Commands and configuration

Run archivetuna --help for the executable's command summary. Settings come from defaults, then archivetuna.config.json in the working directory, then CLI flags.

Commands

Command	Inputs	Writes
init [file]	Packaged starter profile	New profile; refuses overwrite
scan root	Directory metadata	manifest.json
dupes	Saved manifest	dupes.md, dupes.json
describe	Saved manifest and tier-appropriate evidence	Description cache/snapshot
rank --profile file	Saved descriptions and profile	Ranking cache/snapshot and reports
report	Matching saved descriptions/rankings	Reports only
run root --profile file	Archive and profile	Complete scan/analysis/reports
status	Output directory	Nothing
doctor	Local Ollama and renderer availability	Nothing
models [collage/meta] target	STL files and optional collages	Adjacent model outputs

Flags

--out selects core output/cache location. --tier selects T0–T3. --model selects the exact local Ollama tag. --limit caps selected folders or per-stage model batches; zero means unlimited. --max-depth changes folder selection depth. --no-vision is for models only. --dry-run previews the supported command without writes/inference/archive-content reads. --help/-h and --version/-v work without a config or model.

Run and rank require --profile. Report accepts it as an optional consistency check against the saved ranking profile. Unsupported flags and extra positional arguments are errors.

Configuration defaults

{
  "out": "./archivetuna-out",
  "tier": 1,
  "model": "gemma4:12b",
  "ollamaUrl": "http://localhost:11434",
  "limit": 0,
  "maxDepth": 3,
  "minFiles": 1,
  "sampleFilesPerDir": 3,
  "sampleBytesPerFile": 4096,
  "fullTextMaxFiles": 20,
  "fullTextMaxBytesPerFile": 65536,
  "fullTextMaxTotalBytes": 262144,
  "maxNamesInEvidence": 80,
  "dryRun": false
}

Counts and byte caps must be nonnegative integers. Tier must be 0, 1, 2, or 3. Unknown settings are rejected. The prune array replaces the built-in list: node_modules, .git, .svn, .hg, pycache, bower_components, vendor, pods, .gradle, .idea, .vs, .settings, .nuget, dist, .cache, $recycle.bin, and system volume information. Matching is case-insensitive on every platform.

The current output directory is excluded independently of the prune list. Core outputs cannot use the archive root itself as their directory.

Migration from Sifter

The product and executable are ArchiveTuna/archivetuna. The source folder may remain sifter. Rename sifter.config.json to archivetuna.config.json; legacy config is still read with a warning if the new name does not exist.

Use --out ./sifter-out to read an old manifest. Rerun describe and rank to create the new caches/snapshots. The old model cache format is intentionally ignored. Bare paths no longer run STL processing. Use models explicitly. Top-level collage/meta remain deprecated aliases.

Everything backends, find, content-hash verification, and additional file modes are not implemented in 0.2.0.