Examples

HyperFlow comes with multiple real-world examples showing standard configurations, local improvement loops, or evolutionary setups.

You can run single evaluations to test if an agent behaves well out-of-the-box, or you can invoke evolutionary loops so the agent writes patches to its own logic!

Example Domains

• Bash: Generates terminal commands.
• Scoring: Grade student math answers (accept/reject).
• Calculator: Solve math problems using a tool.
• Fact-check: Classify statements as true/false.
• Paper Review: Predict accept/reject for research paper abstracts.
• Git Evolution: A comprehensive standard flow utilizing patches across an isolated git structure branch.

Running the Examples

Evaluate Single Run

Execute run.py to evaluate the current solver logic strictly once.

cd examples/bash && python run.py
cd examples/factcheck && python run.py
cd examples/paper_review && python run.py

Run Evolutionary Self-Improvement

Execute run.py evolve (or via specific script parameters depending on the example) to trigger the evolutionary loop.

cd examples/bash && python run.py evolve
cd examples/factcheck && python run.py evolve
cd examples/scoring && python run.py
cd examples/calculator && python run.py
cd examples/git_evolution && python run.py

For the git-based evolution:

cd examples/git_evolution && python run.py        # 2 generations
cd examples/git_evolution && python run.py 5      # 5 generations
cd examples/git_evolution && python run.py --reset # Start over