Pocketflow

Plan and implement an LLM system with PocketFlow

Move from a high-level user problem and flow design to initial nodes and flows.

Before you start

• Python project with PocketFlow listed as a dependency; the guide gives pip install pocketflow as an installation option.
• A project-specific user problem and example inputs to reason through.
• A docs/design.md file for the high-level design.

Steps

1. Clarify the user-facing requirements and assess whether the task fits an AI system; work through example inputs manually.

2. Describe the high-level flow, select applicable patterns such as Agent, Map Reduce, RAG, or Workflow, and draw the flow in a Mermaid diagram in docs/design.md.

3. Identify needed utility functions from the flow; document their inputs, outputs, and purpose, implement them, and write a simple test for each.

4. Design the shared store and specify for each node what it reads, writes, and which utilities it uses.

5. Implement the initial nodes and flows based on the design; add logging and use PocketFlow’s built-in retry and fallback mechanisms as appropriate.

Guide limitations

• The guide describes an iterative development process, not a complete runnable application or deployment procedure.
• External utilities and integrations depend on the project; the example LLM utility uses an API key placeholder and does not establish provider setup or availability.
• The guide recommends keeping the initial solution simple and evaluating and refining results; it does not specify a universal test suite or success criterion.