gensyn-train

TrainMesh trainer node CLI. Join the decentralized ML training network — your machine trains models, you earn x402 payments. Two commands. No code.

Quick Start

# Install globally (requires bun)
npm install -g gensyn-train

# One-time setup
gensyn-train setup --coordinator <coordinator-ip>

# Start earning
gensyn-train start

Or zero-install:

npx gensyn-train setup --coordinator <coordinator-ip>

Prerequisites

• bun (>= 1.0) — JavaScript runtime
• Python 3.9+ — training execution
• pip — Python package manager
• OpenSSL — key generation

Commands

gensyn-train setup

One-time configuration. Does everything:

1. Checks system prerequisites (bun, Python 3.9+, pip)
2. Downloads AXL P2P node binary for your OS/arch
3. Generates ed25519 keypair in ~/.gensyn-train/private.pem
4. Writes AXL node config with coordinator peer address
5. Installs Python dependencies (msgpack, rich, requests)
6. Registers your node with the platform
7. Saves config to ~/.gensyn-train/config.json

gensyn-train setup --coordinator 34.46.48.224

Options:

| Flag | Default | Purpose | |------|---------|---------| | --coordinator <host> | 127.0.0.1 | Coordinator AXL peer address | | --register-url <url> | http://127.0.0.1:3000/api/workers/register | Platform registration endpoint |

gensyn-train start

Launches the trainer node daemon with a live TUI dashboard:

• Starts the AXL P2P node (encrypted mesh networking)
• Starts the worker daemon (waits for jobs, trains, sends results)
• Shows real-time earnings, active job, loss/accuracy, log

gensyn-train start

Press Ctrl+C to stop gracefully.

What It Runs

┌──────────────────────────────────────────────────────────────┐
│  gensyn-train  ● TRAINING              node: 3f9a2c...1e    │
├─────────────────────┬────────────────────────────────────────┤
│  EARNINGS           │  ACTIVE JOB                            │
│  Today    $0.61     │  Job ID:   mnist-cnn-a3f9             │
│  All time $4.06     │  Round:    3 / 10                      │
│  Rounds   82        │  Epoch 2/3  Loss:0.312  Acc:88.4%    │
├─────────────────────┴────────────────────────────────────────┤
│  LOG                                                        │
│  [10:44:40] Round 3 started                                 │
│  [10:44:55] Epoch 1 done — loss: 0.401, acc: 84.2%          │
└──────────────────────────────────────────────────────────────┘

How It Works

Your machine becomes a trainer node in the TrainMesh network:

1. The Coordinator sends you a training job (script + data shard URL) over the AXL P2P mesh
2. Your machine downloads the data shard, runs the training script locally
3. Your machine sends updated model weights back to the Coordinator
4. You earn x402 micropayments per completed training round

Your data never leaves your machine. Training scripts run as subprocesses — any ML framework works.

Files

Everything lives under ~/.gensyn-train/:

~/.gensyn-train/
├── config.json          # worker ID, keys, wallet
├── private.pem          # ed25519 private key (600 perms)
├── node-config.json     # AXL P2P node config
├── axl-node             # AXL binary (OS-specific)
└── jobs/                # job artifacts per job_id

Supported Platforms

| OS | Arch | Status | |----|------|--------| | macOS | arm64 (Apple Silicon) | Supported | | macOS | amd64 (Intel) | Supported | | Linux | amd64 | Supported | | Linux | arm64 | Supported | | Windows | — | Not supported |

Security

• All network traffic encrypted via AXL (Yggdrasil + TLS)
• Private key stored with 0600 permissions
• Training scripts run without sandbox (trusted network)
• HTTP API locked to 127.0.0.1