Skip to content

Give AxonX a Star on GitHub! ↗

AGENT-NATIVE QUANT HARNESS

Quantitative research.
Built with Agents.

An agent-native harness for quantitative code development, experiment execution, and result analysis. Use your prompt and AxonX Skill with an Agent to improve Qlib Alpha158-based plugins or develop new research plugins.

PLUGIN JOBTASKRESULT

Latest Updates

  • 0.3.0 released: supports inheritable configuration, unified Task execution, and scheduled Tasks that continue across service restarts.
  • Alpha158 experiments: risk factors and rank retention improved Top20 net annualized returns by +1.36 pp and +24.70 pp over the adapted baseline. See results and evidence limits.
  • Studio Playground: explore Tasks and charts with simulated data. Try it online.

01 / GET STARTED

Run your first task

Run the built-in demo, check its result and exit code, then inspect the record.

↗

02 / AGENT DEVELOPMENT

Develop with an Agent

Prepare source and tools; use the Skill and a research prompt to develop plugins.

↗

03 / RESEARCH

Build a research workflow

Run a baseline, develop plugins, and evaluate features, models, and portfolio policies.

↗

01 / AXONX STUDIO

See your research unfold.

Explore Studio ↗

From task execution to result interpretation, inspect your research in one workspace. Complete Studio 0.3.0 screenshots use Playground's fictional data and simulated execution.

04 / AGENT RESEARCH CASE

Improving Qlib Alpha158 with factors and portfolio policies

Experiment design ↗

Top20 net annualized returns are 7.67%, 9.03%, and 32.37% for Alpha158, two added risk factors, and 3-day rank retention, with buy/sell fees of 0.05%/0.15%. Base and factor models use different feature_fraction settings; market data is incomplete and there is no independent confirmation window.

Alpha158 and risk-factor IC / RankIC comparisonTop20 / Top30 net annualized returns for baseline, risk factors and 3-day strategy

05 / BUILD YOUR RESEARCH LOOP

Start with one Task.

The built-in demo needs no external data or model credentials.

Agent-native quant research.