MultiplAI

Autonomous development agents that turn GitHub issues into pull requests.

MultiplAI is your parallel coding pipeline. You create issues, it executes in batch, you review ready PRs.

MultiplAI is not a chatbot. It's your extra dev team working in parallel.

What it does

┌─────────────────────────────────────────────────────────────────┐
│                                                                 │
│    Issues           MultiplAI              Ready PRs            │
│    ┌───┐           ┌─────────┐             ┌───┐               │
│    │ 1 │──────────▶│         │────────────▶│PR1│               │
│    └───┘           │  ⚡⚡⚡⚡  │             └───┘               │
│    ┌───┐           │         │             ┌───┐               │
│    │ 2 │──────────▶│ Planner │────────────▶│PR2│               │
│    └───┘           │ Coder   │             └───┘               │
│    ┌───┐           │ Tester  │             ┌───┐               │
│    │ 3 │──────────▶│ Review  │────────────▶│PR3│               │
│    └───┘           └─────────┘             └───┘               │
│                                                                 │
│    You plan         Parallel              You review            │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

Create issues describing what you need
MultiplAI processes them with AI agents in parallel
Review ready PRs and merge

Features

Autonomous Pipeline: Issue → Plan → Code → Test → Review → PR
Multi-Agent Architecture: Specialized agents for planning, coding, fixing, and reviewing
Batch Processing: Process multiple issues in parallel with Jobs API
Self-Healing: Automatic retry with Fixer agent when tests fail (up to 3 attempts)
Code Review: LLM-based review before opening PR
Dashboard: Real-time monitoring, analytics, and job management
Linear Integration: Sync with Linear issues for project management

Quick Start

1. Install

git clone https://github.com/limaronaldo/MultiplAI.git
cd MultiplAI
bun install

2. Configure

cp .env.example .env
# Fill in: GITHUB_TOKEN, ANTHROPIC_API_KEY, DATABASE_URL
bun run db:migrate

3. Run

# Backend
bun run dev

# Dashboard (separate terminal)
cd autodev-dashboard
bun install
bun run dev

4. Use

Configure webhook in your GitHub repo → https://your-server/webhooks/github
Create an issue with label auto-dev
MultiplAI delivers a PR

Architecture

Pipeline Flow

Issue labeled ──▶ Planner ──▶ Coder ──▶ Tester ──▶ Reviewer ──▶ PR
     │              │           │          │           │        │
   auto-dev      Analyzes   Implements   Runs CI    Reviews   Ready
   trigger       + DoD      as diff      + Fix      code      for you

Agents

Agent	Model	Function
Planner	Claude Sonnet	Analyzes issue, creates plan and Definition of Done
Coder	Claude Opus	Writes code as unified diff
Fixer	Claude Opus	Fixes code when tests fail (up to 3x)
Reviewer	Claude Sonnet	Code review before opening PR

State Machine

NEW → PLANNING → PLANNING_DONE → CODING → CODING_DONE → TESTING
    → TESTS_PASSED → REVIEWING → REVIEW_APPROVED → PR_CREATED → WAITING_HUMAN
                                                              → COMPLETED

Fix Loop: TESTS_FAILED → FIXING → CODING_DONE (retry)
Review Loop: REVIEW_REJECTED → CODING (re-code with feedback)

Project Structure

├── src/
│   ├── index.ts              # Bun HTTP server entry
│   ├── router.ts             # HTTP routes
│   ├── core/
│   │   ├── types.ts          # Types and Zod schemas
│   │   ├── state-machine.ts  # State transitions
│   │   └── orchestrator.ts   # Main processing logic
│   ├── agents/
│   │   ├── base.ts           # Base agent class
│   │   ├── planner.ts        # Planning agent
│   │   ├── coder.ts          # Code generation
│   │   ├── fixer.ts          # Error fixing
│   │   └── reviewer.ts       # Code review
│   └── integrations/
│       ├── anthropic.ts      # Claude SDK
│       ├── github.ts         # Octokit wrapper
│       ├── linear.ts         # Linear SDK
│       └── db.ts             # PostgreSQL
├── autodev-dashboard/        # React dashboard
│   ├── src/
│   │   ├── components/       # UI components
│   │   ├── hooks/            # React hooks
│   │   ├── pages/            # Route pages
│   │   └── stores/           # Zustand stores
│   └── ...
└── prompts/                  # Agent prompt templates

Dashboard

Real-time monitoring dashboard built with React + TypeScript + Tailwind.

Features

Dashboard: KPI cards, status distribution, activity charts
Tasks: List, filter, and view task details with diff viewer
Jobs: Create and manage batch jobs, real-time progress
Logs: Live log streaming via SSE
Settings: Theme toggle, repository config

Running the Dashboard

cd autodev-dashboard
bun install
bun run dev
# Opens at http://localhost:5173

Configure the backend URL in .env.local:

VITE_API_BASE_URL=http://localhost:3000/api

API Reference

Tasks

Method	Endpoint	Description
GET	`/api/tasks`	List all tasks
GET	`/api/tasks/:id`	Get task details
POST	`/api/tasks/:id/process`	Manually trigger processing

Jobs (Batch Processing)

Method	Endpoint	Description
POST	`/api/jobs`	Create job with multiple issues
GET	`/api/jobs/:id`	Get job status with task summaries
POST	`/api/jobs/:id/run`	Start job processing
POST	`/api/jobs/:id/cancel`	Cancel running job

Create a Job

curl -X POST http://localhost:3000/api/jobs \
  -H "Content-Type: application/json" \
  -d '{"repo": "owner/repo", "issueNumbers": [21, 22, 23]}'

Start a Job

curl -X POST http://localhost:3000/api/jobs/{id}/run

Webhooks

Method	Endpoint	Description
POST	`/webhooks/github`	Receives GitHub events

Events: issues (labeled), check_run (completed), pull_request (closed)

Other

Method	Endpoint	Description
GET	`/api/health`	Health check
GET	`/api/review/pending`	Linear issues awaiting review

Configuration

Environment Variables

Required:

GITHUB_TOKEN - GitHub personal access token
ANTHROPIC_API_KEY - Claude API key
DATABASE_URL - Neon PostgreSQL connection string

Optional:

LINEAR_API_KEY - Linear API key for issue sync
GITHUB_WEBHOOK_SECRET - Webhook signature validation
MAX_ATTEMPTS - Max fix attempts (default: 3)
MAX_DIFF_LINES - Max lines in diff (default: 300)

Safety Limits

Config	Default	Description
`maxAttempts`	3	Retry attempts before failing
`maxDiffLines`	300	Maximum diff size
Complexity	XS/S	L/XL issues are rejected
Allowed paths	`src/`, `lib/`, `tests/`	Safe to modify
Blocked paths	`.env`, `secrets/`, `.github/workflows/`	Never touched

Writing Good Issues

Good Issue ✅

## Add email validation function

### Requirements
- Create `validateEmail(email: string): boolean` in `src/utils.ts`
- Use regex for validation
- Return true if valid, false if invalid

### Acceptance Criteria
- [ ] Function exists and is exported
- [ ] Validates correct format ([email protected])
- [ ] Rejects invalid formats
- [ ] Has unit tests

Bad Issue ❌

Improve the email system

Deployment

Fly.io

# First deploy
fly apps create multiplai --region gru
fly secrets set GITHUB_TOKEN=ghp_xxx ANTHROPIC_API_KEY=sk-ant-xxx DATABASE_URL=postgresql://...
fly deploy

# Future deploys
fly deploy

# View logs
fly logs

Dashboard Deployment

The dashboard can be deployed to any static hosting (Vercel, Netlify, Cloudflare Pages):

cd autodev-dashboard
bun run build
# Deploy dist/ folder

Docker with Visual Testing (CUA)

AutoDev includes Docker support for running with Computer Use Agent (CUA) visual testing capabilities.

Prerequisites

Docker and Docker Compose installed
Playwright browsers will be installed automatically in the container

Quick Start

# 1. Copy environment file
cp .env.example .env
# Fill in required variables: GITHUB_TOKEN, ANTHROPIC_API_KEY, DATABASE_URL

# 2. Start with Docker Compose
docker-compose -f docker-compose.cua.yml up -d

# 3. View logs
docker-compose -f docker-compose.cua.yml logs -f

# 4. Stop
docker-compose -f docker-compose.cua.yml down

Build Custom Image

# Build the CUA-enabled image
docker build -f Dockerfile.cua -t autodev-cua .

# Run manually
docker run -d \
  --name autodev-cua \
  --shm-size 2gb \
  -p 3000:3000 \
  -e DATABASE_URL="postgresql://..." \
  -e GITHUB_TOKEN="ghp_xxx" \
  -e ANTHROPIC_API_KEY="sk-ant-xxx" \
  -v $(pwd)/screenshots:/app/screenshots \
  autodev-cua

Visual Testing Configuration

Enable visual testing in your .env:

# Visual Testing (CUA)
ENABLE_VISUAL_TESTING=true
CUA_HEADLESS=true           # Run browsers in headless mode
CUA_TIMEOUT=60000           # Test timeout in ms (default: 60s)
CUA_MAX_ACTIONS=30          # Max actions per test (default: 30)

Docker Features

Playwright Support: Pre-installed Chromium, Firefox, and WebKit
Shared Memory: 2GB shm-size for stable browser execution
Volume Mounts:
- ./screenshots - Visual test screenshots
- ./logs - Application logs
Health Checks: Automatic health monitoring on /api/health
Security: Runs with seccomp:unconfined for Chrome compatibility

Troubleshooting

Browser crashes:

# Increase shared memory
docker-compose -f docker-compose.cua.yml down
# Edit docker-compose.cua.yml and increase shm_size to 4gb
docker-compose -f docker-compose.cua.yml up -d

Missing screenshots:

# Check volume mount permissions
ls -la screenshots/
chmod 755 screenshots/

Container health issues:

# Check health status
docker inspect autodev-cua | grep -A 10 Health

# Check logs
docker logs autodev-cua --tail 100

Tech Stack

Backend:

Bun runtime
TypeScript
Neon PostgreSQL
Anthropic Claude SDK

Dashboard:

React 19
TypeScript
Tailwind CSS
Zustand (state management)
React Router v7
react-diff-viewer-continued

Integrations:

GitHub API (Octokit)
Linear API
SSE for real-time logs

Roadmap

Documentation

CLAUDE.md - Complete codebase guide for AI agents
LEARNINGS.md - Model performance data and lessons learned

License

MIT

MultiplAI — Multiply your team's capacity, not your headcount.

Name		Name	Last commit message	Last commit date
Latest commit History 287 Commits
.github		.github
docs		docs
packages		packages
scripts		scripts
.dockerignore		.dockerignore
.gitignore		.gitignore
CLAUDE.md		CLAUDE.md
Dockerfile		Dockerfile
README.md		README.md
fly.toml		fly.toml
package.json		package.json
pnpm-lock.yaml		pnpm-lock.yaml
pnpm-workspace.yaml		pnpm-workspace.yaml
tsconfig.json		tsconfig.json
turbo.json		turbo.json

limaronaldo/MultiplAI

Folders and files

Latest commit

History

Repository files navigation

MultiplAI

What it does

Features

Quick Start

1. Install

2. Configure

3. Run

4. Use

Architecture

Pipeline Flow

Agents

State Machine

Project Structure

Dashboard

Features

Running the Dashboard

API Reference

Tasks

Jobs (Batch Processing)

Create a Job

Start a Job

Webhooks

Other

Configuration

Environment Variables

Safety Limits

Writing Good Issues

Good Issue ✅

Bad Issue ❌

Deployment

Fly.io

Dashboard Deployment

Docker with Visual Testing (CUA)

Prerequisites

Quick Start

Build Custom Image

Visual Testing Configuration

Docker Features

Troubleshooting

Tech Stack

Roadmap

Documentation

License

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases 108

Packages 0

Contributors 2

Languages

Packages