langgraph_outer_ainl_core

AINL as deterministic core inside LangGraph

This folder shows a hybrid layout: LangGraph owns orchestration (checkpointing, multi-agent routing, optional LLM nodes), while AINL holds the compact, deterministic policy graph (no recurring tokens for the AINL-authored portion at runtime).

Files

• monitoring_escalation.ainl — example workflow (strict-safe): threshold diff + summary string.
• monitoring_escalation_langgraph.py — emitted wrapper (regenerate with the command below).

Regenerate the LangGraph wrapper

From the repository root:

python3 scripts/validate_ainl.py \
  examples/hybrid/langgraph_outer_ainl_core/monitoring_escalation.ainl \
  --emit langgraph \
  -o examples/hybrid/langgraph_outer_ainl_core/monitoring_escalation_langgraph.py

Run the emitted graph

Requires langgraph (pip install langgraph). Run from repo root so runtime/ and adapters/ resolve:

python3 examples/hybrid/langgraph_outer_ainl_core/monitoring_escalation_langgraph.py

The default graph is START → ainl_core → END, where ainl_core calls run_ainl_graph() with the embedded IR. Pass initial frame variables via LangGraph state under ainl_frame (e.g. app.invoke({"ainl_frame": {"metric_value": 100}}) after you extend the AINL program to read those vars).

Learn more

• docs/hybrid_langgraph.md