An Open-Source
Autonomous AI
Research Framework.
From hypothesis to running experiments to a full reproducible trace. Watch Archimedes work.
Give Archimedes a hypothesis, paper, dataset, or benchmark.
It produces a complete, reproducible research trace — from the first literature search to a reproducible draft and the full record behind it.
Literature Synthesis
Archimedes uses the Google Agent SDK to orchestrate an autonomous ideation loop. It surveys ArXiv and Semantic Scholar to identify unexploited research frontiers and mathematically novel hypotheses.
Neural Architecture
Leveraging Claude Code for surgical refactoring, the agent implements complex neural networks from scratch. It utilises structural code intelligence to perform AST inspections, ensuring mathematical correctness.
Empirical Validation
The Stage Reflector acts as an autonomous Principal Investigator, adapting the research plan in real-time based on training logs and convergence metrics. Results are compiled into a reproducible LaTeX manuscript alongside a full session trace.