Propel Documentation

Not an autonomous agent. Propel automates the tedious half of research engineering and stops at every decision that is yours. You decide; the machines execute.

Getting Started

New to Propel? Start here to understand what it does and how to set it up.

Installation & Setup

Install Propel, initialize it in your project, and run your first workflow in under 5 minutes.

Core Principles

The five non-negotiable principles injected into every session — the mindset beneath all mechanisms.

How It Works

The two things that changed most recently: almost everything now dispatches itself, and every decision gets a second model before it reaches you.

Automatic Dispatch

Mode selection, skill routing, subagent and auditor dispatch — what fires without you asking, and the line it will not cross.

Codex — Two Models

A second model critiques every design, diff and diagnosis. Every claim is verified against your repo. No command to invoke.

The Pipeline

Understand the end-to-end workflow: gates, questioners, and phase transitions.

Pipeline Overview

The full seven-stage workflow from intake to retrospective — how each phase connects and why ordering matters.

Gates & Questioners

Deep dive into Gate 0–4 and Questioners Q0–Q1. What each checkpoint asks, why, and what format.

Modes

Not every session needs the full pipeline. Each mode filters which skills and gates are active.

Researcher Mode

Literature review, investigation, deep research. Understanding the problem space before building anything.

Engineer Mode

Full pipeline with all gates and auditors. The default mode for implementing paper-derived code.

Debugger Mode

Root-cause analysis with evidence-backed diagnosis. Classifies code bugs vs. design issues vs. config problems.

Trainer Mode

Launch training runs, monitor progress, fix CUDA/OOM/path errors. Code is ready — time to train.

Skills

20 specialized skills organized by workflow phase. They activate on intent, not on magic phrases — describe the problem and the right one runs.

Investigation

Scaffolds scratch/ directories with living READMEs that persist across sessions.

Research Design

Paper-to-code mapping, regression risk assessment, and explicit design decisions before implementation.

Systematic Debugging

Diagnosis-first methodology with 3-strike limit. Never apply a fix without presenting the root cause.

Codex Consult

The automatic dual-model layer: how the brief is written, how claims are verified, and what the attributed card looks like.

All Skills →

Complete reference of all 20 skills: triggers, active modes, and detailed workflow descriptions.

Agents

Thirteen subagents, dispatched for you. Eight read-only auditors that check the work, and five workers that do it — each in its own context window, so the session keeps the conclusions rather than the file dumps.

Silent Bug Detector

Scans for 11 categories of silent failures — broadcasting bugs, wrong reductions, detached gradients, and more.

Paper Alignment Auditor

Cross-references code against paper equations and algorithms to catch implementation mismatches.

JAX Logic Auditor

Traces tensor shapes through JAX transforms (vmap, jit, scan) to catch shape mismatches.

Codex Bridge

Runs the Codex consult in an isolated context and returns only what survived verification against your actual code.

All Agents →

Complete reference of all 13 agents: dispatch rules, auto-trigger conditions, and output formats.