An extensible harness for quantitative research.
QuantNavigator is a planned standalone application for running quantitative research with coding agents. The goal is to keep a research project's questions, experiments and evidence accessible across sessions, with a person deciding what to investigate and which conclusions the results support.
This repository contains the product vision and roadmap. There is no released application code, working runtime integration or installation command yet. All capabilities described below are planned.
Consider a project testing whether a new feature improves a forecasting model. The intended workflow starts with a question and a baseline, records the data and validation protocol, then runs a bounded comparison. You would inspect metrics and plots, check failures and limitations, and record a decision about the next step. The project would retain the experiment configuration and result artifacts needed to revisit that decision later.
QuantNavigator is intended to bring that work into one environment:
- Multiple personal workspaces, each containing separate research projects.
- Persistent project records, experiment history and results that can be inspected after an agent session ends.
- Runtime connections for delegating work to coding agents, with visible execution status and failures.
- Research tooling for reproducible comparisons, result inspection and reports, added through plugins and native components.
Planned external runtimes include Codex, Claude Code and Pi. Each integration will need its own implementation and verification; support and capabilities will be documented as they become available.
We are also considering a native QuantNavigator runtime built on Pi or Mastra. That choice remains open. Project records and result artifacts are intended to remain independent of the runtime. Transferring a runtime's complete internal session is outside the current plan.
Research capabilities will be introduced gradually through plugins and native components. The extension interfaces and the boundary between those components still need to be designed and tested.
See the roadmap for planned milestones and the evidence required to complete them. There are no announced release dates.
Use GitHub Discussions to describe a research workflow you would want to use, a result you need to inspect, or a runtime requirement.
Prefer small, focused pull requests. For runtime behavior changes or new plugins, open a discussion first in GitHub Discussions. See the roadmap for planned work and use Discussions for deeper topics.
See LICENSE.