tracevox-ai

TraceVox Research CLI and compute runner. TraceVox is a visual, reproducible research environment for studying intelligent agents — reinforcement learning, trustworthy AI, multimodal agents, adversarial robustness. Observable · Reproducible · Adversarial.

The browser is the research control and visualization layer; this package is the compute layer: it executes typed research jobs (train / evaluate / sweep / export) on your own hardware — laptop, workstation, lab GPU server, or cluster node — and streams evidence to a TraceVox control plane (hosted, institution-hosted, or fully local/offline).

pip install tracevox-ai            # CLI + runner agent
pip install "tracevox-ai[rl]"      # + PyTorch/Gymnasium/MiniGrid for RL work

tracevox doctor                    # readable environment health report
tracevox connect --code <code>     # pair with a TraceVox control plane
tracevox runner start              # accept typed research jobs

Headless/HPC use needs no browser:

tracevox experiment run experiment.json
tracevox experiment export --experiment-id exp_…
tracevox experiment reproduce --experiment-id exp_…   # new lineage, original untouched

Security model: the runner connects outbound-only (no inbound ports), pairing codes are one-time and short-lived, runner tokens are hashed at rest and revocable, and jobs are typed research requests validated locally against registered environments/algorithms — there is no arbitrary-command job, so a TraceVox server can never execute arbitrary code on your machine.

Explore published research without installing anything: https://tracevox.ai/research — machine-readable docs at https://tracevox.ai/llms.txt.