clearflow 1.0.1

Compose type-safe flows for emergent AI

ClearFlow

Compose type-safe flows for emergent AI. 100% test coverage, zero dependencies.

Why ClearFlow?

• 100% test coverage – Every path proven to work
• Type-safe transformations – Errors caught at development time, not runtime
• Immutable state – No hidden mutations
• Zero dependencies – No hidden failure modes
• Single exit enforcement – No ambiguous endings
• AI-Ready Documentation – llms.txt for optimal coding assistant integration

Quick Start

pip install clearflow

Upgrading from v0.x? See the Migration Guide for breaking changes.

AI Assistant Integration

ClearFlow provides comprehensive documentation in llms.txt format for optimal AI assistant support.

Claude Code Setup

Add ClearFlow documentation to Claude Code with one command:

claude mcp add-json clearflow-docs '{
    "type":"stdio",
    "command":"uvx",
    "args":["--from", "mcpdoc", "mcpdoc", "--urls", "ClearFlow:https://raw.githubusercontent.com/artificial-sapience/clearflow/main/llms.txt"]
}'

For IDEs (Cursor, Windsurf), see the mcpdoc documentation.

Direct URL Access

Use these URLs directly in any AI tool that supports llms.txt:

• Minimal index (~2KB): https://raw.githubusercontent.com/artificial-sapience/clearflow/main/llms.txt
• Full documentation (~63KB): https://raw.githubusercontent.com/artificial-sapience/clearflow/main/llms-full.txt

Examples

Name	Description
Chat	Simple conversational flow with OpenAI
Portfolio Analysis	Multi-specialist workflow for financial analysis
RAG	Full retrieval-augmented generation with vector search

Core Concepts

Node[TIn, TOut]

A unit that transforms state from TIn to TOut (or Node[T] when types are the same).

• prep(state: TIn) -> TIn – optional pre-work/validation
• exec(state: TIn) -> NodeResult[TOut] – required; return new state + outcome
• post(result: NodeResult[TOut]) -> NodeResult[TOut] – optional cleanup/logging

Nodes are frozen dataclasses that execute async transformations without mutating input state.

NodeResult[T]

Holds the new state and an outcome string used for routing.

flow()

A function that creates a flow with explicit routing:

flow("Name", start_node)
  .route(start_node, "outcome1", next_node)
  .route(next_node, "outcome2", final_node)
  .end(final_node, "done")  # exactly one termination

Routing: Routes are (node, outcome) pairs. Each outcome must have exactly one route.
Type inference: The flow infers types from start to end, supporting transformations.
Composability: A flow is itself a Node – compose flows within flows.

ClearFlow vs PocketFlow

Aspect	ClearFlow	PocketFlow
State	Immutable, passed via NodeResult	Mutable, passed via shared param
Routing	Outcome-based explicit routes	Action-based graph edges
Termination	Exactly one exit enforced	Multiple exits allowed
Type safety	Full Python 3.13+ generics	Dynamic (no annotations)

ClearFlow emphasizes robust, type-safe orchestration with validation and guardrails. PocketFlow emphasizes brevity and flexibility with minimal overhead.

Development

Install uv

• Please see official uv docs for other ways to install uv.

curl -LsSf https://astral.sh/uv/install.sh | sh

Clone and set up development environment

git clone https://github.com/artificial-sapience/clearflow.git
cd ClearFlow
uv sync --all-extras     # Creates venv and installs deps automatically
./quality-check.sh       # Run all checks

License

MIT

Acknowledgments

Inspired by PocketFlow's Node-Flow-State pattern.