Daemon Management

The hali daemon (halid) is the background process that seeds models, announces them on your LAN, and communicates with the public registry. This page covers everything you need to know about managing it.

---

Daemon lifecycle

Auto-start behavior

The daemon starts automatically when:

• You run hali pull and the daemon isn't already running
• You run hali stats or hali stats --web and it isn't running

In user mode (no system service), the daemon stops when you close your terminal. In service mode (Windows SCM / Linux systemd), it runs continuously regardless of terminal state.

---

Basic controls

# Start the daemon manually
hali daemon start

# Check what it's doing
hali daemon status

# Stop it
hali daemon stop

If the daemon is already running, start is a no-op. If it's not running, stop is a no-op.

---

Understanding daemon status

hali daemon status

Daemon running  PID 12345  uptime 2h15m0s  port 51234

SEEDING                                     STATUS    PEERS
------------------------------------------  --------  -----
mistral:7b:instruct:q4_k_m                 seeding   3 peers
  magnet: magnet:?xt=urn:btih:a3f9c21b4d67...
llama:8b:instruct:q5_k_m                   hashing   —
llama:13b:instruct:q4_k_m                  seeding   1 peer
  magnet: magnet:?xt=urn:btih:f8c41dc0a13b...

LAN AVAILABLE                               PEERS     INFOHASH
------------------------------------------  -----     --------
mistral:7b:instruct:q4_k_m                 1         a3f9c21b4d67…

Fields explained

Field	Meaning
PID	Process ID of the daemon
uptime	How long the daemon has been running
port	BitTorrent port for peer connections
SEEDING	Models your daemon is actively sharing
STATUS — seeding	Torrent is ready, actively sharing pieces
STATUS — hashing	Torrent metadata still being computed (piece hashing in progress)
PEERS (SEEDING)	How many peers are actively downloading from you
magnet	The magnet URI for this model — copy-pasteable
LAN AVAILABLE	Models announced by other machines on your LAN
PEERS (LAN AVAILABLE)	How many machines on your LAN have this model
INFOHASH	The torrent infohash — links LAN announcements to the SEEDING list above

Status transitions

[NEW MODEL] → hashing → seeding → (active)
     ↑                          ↓
     └──── already cached ──────┘

• hashing — the daemon is computing SHA-1 piece hashes (16 MiB chunks). For a 4 GB model, ~256 chunks. Takes 1–2 minutes. If streaming_hash was enabled during download, this phase is skipped.
• seeding — ready to share. Peers can download from you.

---

The web dashboard

hali stats --web

Opens http://127.0.0.1:47433 in your browser.

The dashboard shows:

• Live transfer speeds — upload and download rates with an SVG sparkline
• Session totals — cumulative data transferred this session
• Active models table — each model with:
  • Name and quantization
  • Status (seeding / hashing)
  • Peer count
  • Clickable magnet link
• Auto-refresh — updates every few seconds

The dashboard is local-only by default (binds to 127.0.0.1). To expose it to your LAN, see Linux LAN mode.

---

Service mode vs. user mode

	User Mode	Service Mode
Start	hali daemon start	hali service install (once, then auto)
Stop	hali daemon stop	hali service stop
Survives reboot	No	Yes
Auto-restart on crash	No	Yes (5s / 30s / 60s)
Data directory	~/.hali/	/var/lib/hali/ (Linux) or %ProgramData%\Hali\ (Windows)
Requires sudo	No	Yes (install only)
Best for	Personal machines	Team servers, always-on machines

---

Configuration

View all settings

hali config show

Prints every configuration key with its current value and the config file path.

Common daemon tuning

# Hash faster seeding handoff
hali config set streaming_hash true

# Verbose logging (troubleshooting)
hali config set debug true

# Cap bandwidth
hali config set max_upload_mbps 50
hali config set max_download_mbps 200

# Custom model storage
hali config set models_dir /mnt/data/llama-models

# Expose dashboard to LAN (Linux only)
hali config set daemon_listen_addr 0.0.0.0

Changes take effect after restart:

hali service restart    # Service mode
# or
hali daemon stop && hali daemon start    # User mode

---

IPC and how the CLI talks to the daemon

The CLI communicates with the daemon via JSON-over-TCP on a local 127.0.0.1 ephemeral port. The daemon writes its address to:

• Linux/macOS: ~/.hali/daemon.addr
• Windows: %ProgramData%\Hali\daemon.addr

When you run hali daemon status, the CLI:

1. Reads daemon.addr to find the daemon's port
2. Connects to 127.0.0.1:<port>
3. Sends a JSON status request
4. Receives and displays the response

If daemon.addr is missing or the connection fails, the CLI reports "Daemon is not running."

---

Logs

Linux (service mode)

journalctl -u halid -f                  # Follow logs in real time
journalctl -u halid --since "1h ago"    # Last hour
journalctl -u halid -n 100 --no-pager   # Last 100 lines

Linux (user mode)

Logs go to ~/.hali/logs/.

Windows

%ProgramData%\Hali\logs\hali.log

Enable debug logging for more verbosity:

hali config set debug true
hali service restart

---

Health check workflow

To verify everything is healthy:

# 1. Is the daemon running?
hali daemon status

# 2. Are models seeding?
# Look for "seeding" in the SEEDING section

# 3. Are LAN peers visible?
# Look for entries in the LAN AVAILABLE section

# 4. Is the dashboard accessible?
hali stats --web

# 5. Are there any error logs?
journalctl -u halid -n 50 --no-pager    # Linux
# or check %ProgramData%\Hali\logs\hali.log    # Windows

---

IPC protocol (for developers)

The daemon exposes a simple JSON-over-TCP API on its loopback port. Requests and responses are newline-delimited JSON objects.

Example status request:

{ "method": "status" }

Example response:

{
  "pid": 12345,
  "uptime": "2h15m0s",
  "torrent_port": 51234,
  "seeding": [
    {
      "model_id": "mistral:7b:instruct:q4_k_m",
      "status": "seeding",
      "peers": 3,
      "magnet": "magnet:?xt=urn:btih:..."
    }
  ]
}

This is an internal interface — use the CLI for normal operations.