See what's heavy. A modern, fast disk usage viewer for the terminal.
1.8x faster than dust · ~900 KiB binary · TTY + JSON + Pipe + AI Agent
cargo install tonz# The old way
du -sh * | sort -hr | head -n 10
# The tonz way
tonzOne command. Sorted by size. Proportional bars. Semantic coloring. Done.
- Fast — Parallel scanning via Rayon, 1.8x faster than dust on typical workloads
- Readable — Clean columnar output with proportional bars and 3-level semantic coloring
- AI-friendly —
--llmmode produces token-efficient output for AI agents (~5 tokens/entry, 5x fewer than JSON) - Smart hidden files — Hidden items aggregated by default; large ones (>5%) auto-promoted with
(!)marker - Symlink-aware — Symlinks shown with
@prefix in magenta, clearly distinguished from regular files - Filterable —
--threshold-pctand--topto focus on what matters - Unix-friendly — TTY mode for humans, raw TSV for pipes,
--jsonfor scripts. RespectsNO_COLOR - Safe — Single-filesystem by default. Hardlink deduplication. Memory guard for huge directories
- Tiny — ~900 KiB stripped binary
| Feature | tonz | dust | dua | ncdu | gdu |
|---|---|---|---|---|---|
| Parallel scan | Yes | Yes | Yes | No | Yes |
| Proportional bars | Yes | Yes | No | Yes | Yes |
| Semantic coloring | 3-level | No | No | No | Basic |
| Hidden file aggregation | Auto | No | No | No | No |
| Symlink detection | @prefix | No | No | No | No |
AI agent mode (--llm) |
Yes | No | No | No | No |
| JSON output | Yes | No | No | No | Yes |
| Pipe-friendly (TSV) | Yes | No | No | No | No |
| Hardlink dedup | Yes | No | No | Yes | No |
| Single binary size | ~900 KiB | ~2 MiB | ~1 MiB | ~300 KiB | ~5 MiB |
cargo install tonzgit clone https://github.com/Axect/tonz.git
cd tonz
cargo build --release
# Binary at target/release/tonzDownload from GitHub Releases for Linux (x86_64, aarch64) and macOS (x86_64, Apple Silicon).
tonz [PATH] [OPTIONS]
| Flag | Description |
|---|---|
-H, --hidden |
Show all hidden files and directories |
--json |
Output as line-delimited JSON |
--llm |
Token-efficient output for AI agents |
--sparkline |
Use compact sparkline visualization |
--threshold-pct <N> |
Hide entries below N% of total size |
--top <N> |
Show only the top N entries by size |
-j, --jobs <N> |
Number of threads (default: auto) |
--across-mounts |
Cross filesystem boundaries |
--no-color |
Disable colors |
tonz # Current directory
tonz /var/log # Specific path
tonz -H ~ # Show hidden files
tonz ~ | sort -rn | head -5 # Pipe to other tools
tonz --json ~/projects # JSON for scripting
tonz --llm --threshold-pct 1 ~ # AI agent modeTTY (default when connected to a terminal):
Documents ████████████████████░░░░ 4.2 GiB 38%
.cache ██████████░░░░░░░░░░░░░░ 2.1 GiB 19% (!)
@my_symlink ░░░░░░░░░░░░░░░░░░░░░░░░ 43 B 0%
LLM (--llm — token-efficient for AI agents):
/home/user (466G, 64 items, 4 shown)
173G 37% Dropbox/
153G 33% Insync/
17G 4% Documents/
7.5G 2% tbin/
Pipe (when piped to another command):
4509715456 Documents
1932735283 Downloads
JSON (--json):
{"name":"Documents","size":4509715456,"is_dir":true,"is_hidden":false,"is_estimate":false,"is_symlink":false,"percentage":38.2}How it works
tonz uses a 4-phase scanner pipeline:
- Classify — Read immediate children, check filesystem boundaries, detect symlinks
- Parallel Scan — Rayon work-stealing traversal per subdirectory with hardlink dedup
- Merge — Aggregate sizes, errors, and estimate flags
- Render — TTY/Pipe/JSON/LLM output with proportional bar visualization
The depth-1-only design is both a UX and performance decision: no tree construction means minimal allocation and sub-second results even on large directories.
Tip
For AI agents: Use tonz --llm --threshold-pct 1 <path> for disk analysis. Start at depth 1, drill into directories >20% of total. The --llm mode uses ~5 tokens per entry — 5x fewer than JSON.
Tip
tonz stays on a single filesystem by default. Running tonz / won't accidentally traverse network mounts or slow external drives. Use --across-mounts to opt in.
MIT