Skip to main content

LangGraph Integration

Use the LangGraph adapter when you want to embed ACM tasks inside existing LangGraph flows while retaining deterministic guards and ledger capture.

Setup​

pnpm add @ddse/acm-adapters @langchain/langgraph

Wrap ACM plan in LangGraph​

import {asLangGraph} from '@ddse/acm-adapters';

const acmGraph = asLangGraph({
  goal,
  context,
  plan,
  capabilityRegistry,
  toolRegistry,
  ledger,
  policy: policyEngine,
  stream: streamSink
});

const result = await acmGraph.execute();

Tips​

  • Provide a MemoryLedger to capture adapter events and forward them to LangGraph monitoring.
  • Use guards inside the plan rather than LangGraph branching to ensure replay bundles remain truthful.
  • When you need resumability, run the plan with the native ACM runtime before invoking LangGraph-specific post-processing.

Combining with other LangGraph nodes​

  1. Execute ACM plan to obtain deterministic outputs and ledger.
  2. Feed outputs into existing LangGraph nodes (e.g., summarisation, user prompts).
  3. Optionally append ledger metadata to downstream logs for a unified timeline.

Limitations​

  • Checkpoint/resume is not available through the adapter (use the ACM runtime directly when required).
  • Streamed updates follow LangGraph conventions; adjust listeners accordingly.

References​