digital-adriana

Zero-touch deployer for self-teaching AI digital workers.

Sets up a full agent stack — NanoClaw (containerized runtime) + Letta (self-teaching memory) — with one command.

Install & Deploy

npx digital-adriana deploy

That's it. The wizard handles everything:

1. Scans your machine for installed agent runtimes (Claude Code, Codex, OpenCode, Pi)
2. Asks for essentials — worker name, API key, Telegram bot token
3. Auto-provisions Letta memory (embedded SQLite, zero config)
4. Generates all configs (docker-compose, soul.md, nanoclaw.yml)
5. Pulls & starts Docker containers

Prerequisites

• Docker Desktop
• Node.js 20+
• At least one agent runtime:

  npm install -g @anthropic-ai/claude-code   # Claude Code
  npm install -g @openai/codex               # Codex

Commands

npx digital-adriana deploy              # Full interactive setup
npx digital-adriana deploy --config c.json  # Zero-touch (pre-filled config)
npx digital-adriana deploy --no-start   # Generate configs only
npx digital-adriana start               # Start services
npx digital-adriana stop                # Stop services
npx digital-adriana status              # Health check
npx digital-adriana export -o ./bundle  # Export trained skills

Zero-Touch Mode

Pre-fill a config JSON and skip all prompts:

{
  "workerName": "aria",
  "workerDescription": "Finance and admin assistant",
  "agentRuntime": "claude",
  "agentRuntimeBin": "/usr/local/bin/claude",
  "agentRuntimeVersion": "1.0.0",
  "anthropicKey": "sk-ant-...",
  "model": "claude-sonnet-4-6",
  "enableTelegram": true,
  "telegramToken": "123456789:ABCdefGHI-jklMNO",
  "apiSpendLimitUsd": 50,
  "skillsDir": ""
}

npx digital-adriana deploy --config ./config.json

What Gets Created

~/.digital-worker/<name>/
├── docker-compose.yml    # NanoClaw + Letta containers
├── .env                  # API keys, tokens (gitignored)
├── soul.md               # Agent personality & training protocol
├── nanoclaw.yml          # Runtime config
├── worker.json           # Worker metadata
├── skills/               # Learned & pre-built skills (SKILL.md files)
├── memory/               # Persistent memory files
├── data/
│   ├── letta/            # Letta SQLite database
│   └── nanoclaw/         # Runtime data
└── logs/

Exporting Trained Workers

Once the agent has been trained, export its skills and memory as a portable bundle:

npx digital-adriana export -o ./my-worker-bundle

Import on another machine:

npx digital-adriana deploy --config ./my-worker-bundle/worker.json

Stack

• NanoClaw — Secure containerized agent runtime. Every agent runs in its own Docker container.
• Letta — Self-teaching memory layer. Agents learn skills from experience, not just RAG.

License

MIT