Green Aents

A VS Code extension for deploying, monitoring, and managing machine learning workloads — without leaving your editor.

Features

• Deploy to Docker, Fly.io, Railway, AWS (ECS/Lambda/EKS), and Kubernetes with pre-deploy validation, migration runner, deploy diff, and auto-rollback on health failure
• 10 cloud providers — AWS, GCP, Azure, DigitalOcean, Hetzner, Linode, Vultr, Fly.io, OCI, Cloudflare — with start/stop/scale/deploy/rollback from the tree view
• ML training — push code to GPU servers, launch jobs, tail logs, watchdog alerts (OOM, diverging loss, VRAM trajectory), OOM auto-retry with halved batch size, job queue, ETA status bar, auto-download checkpoints
• Experiment comparison — side-by-side loss curves, metrics tables, and hyperparameter diffs across training runs
• Security — env audit, vulnerability scanning (npm/pip/cargo/bundle), secret redaction in output, secret scanning on git stage, policy checks
• Dashboard — live container stats, health status, GPU stats, training metrics, deployment history
• Team sync — shared deployment history and credential distribution via a self-hosted backend

Prerequisites

Install the CLIs for the platforms you intend to use:

Run Green: Check Dependencies (Cmd/Ctrl+Shift+P) to see which tools are installed and get install links for any that are missing.

Installation

Install from the VS Code marketplace (search Green) or install the .vsix directly:

code --install-extension green-0.1.0.vsix

Quick Start

Deploy

1. Open a project folder in VS Code.
2. Connect a cloud provider: Green: Connect Cloud Provider and enter your credentials.
3. Run Green: Deploy — Green validates your project, shows a diff, runs migrations, deploys, and starts health monitoring.

Machine Learning

1. Green ML: Add GPU Server — enter SSH host, user, key path, remote project path, and results path.
2. Green ML: Push to GPU Server — rsyncs your project to the server.
3. Green ML: Launch Training Job — enter the training command; a watchdog starts automatically.
4. View live progress in the ETA status bar; get VS Code notifications on OOM, loss divergence, or stagnation.
5. Green ML: Compare Experiments — select runs in the left panel to compare loss curves, metrics, and hyperparameters side by side.

Configuration (.green.json)

Place a .green.json in your project root. schemaVersion must be 1.

{
  "schemaVersion": 1,
  "defaultTarget": "fly",
  "defaultRegion": "us-east-1",
  "deploymentName": "my-app",
  "namespace": "production",
  "autoRollback": {
    "enabled": true,
    "windowSecs": 300
  },
  "environments": {
    "staging": {
      "target": "railway",
      "region": "us-west-2"
    },
    "prod": {
      "target": "aws",
      "region": "eu-west-1",
      "service": "my-ecs-service",
      "cluster": "my-cluster"
    }
  }
}

Team Sync

Green can share deployment history and credentials across your team via a self-hosted backend.

Deploy the server

cd server/
cp .env.example .env          # fill in POSTGRES_URL and ENCRYPTION_KEY (64-char hex)
npm ci && npm run db:migrate
npm start

The server exposes:

• POST /teams — create a team, returns { teamId, token, syncUrl }
• GET/POST /teams/:id/deployments — shared deployment history
• GET/PUT /teams/:id/credentials — encrypted shared credentials

Connect team members

Each developer runs Green: Connect Team Sync and pastes the syncUrl and token. The token is stored in VS Code SecretStorage and never written to disk.

Add to .green.json:

{
  "schemaVersion": 1,
  "remote": {
    "syncUrl": "https://your-server.example.com/teams/YOUR_TEAM_ID",
    "credentialsSource": "https://your-server.example.com/teams/YOUR_TEAM_ID/credentials"
  }
}

Contributing

git clone <repo>
cd green
npm ci
npm run check-types   # type check
npm run lint          # eslint
npm run test:unit     # unit tests (95 tests, no VS Code required)
npm run compile       # build extension

Press F5 in VS Code to launch the Extension Development Host.

Security

• All secrets entered through provider dialogs are stored in VS Code SecretStorage (OS keychain), never in plaintext files.
• Output panels redact stored secret values automatically.
• Pre-deploy validation runs security scans, env audits, and vulnerability scanning before every deploy.
• Shell commands are always quoted with q() to prevent injection.