llm-memory-and-context-engineering-module

context windowThe finite token sequence an LLM can process in a single forward pass; defines the maximum input + output capa

context lengthTotal number of tokens (input + output) an LLM supports; the hard ceiling on how much text can be loaded into

long term memoryInformation persisted across sessions, tasks, or deployments; requires external storage and retrieval mechanis

working memoryThe transient token budget allocated for active reasoning within a single model call; scratch space for chain_

attention mechanismThe core operation allowing each token to attend to all others; scaled dot-product attention is the standard.

vector databaseStorage system indexing embeddings for similarity search; backbone of RAG. Examples: Pinecone, Weaviate, Chrom

conversation historyThe chronological log of user_message and assistant_message turns in a session; the primary raw material for c

em eventA discrete unit of experience stored in episodic memory — a single interaction, task completion, or notable ou

effective contextPortion of context_window the model actually leverages for any given generation; often less than the full wind

retrieval augmented generationArchitecture pattern where an LLM queries external knowledge sources and injects results into context before g

sliding windowFixed-size contiguous subset of recent tokens kept in working_memory as new tokens arrive and old ones are evi

context compressionTransforming a large context into a shorter equivalent representation — via summarization, extraction, or loss

attention sinksSpecial tokens that accumulate disproportionate attention mass, enabling efficient long-context modeling witho

chain of thoughtProducing intermediate reasoning steps before final answer; when demonstrated in context, subsequent reasoning

kv cacheThe key-value tensors cached during transformer self-attention to avoid recomputing activations for previously

prompt engineeringCrafting input text to elicit desired outputs; includes instruction design, example selection, formatting, and

chunking strategyHow documents are split into retrieval units; choices affect recall/precision: fixed-size, semantic (by senten

episodic memoryMemory storing discrete experiences or events — specific interactions,完成任务, or notable outcomes — organized ch

in context learningModel adapts behavior based on examples provided in the context window; no weight updates. Related to memory s

memory retrievalThe act of fetching stored memory entries — by semantic similarity, keyword, metadata, or temporal proximity —