README

Harvy

Parallel Coding Agent Platform & Architecture Knowledge Graph Orchestrator

Harvy is an intelligent multi-agent coding platform built into an enhanced fork of T3 Code. Harvy orchestrates parallel coding agents across independent git worktrees, synchronizes their work through a unified AST knowledge graph, coordinates contract handoffs with the Jev decision engine, and automatically verifies integrated code against approved test suites.

Key Features

• Parallel Agent Orchestration: Divides features across specialized backend, frontend, and testing agents working concurrently in isolated worktrees without workspace pollution.
• AST Knowledge Graph & Explorer: Powered by Graphify, indexing codebases locally to visualize call chains, dependencies, and file relationships in an interactive graph and tree inspector.
• Jev AI Decision Engine: Intelligent task routing, API contract arbitration, architectural risk scoring, and automated failure triage with deterministic rule-based fallbacks.
• Project Decision Governance: Define global architecture and security constraints (e.g., JWT auth, database patterns) that are injected into every agent prompt and tracked across missions.
• Conflict-Free Integration & Verification: Merges parallel agent commits into a dedicated integration branch and runs test verification commands before touching your checkout.
• Integrated Mission Control: Access Harvy directly via the topbar header button, sidebar, or Ctrl + K command palette with a sleek, rounded dark-mode UI.

Repository Layout

Prerequisites

• Vite+ (vp): Unified toolchain providing Node 26.12.1 and pnpm 12.14.0.
• uv: Fast Python package manager for Graphify installation.
• Agent CLI: A supported agent runner (e.g., Claude Code or OpenCode).
• Git: Version 2.30+ with worktree support.

Setup Guide

1. Toolchain Installation

# Install Vite+
irm https://vite.plus/ps1 | iex

# Install Graphify locally via uv
uv tool install .\graphify-8\graphify-8

2. Harvy Dependencies

cd t3code-main\t3code-main

# Install dependencies via Vite+
vp i

# Configure environment variables (optional: set TYPESAFE_API_KEY for Jev AI decisions)
copy .env.example .env

cd ..\..

3. Initialize Demo Fixture

powershell -ExecutionPolicy Bypass -File scripts\setup-fixture.ps1

Running Harvy

Start the development server:

cd t3code-main\t3code-main
vp run dev --home-dir D:\GoAgent\.t3-dev --no-seed

Once running, navigate to the local URL (e.g., http://localhost:5734/):

1. Add Project: Click Add Project in the sidebar or press Ctrl + K and select the demo fixture path (D:\GoAgent-fixtures\profile-app).
2. Model Configuration: Navigate to Settings → Projects and configure the default model.
3. Launch Harvy: Click the Open Harvy button in the top navigation bar.
4. Index Repository: In the Knowledge Graph tab, select Index repository to generate the dependency graph.
5. Execute Mission:
   • Enter your feature requirement (e.g., "Add avatar image upload with thumbnail generation").
   • Click Plan Mission to let Harvy decompose the tasks.
   • Review and click Approve Plan, then Start Agents.
   • Monitor real-time agent threads, review contracts, and click Integrate & Verify.

Running Tests

Harvy includes automated unit and integration tests across both the server and web applications:

# Server test suite (GoAgent mission service, Jev decisions, planner, store)
cd t3code-main\t3code-main\apps\server
vp test run src/goagent

# Web UI test suite (Knowledge Explorer, layout, graph view)
cd ..\web
vp test run src/components/goagent

License

This project is licensed under the MIT License — see the LICENSE file for details.