Context-M LangChain Adapter

Drop-in replacement for ConversationBufferMemory that routes through Context-M's REST API. Gives any LangChain agent access to Context-M's bi-temporal Trace, cryptographic provenance, Memory Git (branch/merge/diff/blame), ZK-lite proofs, federation-ready schema export, and the mu=0 ingest protocol (zero LLM calls during save).

Quickstart

from context_m_langchain import ContextMMemory
memory = ContextMMemory(user_id='alice')
memory.save_context(
    {'input': 'Hi I\'m Alice'},
    {'output': 'Nice to meet you, Alice!'})
memory.load_memory_variables(
    {'input': 'what\'s my name?'})

Configuration

Environment variables:

• CONTEXT_M_REST_URL — default http://localhost:8900
• CONTEXT_M_API_KEY — default empty (set if REST server requires RBAC)
• CONTEXT_M_USER_ID — default "default"

What's new in v0.2.0 (2026-08-28)

• Timeout raised to 10s default (was 5s) for large-context search.
• intent field surfaced in load_memory_variables output.
• save_context streams through the unmess pipeline by default.