LangGraph Integration (Adapter Pattern)

This document provides a minimal integration pattern for LangGraph-style agent flows with Aionis.

Flow Mapping

1. recall_text before planning (retrieve compact context).
2. tools/select during planning (apply active/shadow rules).
3. execute tool / agent step.
4. write + tools/feedback after step completion (persist outcome).

Runtime Requirements

1. Aionis API reachable (AIONIS_BASE_URL or http://localhost:${PORT}).
2. If auth is enabled:
   • API_KEY (or PERF_API_KEY) for MEMORY_AUTH_MODE=api_key.
   • AUTH_BEARER (or PERF_AUTH_BEARER) for MEMORY_AUTH_MODE=jwt.
3. Python runtime with local SDK source available in this repo.

Smoke Command

cd /Users/lucio/Desktop/Aionis
set -a; source .env; set +a
bash examples/langgraph_adapter_smoke.sh

Expected output (JSON):

1. ok: true
2. project: "langgraph_style"
3. steps.recall_text / steps.tools_select / steps.write / steps.tools_feedback present

Reference Script

1. /Users/lucio/Desktop/Aionis/examples/langgraph_adapter_smoke.sh
2. /Users/lucio/Desktop/Aionis/src/dev/langgraph-adapter-smoke.py