AlgoTune Agent
Command-driven algorithm optimizer with evaluation tools, profiling, multi-file edits, and best-snapshot restore.
# Upstream AlgoTune Agent defaults: $1 spend, 9,999 message cap, five full messages per role.
# Spend/history are method invariants in the controller; max_iterations is the message safety cap.
seed: 0
general:
max_iterations: 9999
checkpoint_interval: 10
mutation_approach: diff_based_edit
inner_retry_times: 1
# A multi-file workspace is base64-packed into one Genome (roughly 4/3 overhead).
max_solution_length: 1000000
proposer:
temperature: 0.0
max_tokens: 30000
reasoning_effort: high