Source trail

MarkAICode — LangGraph Memory Patterns

Public MindPattern findings, entities, and graph evidence that cite this source.

Findings

All-time hits

High value

Last seen

2026-03-22

Connected entities

RadixAttention SGLang vLLM Markaicode — Production Multi-Agent System with LangGraph LangGraph Production: Pydantic State Machines with Three-Tier Error Recovery and LangSmith Tracing LangGraph Pydantic

Related findings

2026-03-22 / SKILLSLangGraph Memory Scoping: Thread-Level Short-Term vs. Cross-Thread Long-Term State for Precise Multi-Agent Context ControlLangGraph's memory architecture formally separates short-term memory (state persisted within a single conversation thread via in-memory storage or checkpoints) from long-term memory (facts, preferences, and summaries stored across threads and restarts), enabling precise control over what context survives agent session boundaries. In multi-agent systems, shared graph state serves as working memory for active coordination while cross-thread storage provides the persistent knowledge layer — the separation prevents context bleed between independent agent runs and reduces hallucination from stale in-thread state.
2026-03-19 / SKILLSSGLang RadixAttention: 6.4x Inference Throughput via Shared KV Cache Prefix Reuse for Agentic WorkloadsSGLang's RadixAttention stores the KV cache of both prompts and generated tokens in a radix tree, enabling automatic cache reuse whenever multiple requests share a common prefix (e.g., the same system prompt, few-shot examples, or tool definitions). For agentic pipelines where every subagent call shares a long system context, this eliminates redundant computation across requests. Benchmarks show 6.4x higher throughput vs. baseline inference; February 2026 results show 25x improvement on NVIDIA GB300 NVL72. Drop-in replacement for vLLM via the OpenAI-compatible server.
2026-03-16 / SKILLSLangGraph Production: Pydantic State Machines with Three-Tier Error Recovery and LangSmith Tracing

Open latest cited source