion to the runtime PR that moves Python code execution into the Dynamiq runtime's OS-level sandbox. Keep this draft until that sandbox is enabled in the runtime (see that PR's rollout note), so Python workflows keep running when the runtime bumps to this version.

What

Replaces the RestrictedPython code-execution surface in the open-source package:

• Removes dynamiq.nodes.tools.Python and dynamiq.nodes.tools.PythonCodeExecutor (the RestrictedPython nodes) and the RestrictedPython dependency (pyproject.toml + uv.lock).
• Reimplements dynamiq.evaluations.PythonEvaluator on the Monty interpreter instead of RestrictedPython, keeping its public interface, so custom evaluation metrics keep working on a real isolation boundary.
• Updates GraphOrchestrator conditional edges to accept a PythonMonty node in place of the removed Python node.
• Keeps PythonInputSchema (reused by PythonMonty and the runtime sandbox node) in a slimmed nodes/tools/python.py. PythonMonty is otherwise unchanged.

Why

RestrictedPython shares the interpreter with the code it tries to restrict, so it is not a real boundary for untrusted code. Open-source dynamiq now runs Python code through PythonMonty (a separate Rust interpreter) or a remote sandbox (E2B, Daytona); on the platform, the Dynamiq runtime maps the removed node types onto its OS-level sandbox.

Tests and examples

Every test and example that used the Python node as a helper is migrated:

• To PythonMonty where the code is a simple computation (the common case).
• To FunctionTool where a test needs real in-process behaviour the Monty subset does not provide — blocking (the cancellation suites' sleeping tool) or exact exception text.
• The one test of a now-removed feature (file-store injection into the code executor) is dropped.

The full suite passes; the only failures in a network-restricted run are pre-existing tiktoken-download splitter tests, unrelated to this change.

Breaking change

dynamiq.nodes.tools.Python and dynamiq.nodes.tools.PythonCodeExecutor are removed and RestrictedPython is no longer a dependency. PythonEvaluator now runs user code in the Monty interpreter, which supports a subset of Python and no third-party libraries, so a metric that relied on RestrictedPython-only behaviour must be adjusted.

[github-actions]

Coverage Report •

File	Stmts	Miss	Cover	Missing
dynamiq/evaluations
python_evaluator.py	100	12	88%	48, 114–115, 129–131, 153, 157, 215, 244–246
dynamiq/nodes/agents
base.py	1753	277	84%	161, 167, 265, 281, 289, 292, 451, 541, 561, 631, 633, 635, 676–678, 680–685, 690–692, 716–717, 726, 735, 749, 755–756, 767, 783, 785, 787, 789, 832–834, 854–855, 941–942, 944, 947, 975, 980–981, 984, 987, 1038–1039, 1043–1046, 1055–1056, 1076–1077, 1096–1098, 1109–1111, 1158, 1169–1170, 1388, 1391, 1395–1396, 1420–1436, 1460, 1464, 1507–1508, 1512–1513, 1569, 1606, 1617, 1641–1644, 1646, 1648, 1650–1652, 1654–1655, 1662, 1664–1665, 1669–1671, 1677, 1694, 1700, 1703–1705, 1733–1734, 1809–1810, 1816–1818, 1822–1824, 1843–1844, 1869–1870, 1886–1887, 1895, 1904, 1908, 2082, 2118–2120, 2170–2171, 2191, 2198–2199, 2204, 2256, 2346, 2527–2530, 2533, 2538, 2545, 2551–2552, 2554–2555, 2562, 2572–2574, 2587, 2615–2616, 2624–2625, 2629–2630, 2637, 2644, 2668, 2672–2673, 2702, 2715–2720, 2731–2734, 2736–2744, 2751, 2753, 2772, 2791–2792, 2797, 2803, 2847, 2861, 2870, 2877, 2889–2890, 2892–2893, 2897–2898, 2900, 2902, 2920, 2923, 2932–2933, 2935–2936, 2941–2942, 2947–2948, 2950–2951, 2956–2957, 2960, 3082, 3137, 3144, 3184–3185, 3287, 3321–3323, 3328–3329, 3342–3344, 3352–3353, 3357, 3370, 3374, 3380–3382, 3384, 3386, 3388–3391, 3393, 3399, 3402, 3404, 3408, 3411, 3413, 3417, 3420–3422, 3429, 3436, 3439–3440
dynamiq/nodes/agents/orchestrators
graph.py	191	55	71%	79, 82–83, 101, 109–112, 128, 136, 141, 163–164, 166–167, 195, 219–220, 226, 246, 256, 258–263, 265–266, 270–271, 279–281, 283–284, 286–287, 307, 318, 323, 327, 333, 349, 352–353, 358, 368, 391, 393, 404, 428–431
graph_state.py	148	48	68%	63, 79, 99–100, 115, 126, 130, 144–148, 150, 152–154, 185–188, 199, 201–206, 208–209, 212–214, 216, 224–229, 231–232, 242, 244–246, 252, 257, 259
examples/components/core/checkpoints
loan_cooling_off.py	57	22	61%	85–87, 89–92, 94–96, 98–103, 105, 109–112, 116
TOTAL	52921	10295	81%

Tests	Skipped	Failures	Errors	Time
7738	1 💤	0 ❌	0 🔥	5m 1s ⏱️