Key facts
- The glossary keeps node-graph vocabulary concrete so comparisons stay legible.
- Orchestration terms are separated from provider, retrieval, and observability terms.
- Definitions are grounded in primary docs where possible and kept practical for implementation decisions.
How to use this glossary
Minimal agent orchestration is easier to discuss when terms stay concrete. This glossary uses the node-graph vocabulary from Pocket Flow as the baseline, then adds terms that become relevant when comparing tiny runtimes with heavier systems.
The most important distinction is between the orchestration core and application utilities. A node graph decides order and state. A model wrapper calls a provider. A vector store retrieves context. Observability records what happened. Keeping those boundaries clear prevents a minimal framework from quietly becoming a large one.
"Shared Store"
Terms
- Action
- A string returned by a node that tells the flow which labeled edge to follow next.
- Agent
- A graph pattern where a decision node chooses among actions based on current state and loops until it can stop.
- Batch node
- A node variant that runs the same operation across many items, often as the map phase of map-reduce.
- Durable execution
- Runtime support for pausing, resuming, replaying, and recovering long-running workflows after failure.
- Flow
- The graph runner that starts at a node and follows action-labeled transitions until no next node is found.
- Graph
- The explicit structure of nodes and edges that defines an LLM application workflow.
- LLM wrapper
- Application code that calls a model provider. In the minimal philosophy, this usually lives outside the orchestration core.
- Map-reduce
- A pattern that maps work over independent inputs and reduces partial outputs into one result.
- Node
- A small unit of work with a lifecycle for preparing inputs, executing compute, and writing results.
- Params
- Local task identifiers passed to nodes or flows, especially in batch work.
- RAG
- Retrieval-augmented generation: retrieve external context and use it with a model to generate an answer.
- Structured output
- A pattern where generation, parsing, validation, and repair are explicit graph steps.
- Workflow
- A mostly deterministic chain of nodes that decomposes a task into known steps.
Related reading
For concepts that are specific to Pocket Flow, start with the node graph model. For tradeoffs against LangChain, LangGraph, and LlamaIndex, read minimal vs heavy frameworks. For implementation, read build your own minimal orchestrator.
Sources used on this page
- Pocket Flow documentation The Pocket. Accessed July 6, 2026. Primary source for the 100-line claim, Graph + Shared Store abstraction, core patterns, and utility philosophy.
- Pocket Flow Node documentation The Pocket. Accessed July 6, 2026. Primary source for prep, exec, post, retries, and fallback behavior.
- Pocket Flow Flow documentation The Pocket. Accessed July 6, 2026. Primary source for action-based transitions, branching, loops, and nested flows.
- Pocket Flow Communication documentation The Pocket. Accessed July 6, 2026. Primary source for Shared Store and Params guidance.
- LangGraph overview LangChain. Accessed July 6, 2026. Primary source for durable execution, persistence, human-in-the-loop, memory, and production orchestration claims.
- Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks arXiv / NeurIPS 2020. Submitted May 22, 2020; accessed July 6, 2026. Primary research source for RAG as parametric plus non-parametric memory for generation.
Cite this page
Glossary. PocketFlow AI Guide. Updated July 6, 2026. https://pocketflowai.com/glossary/
PocketFlow AI Guide. "Glossary." Accessed July 6, 2026. https://pocketflowai.com/glossary/