Unlike happy-cli which both runs and controls agents, happy-agent only controls them — listing machines, spawning sessions on a machine, creating sessions, sending messages, reading history, monitoring state, and stopping sessions.

Installation

From the monorepo:

yarn workspace happy-agent build

Or link globally:

cd packages/happy-agent && npm link

Authentication

Happy Agent uses account authentication via QR code, the same flow as linking a device in the Happy mobile app.

# Authenticate by scanning QR code with the Happy mobile app
happy-agent auth login

# Check authentication status
happy-agent auth status

# Clear stored credentials
happy-agent auth logout

Credentials are stored at ~/.happy/agent.key.

Commands

List sessions

# List all sessions
happy-agent list

# List only active sessions
happy-agent list --active

# Output as JSON
happy-agent list --json

List machines

# List all machines
happy-agent machines

# List only active machines
happy-agent machines --active

# Output as JSON
happy-agent machines --json

Spawn on a machine

# Spawn a session on a specific machine
happy-agent spawn --machine <machine-id> --path ~/project

# Let the daemon create the directory if needed
happy-agent spawn --machine <machine-id> --path ~/new-project --create-dir

# Choose a specific agent
happy-agent spawn --machine <machine-id> --path ~/project --agent codex

# Output as JSON
happy-agent spawn --machine <machine-id> --path ~/project --json

Session status

# Get live session state (supports ID prefix matching)
happy-agent status <session-id>

# Output as JSON
happy-agent status <session-id> --json

Create a session

# Create a new session with a tag
happy-agent create --tag my-project

# Specify a working directory
happy-agent create --tag my-project --path /home/user/project

# Output as JSON
happy-agent create --tag my-project --json

Send a message

# Send a message to a session
happy-agent send <session-id> "Fix the login bug"

# Send with yolo permissions
happy-agent send <session-id> "Ship it" --yolo

# Send and wait for the agent to finish
happy-agent send <session-id> "Run the tests" --wait

# Output as JSON
happy-agent send <session-id> "Hello" --json

Message history

# View message history
happy-agent history <session-id>

# Limit to last N messages
happy-agent history <session-id> --limit 10

# Output as JSON
happy-agent history <session-id> --json

Stop a session

happy-agent stop <session-id>

Wait for idle

# Wait for agent to become idle (default 300s timeout)
happy-agent wait <session-id>

# Custom timeout
happy-agent wait <session-id> --timeout 60

Exit code 0 when agent becomes idle, 1 on timeout.

Environment Variables

HAPPY_SERVER_URL - API server URL (default: https://api.cluster-fluster.com)
HAPPY_HOME_DIR - Home directory for credential storage (default: ~/.happy)

Session ID Matching

All commands that accept a <session-id> support prefix matching. You can provide the first few characters of a session ID and the CLI will resolve the full ID.

Machine-aware commands such as spawn --machine <machine-id> also support ID prefix matching.

Encryption

All machine and session data is end-to-end encrypted. New records use AES-256-GCM with per-record keys. Existing records created by other clients are decrypted using the appropriate key scheme (AES-256-GCM or legacy NaCl secretbox).

Requirements

Node.js >= 20.0.0
A Happy mobile app account for authentication

Publishing to npm

Maintainers can publish a new version:

yarn release               # From repo root: choose library to release
# or directly:
yarn workspace happy-agent release

This flow:

runs tests/build checks via prepublishOnly
creates a release commit and happy-agent-vX.Y.Z tag
creates a GitHub release with generated notes
publishes happy-agent to npm

License

MIT