Bump versions

More string extractors
Add nix files
2026-03-11 12:22:02 -07:00 · 2026-03-11 12:22:02 -07:00 · 2026-03-11 12:21:58 -07:00 · 2026-03-11 11:05:32 -07:00 · 2026-03-11 11:05:32 -07:00 · 2026-03-11 11:05:30 -07:00
56 changed files with 1104 additions and 706 deletions
--- a/.editorconfig
+++ b/.editorconfig
@@ -0,0 +1,13 @@
+root = true
+
+[*]
+indent_style = tab
+indent_size = 4
+end_of_line = lf
+charset = utf-8
+trim_trailing_whitespace = true
+insert_final_newline = true
+
+[*.yml]
+indent_style = space
+indent_size = 2
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -2481,7 +2481,7 @@ checksum = "9b4f627cb1b25917193a259e49bdad08f671f8d9708acfd5fe0a8c1455d87220"

 [[package]]
 name = "pile"
-version = "0.0.1"
+version = "0.0.2"
 dependencies = [
 "anstyle",
 "anyhow",
@@ -2491,6 +2491,7 @@ dependencies = [
 "pile-config",
 "pile-dataset",
 "pile-toolbox",
+ "pile-value",
 "serde",
 "serde_json",
 "tokio",
@@ -2503,7 +2504,7 @@ dependencies = [

 [[package]]
 name = "pile-config"
-version = "0.0.1"
+version = "0.0.2"
 dependencies = [
 "serde",
 "smartstring",
@@ -2513,28 +2514,15 @@ dependencies = [

 [[package]]
 name = "pile-dataset"
-version = "0.0.1"
+version = "0.0.2"
 dependencies = [
- "async-trait",
- "aws-sdk-s3",
 "axum",
- "blake3",
 "chrono",
- "epub",
- "id3",
- "image",
- "itertools 0.14.0",
- "kamadak-exif",
- "mime",
- "mime_guess",
- "pdf",
- "pdfium-render",
 "pile-config",
- "pile-flac",
 "pile-toolbox",
+ "pile-value",
 "serde",
 "serde_json",
- "smartstring",
 "tantivy",
 "thiserror",
 "tokio",
@@ -2543,12 +2531,11 @@ dependencies = [
 "tracing",
 "utoipa",
 "utoipa-swagger-ui",
- "walkdir",
 ]

 [[package]]
 name = "pile-flac"
-version = "0.0.1"
+version = "0.0.2"
 dependencies = [
 "base64",
 "itertools 0.14.0",
@@ -2562,12 +2549,39 @@ dependencies = [

 [[package]]
 name = "pile-toolbox"
-version = "0.0.1"
+version = "0.0.2"
 dependencies = [
 "thiserror",
 "tokio",
 ]

+[[package]]
+name = "pile-value"
+version = "0.0.2"
+dependencies = [
+ "async-trait",
+ "aws-sdk-s3",
+ "blake3",
+ "chrono",
+ "epub",
+ "id3",
+ "image",
+ "kamadak-exif",
+ "mime",
+ "mime_guess",
+ "pdf",
+ "pdfium-render",
+ "pile-config",
+ "pile-flac",
+ "serde_json",
+ "smartstring",
+ "tokio",
+ "tokio-stream",
+ "toml",
+ "tracing",
+ "walkdir",
+]
+
 [[package]]
 name = "pin-project-lite"
 version = "0.2.16"
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -5,7 +5,7 @@ resolver = "2"
 [workspace.package]
 rust-version = "1.94.0"
 edition = "2024"
-version = "0.0.1"
+version = "0.0.2"

 [workspace.lints.rust]
 unused_import_braces = "deny"
@@ -67,20 +67,21 @@ pile-toolbox = { path = "crates/pile-toolbox" }
 pile-config = { path = "crates/pile-config" }
 pile-flac = { path = "crates/pile-flac" }
 pile-dataset = { path = "crates/pile-dataset" }
+pile-value = { path = "crates/pile-value" }

 # Clients & servers
 tantivy = "0.25.0"
 axum = { version = "0.8.8", features = ["macros", "multipart"] }
 utoipa = { version = "5.4.0", features = [
-    "axum_extras",
-    "chrono",
-    "url",
-    "uuid",
+	"axum_extras",
+	"chrono",
+	"url",
+	"uuid",
 ] }
 utoipa-swagger-ui = { version = "9.0.2", features = [
-    "axum",
-    "debug-embed",
-    "vendored",
+	"axum",
+	"debug-embed",
+	"vendored",
 ] }

 # Async & Parallelism
--- a/bump-version.sh
+++ b/bump-version.sh
@@ -0,0 +1,26 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+CARGO_TOML="$(dirname "$0")/Cargo.toml"
+DEFAULT_NIX="$(dirname "$0")/default.nix"
+
+# Read current version from workspace Cargo.toml
+current=$(grep '^version = ' "$CARGO_TOML" | head -1 | sed 's/version = "\(.*\)"/\1/')
+
+echo "Current version: $current"
+read -rp "New version: " new
+
+if [[ -z "$new" ]]; then
+	echo "No version entered. Aborting." >&2
+	exit 1
+fi
+
+# Update Cargo.toml workspace version
+sed -i "s/^version = \"$current\"/version = \"$new\"/" "$CARGO_TOML"
+
+# Update default.nix version field
+sed -i "s/version = \"$current\";/version = \"$new\";/" "$DEFAULT_NIX"
+
+echo "Bumped $current -> $new in:"
+echo "  $CARGO_TOML"
+echo "  $DEFAULT_NIX"
--- a/crates/pile-config/src/lib.rs
+++ b/crates/pile-config/src/lib.rs
@@ -1,9 +1,6 @@
 use serde::Deserialize;
 use std::{collections::HashMap, fmt::Debug, path::PathBuf};

-mod post;
-pub use post::*;
-
 mod misc;
 pub use misc::*;

@@ -40,10 +37,6 @@ pub struct DatasetConfig {

 	/// Where to find this field
 	pub source: HashMap<Label, Source>,
-
-	/// How to post-process this field
-	#[serde(default)]
-	pub post: Vec<FieldSpecPost>,
 }

 #[derive(Debug, Clone, Deserialize)]
@@ -100,10 +93,6 @@ pub struct FieldSpec {

 	/// How to find this field in a data entry
 	pub path: Vec<ObjectPath>,
-
-	/// How to post-process this field
-	#[serde(default)]
-	pub post: Vec<FieldSpecPost>,
 }

 #[derive(Debug, Clone, Copy, Deserialize, PartialEq, Eq)]
--- a/crates/pile-config/src/post.rs
+++ b/crates/pile-config/src/post.rs
@@ -1,18 +0,0 @@
-use serde::Deserialize;
-
-#[derive(Debug, Clone, Deserialize, PartialEq, Eq)]
-#[serde(untagged)]
-pub enum FieldSpecPost {
-	TrimSuffix { trim_suffix: String },
-	TrimPrefix { trim_prefix: String },
-	SetCase { case: Case },
-	Join { join: String },
-	NotEmpty { notempty: bool },
-}
-
-#[derive(Debug, Clone, Copy, Deserialize, PartialEq, Eq)]
-#[serde(rename_all = "lowercase")]
-pub enum Case {
-	Lower,
-	Upper,
-}
--- a/crates/pile-dataset/Cargo.toml
+++ b/crates/pile-dataset/Cargo.toml
@@ -10,37 +10,23 @@ workspace = true
 [dependencies]
 pile-config = { workspace = true }
 pile-toolbox = { workspace = true }
-pile-flac = { workspace = true }
+pile-value = { workspace = true }

 serde_json = { workspace = true }
-itertools = { workspace = true }
-walkdir = { workspace = true }
 tantivy = { workspace = true }
 tracing = { workspace = true }
 chrono = { workspace = true }
 toml = { workspace = true }
 thiserror = { workspace = true }
-smartstring = { workspace = true }
-blake3 = { workspace = true }
-epub = { workspace = true }
-kamadak-exif = { workspace = true }
-pdf = { workspace = true }
-pdfium-render = { workspace = true, optional = true }
-image = { workspace = true, optional = true }
-id3 = { workspace = true }
 tokio = { workspace = true }
 tokio-stream = { workspace = true }
-async-trait = { workspace = true }
-aws-sdk-s3 = { workspace = true }
-mime = { workspace = true }
-mime_guess = { workspace = true }
-serde = { workspace = true }

+serde = { workspace = true, optional = true }
 axum = { workspace = true, optional = true }
 utoipa = { workspace = true, optional = true }
 utoipa-swagger-ui = { workspace = true, optional = true }

 [features]
 default = []
-pdfium = ["dep:pdfium-render", "dep:image"]
-axum = ["dep:axum", "dep:utoipa", "dep:utoipa-swagger-ui"]
+pdfium = ["pile-value/pdfium"]
+axum = ["dep:axum", "dep:utoipa", "dep:utoipa-swagger-ui", "dep:serde"]
--- a/crates/pile-dataset/src/dataset.rs
+++ b/crates/pile-dataset/src/dataset.rs
@@ -1,6 +1,10 @@
 use chrono::{DateTime, Utc};
 use pile_config::{ConfigToml, Label, Source, objectpath::ObjectPath};
 use pile_toolbox::cancelabletask::{CancelFlag, CancelableTaskError};
+use pile_value::{
+	source::{DataSource, DirDataSource, S3DataSource, misc::path_ts_earliest},
+	value::{Item, PileValue},
+};
 use serde_json::Value;
 use std::{collections::HashMap, io::ErrorKind, path::PathBuf, sync::Arc, time::Instant};
 use tantivy::{Executor, Index, IndexWriter, TantivyError, collector::TopDocs};
@@ -9,13 +13,7 @@ use tokio::task::JoinSet;
 use tokio_stream::{StreamExt, wrappers::ReceiverStream};
 use tracing::{debug, info, trace, warn};

-use crate::{
-	DataSource, Item, PileValue,
-	extract::MetaExtractor,
-	index::{DbFtsIndex, FtsLookupResult},
-	path_ts_earliest,
-	source::{DirDataSource, S3DataSource},
-};
+use crate::index::{DbFtsIndex, FtsLookupResult};

 #[derive(Debug, Error)]
 pub enum DatasetError {
@@ -183,11 +181,12 @@ impl Datasets {
 		let Some(item) = self.get(source, key).await else {
 			return Ok(None);
 		};
-		let extractor = MetaExtractor::new(&item);
-		let root = PileValue::ObjectExtractor(Arc::new(extractor));
-		let Some(value) = root.query(path).await? else {
+
+		let item = PileValue::Item(item);
+		let Some(value) = item.query(path).await? else {
 			return Ok(None);
 		};
+
 		Ok(Some(value.to_json().await?))
 	}

--- a/crates/pile-dataset/src/extract/mod.rs
+++ b/crates/pile-dataset/src/extract/mod.rs
@@ -1,165 +0,0 @@
-use pile_config::Label;
-use std::{collections::HashMap, sync::Arc};
-
-mod flac;
-pub use flac::*;
-
-mod id3;
-pub use id3::*;
-
-mod fs;
-pub use fs::*;
-
-mod epub;
-pub use epub::*;
-
-mod exif;
-pub use exif::*;
-
-mod pdf;
-pub use pdf::*;
-
-mod toml;
-pub use toml::*;
-
-mod map;
-pub use map::*;
-
-mod sidecar;
-pub use sidecar::*;
-
-use crate::{Item, PileValue};
-
-/// An attachment that extracts metadata from an [Item].
-///
-/// Metadata is exposed as an immutable map of {label: value},
-/// much like a json object.
-#[async_trait::async_trait]
-pub trait ObjectExtractor: Send + Sync {
-	/// Get the field at `name` from `item`.
-	/// - returns `None` if `name` is not a valid field
-	/// - returns `Some(Null)` if `name` is not available
-	async fn field(&self, name: &pile_config::Label) -> Result<Option<PileValue>, std::io::Error>;
-
-	/// Return all fields in this extractor.
-	/// `Self::field` must return [Some] for all these keys
-	/// and [None] for all others.
-	async fn fields(&self) -> Result<Vec<Label>, std::io::Error>;
-
-	/// Convert this to a JSON value.
-	async fn to_json(&self) -> Result<serde_json::Value, std::io::Error> {
-		let keys = self.fields().await?;
-		let mut map = serde_json::Map::new();
-		for k in &keys {
-			let v = match self.field(k).await? {
-				Some(x) => x,
-				None => continue,
-			};
-			map.insert(k.to_string(), Box::pin(v.to_json()).await?);
-		}
-
-		Ok(serde_json::Value::Object(map))
-	}
-}
-
-/// An attachment that extracts metadata from an [Item].
-///
-/// Metadata is exposed as an immutable list of values.
-#[async_trait::async_trait]
-pub trait ListExtractor: Send + Sync {
-	/// Get the item at index `idx`.
-	/// Indices start at zero, and must be consecutive.
-	/// - returns `None` if `idx` is out of range
-	/// - returns `Some(Null)` if `None` is at `idx`
-	async fn get(&self, idx: usize) -> Result<Option<PileValue>, std::io::Error>;
-
-	async fn len(&self) -> Result<usize, std::io::Error>;
-
-	async fn is_empty(&self) -> Result<bool, std::io::Error> {
-		Ok(self.len().await? == 0)
-	}
-
-	/// Convert this list to a JSON value.
-	async fn to_json(&self) -> Result<serde_json::Value, std::io::Error> {
-		let len = self.len().await?;
-		let mut list = Vec::with_capacity(len);
-		for i in 0..len {
-			#[expect(clippy::expect_used)]
-			let v = self
-				.get(i)
-				.await?
-				.expect("value must be present according to length");
-			list.push(Box::pin(v.to_json()).await?);
-		}
-
-		Ok(serde_json::Value::Array(list))
-	}
-}
-
-pub struct MetaExtractor {
-	inner: MapExtractor,
-}
-
-impl MetaExtractor {
-	#[expect(clippy::unwrap_used)]
-	pub fn new(item: &Item) -> Self {
-		let inner = MapExtractor {
-			inner: HashMap::from([
-				(
-					Label::new("flac").unwrap(),
-					crate::PileValue::ObjectExtractor(Arc::new(FlacExtractor::new(item))),
-				),
-				(
-					Label::new("id3").unwrap(),
-					crate::PileValue::ObjectExtractor(Arc::new(Id3Extractor::new(item))),
-				),
-				(
-					Label::new("fs").unwrap(),
-					crate::PileValue::ObjectExtractor(Arc::new(FsExtractor::new(item))),
-				),
-				(
-					Label::new("epub").unwrap(),
-					crate::PileValue::ObjectExtractor(Arc::new(EpubExtractor::new(item))),
-				),
-				(
-					Label::new("exif").unwrap(),
-					crate::PileValue::ObjectExtractor(Arc::new(ExifExtractor::new(item))),
-				),
-				(
-					Label::new("pdf").unwrap(),
-					crate::PileValue::ObjectExtractor(Arc::new(PdfExtractor::new(item))),
-				),
-				(
-					Label::new("toml").unwrap(),
-					crate::PileValue::ObjectExtractor(Arc::new(TomlExtractor::new(item))),
-				),
-				(
-					Label::new("sidecar").unwrap(),
-					crate::PileValue::ObjectExtractor(Arc::new(SidecarExtractor::new(item))),
-				),
-			]),
-		};
-
-		Self { inner }
-	}
-}
-
-#[async_trait::async_trait]
-impl ObjectExtractor for MetaExtractor {
-	async fn field(&self, name: &pile_config::Label) -> Result<Option<PileValue>, std::io::Error> {
-		self.inner.field(name).await
-	}
-
-	#[expect(clippy::unwrap_used)]
-	async fn fields(&self) -> Result<Vec<Label>, std::io::Error> {
-		return Ok(vec![
-			Label::new("flac").unwrap(),
-			Label::new("id3").unwrap(),
-			Label::new("fs").unwrap(),
-			Label::new("epub").unwrap(),
-			Label::new("exif").unwrap(),
-			Label::new("pdf").unwrap(),
-			Label::new("sidecar").unwrap(),
-		]);
-	}
-}
--- a/crates/pile-dataset/src/extract/pdf/pdf_cover.rs
+++ b/crates/pile-dataset/src/extract/pdf/pdf_cover.rs
@@ -1,95 +0,0 @@
-use image::ImageFormat;
-use pdfium_render::prelude::*;
-use pile_config::Label;
-use std::{
-	collections::HashMap,
-	io::{BufReader, Cursor},
-	sync::{Arc, OnceLock},
-};
-use tracing::trace;
-
-use crate::{Item, PileValue, SyncReadBridge, extract::ObjectExtractor};
-
-pub struct PdfCoverExtractor {
-	item: Item,
-	output: OnceLock<HashMap<Label, PileValue>>,
-}
-
-impl PdfCoverExtractor {
-	pub fn new(item: &Item) -> Self {
-		Self {
-			item: item.clone(),
-			output: OnceLock::new(),
-		}
-	}
-
-	async fn get_inner(&self) -> Result<&HashMap<Label, PileValue>, std::io::Error> {
-		if let Some(x) = self.output.get() {
-			return Ok(x);
-		}
-
-		let reader = SyncReadBridge::new_current(self.item.read().await?);
-		let cover = tokio::task::spawn_blocking(move || {
-			let mut bytes = Vec::new();
-			std::io::Read::read_to_end(&mut BufReader::new(reader), &mut bytes)?;
-
-			let pdfium = Pdfium::default();
-
-			let document = pdfium
-				.load_pdf_from_byte_slice(&bytes, None)
-				.map_err(|e| std::io::Error::new(std::io::ErrorKind::InvalidData, e.to_string()))?;
-
-			let render_config = PdfRenderConfig::new().set_target_width(1024);
-
-			let page = document
-				.pages()
-				.get(0)
-				.map_err(|e| std::io::Error::new(std::io::ErrorKind::InvalidData, e.to_string()))?;
-
-			let image = page
-				.render_with_config(&render_config)
-				.map_err(|e| std::io::Error::new(std::io::ErrorKind::InvalidData, e.to_string()))?
-				.as_image();
-
-			let mut png_bytes = Vec::new();
-			image
-				.write_to(&mut Cursor::new(&mut png_bytes), ImageFormat::Png)
-				.map_err(|e| std::io::Error::other(e.to_string()))?;
-
-			Ok::<_, std::io::Error>(png_bytes)
-		})
-		.await
-		.map_err(std::io::Error::other)?;
-
-		let output = match cover {
-			Ok(data) => {
-				#[expect(clippy::unwrap_used)]
-				let label = Label::new("cover").unwrap();
-				HashMap::from([(
-					label,
-					PileValue::Blob {
-						mime: mime::IMAGE_PNG,
-						bytes: Arc::new(data),
-					},
-				)])
-			}
-			Err(error) => {
-				trace!(message = "Could not render pdf cover", ?error, key = ?self.item.key());
-				HashMap::new()
-			}
-		};
-
-		return Ok(self.output.get_or_init(|| output));
-	}
-}
-
-#[async_trait::async_trait]
-impl ObjectExtractor for PdfCoverExtractor {
-	async fn field(&self, name: &Label) -> Result<Option<PileValue>, std::io::Error> {
-		Ok(self.get_inner().await?.get(name).cloned())
-	}
-
-	async fn fields(&self) -> Result<Vec<Label>, std::io::Error> {
-		Ok(self.get_inner().await?.keys().cloned().collect())
-	}
-}
--- a/crates/pile-dataset/src/index/index_fts.rs
+++ b/crates/pile-dataset/src/index/index_fts.rs
@@ -1,9 +1,6 @@
-use itertools::Itertools;
-use pile_config::{Case, ConfigToml, DatasetFts, FieldSpecPost, Label};
-use std::{
-	path::PathBuf,
-	sync::{Arc, LazyLock},
-};
+use pile_config::{ConfigToml, DatasetFts, Label};
+use pile_value::value::{Item, PileValue};
+use std::{path::PathBuf, sync::LazyLock};
 use tantivy::{
 	DocAddress, Index, ReloadPolicy, TantivyDocument, TantivyError,
 	collector::Collector,
@@ -12,8 +9,6 @@ use tantivy::{
 };
 use tracing::{debug, trace, warn};

-use crate::{Item, PileValue, extract::MetaExtractor};
-
 #[derive(Debug, Clone)]
 pub struct FtsLookupResult {
 	pub score: f32,
@@ -76,11 +71,11 @@ impl DbFtsIndex {
 		doc.add_text(self.schema.get_field("_meta_source")?, item.source_name());
 		doc.add_text(self.schema.get_field("_meta_key")?, key);

-		let extractor = PileValue::ObjectExtractor(Arc::new(MetaExtractor::new(item)));
+		let item = PileValue::Item(item.clone());

 		let mut empty = true;
 		for name in self.fts_cfg().fields.keys() {
-			let x = self.get_field(&extractor, name).await?;
+			let x = self.get_field(&item, name).await?;

 			let val = match x {
 				Some(x) => x,
@@ -135,13 +130,6 @@ impl DbFtsIndex {
 				x => x.clone(),
 			};

-			for post in &field.post {
-				val = match apply(post, &val) {
-					Some(x) => x,
-					None => return Ok(None),
-				};
-			}
-
 			loop {
 				val = match val {
 					PileValue::String(x) => return Ok(Some(x.to_string())),
@@ -186,6 +174,15 @@ impl DbFtsIndex {
 						continue 'outer;
 					}

+					PileValue::Item(_) => {
+						trace!(
+							message = "Skipping field, is item",
+							field = field_name.to_string(),
+							?path,
+						);
+						continue 'outer;
+					}
+
 					PileValue::ListExtractor(_) => {
 						trace!(
 							message = "Skipping field, is ListExtractor",
@@ -296,104 +293,3 @@ impl DbFtsIndex {
 		return Ok(out);
 	}
 }
-
-pub fn apply(post: &FieldSpecPost, val: &PileValue) -> Option<PileValue> {
-	Some(match post {
-		FieldSpecPost::NotEmpty { notempty: false } => val.clone(),
-		FieldSpecPost::NotEmpty { notempty: true } => match val {
-			PileValue::Null => return None,
-			PileValue::String(x) if x.is_empty() => return None,
-			PileValue::Array(x) if x.is_empty() => return None,
-			x => x.clone(),
-		},
-
-		FieldSpecPost::SetCase { case: Case::Lower } => match val {
-			PileValue::Null => return None,
-			PileValue::U64(_) => return None,
-			PileValue::I64(_) => return None,
-			PileValue::Blob { .. } => return None,
-			PileValue::ObjectExtractor(_) => return None,
-			PileValue::ListExtractor(_) => return None,
-			PileValue::String(x) => PileValue::String(Arc::new(x.as_str().to_lowercase().into())),
-
-			PileValue::Array(x) => PileValue::Array(Arc::new(
-				x.iter().map(|x| apply(post, x)).collect::<Option<_>>()?,
-			)),
-		},
-
-		FieldSpecPost::SetCase { case: Case::Upper } => match val {
-			PileValue::Null => return None,
-			PileValue::U64(_) => return None,
-			PileValue::I64(_) => return None,
-			PileValue::Blob { .. } => return None,
-			PileValue::ObjectExtractor(_) => return None,
-			PileValue::ListExtractor(_) => return None,
-			PileValue::String(x) => PileValue::String(Arc::new(x.as_str().to_uppercase().into())),
-
-			PileValue::Array(x) => PileValue::Array(Arc::new(
-				x.iter()
-					.map(|x| apply(post, x))
-					.collect::<Option<Vec<_>>>()?,
-			)),
-		},
-
-		FieldSpecPost::TrimSuffix { trim_suffix } => match val {
-			PileValue::Null => return None,
-			PileValue::U64(_) => return None,
-			PileValue::I64(_) => return None,
-			PileValue::Blob { .. } => return None,
-			PileValue::ObjectExtractor(_) => return None,
-			PileValue::ListExtractor(_) => return None,
-
-			PileValue::String(x) => PileValue::String(Arc::new(
-				x.strip_suffix(trim_suffix).unwrap_or(x.as_str()).into(),
-			)),
-
-			PileValue::Array(x) => PileValue::Array(Arc::new(
-				x.iter()
-					.map(|x| apply(post, x))
-					.collect::<Option<Vec<_>>>()?,
-			)),
-		},
-
-		FieldSpecPost::TrimPrefix { trim_prefix } => match val {
-			PileValue::Null => return None,
-			PileValue::U64(_) => return None,
-			PileValue::I64(_) => return None,
-			PileValue::Blob { .. } => return None,
-			PileValue::ObjectExtractor(_) => return None,
-			PileValue::ListExtractor(_) => return None,
-
-			PileValue::String(x) => PileValue::String(Arc::new(
-				x.strip_prefix(trim_prefix).unwrap_or(x.as_str()).into(),
-			)),
-
-			PileValue::Array(x) => PileValue::Array(Arc::new(
-				x.iter()
-					.map(|x| apply(post, x))
-					.collect::<Option<Vec<_>>>()?,
-			)),
-		},
-
-		FieldSpecPost::Join { join } => match val {
-			PileValue::Null => return None,
-			PileValue::U64(_) => return None,
-			PileValue::I64(_) => return None,
-			PileValue::Blob { .. } => return None,
-			PileValue::ObjectExtractor(_) => return None,
-			PileValue::ListExtractor(_) => return None,
-
-			PileValue::String(x) => PileValue::String(x.clone()),
-
-			PileValue::Array(x) => PileValue::String(Arc::new(
-				x.iter()
-					.map(|x| apply(post, x))
-					.map(|x| x.and_then(|x| x.as_str().map(|x| x.to_owned())))
-					.collect::<Option<Vec<_>>>()?
-					.into_iter()
-					.join(join)
-					.into(),
-			)),
-		},
-	})
-}
--- a/crates/pile-dataset/src/lib.rs
+++ b/crates/pile-dataset/src/lib.rs
@@ -1,21 +1,7 @@
-mod traits;
-pub use traits::*;
-
-mod misc;
-pub use misc::*;
-
 mod dataset;
 pub use dataset::{Dataset, DatasetError, Datasets};

-mod item;
-pub use item::*;
-
-mod value;
-pub use value::*;
-
-pub mod extract;
 pub mod index;
-pub mod source;

 #[cfg(feature = "axum")]
 pub mod serve;
--- a/crates/pile-dataset/src/serve/field.rs
+++ b/crates/pile-dataset/src/serve/field.rs
@@ -5,12 +5,13 @@ use axum::{
 	response::{IntoResponse, Response},
 };
 use pile_config::{Label, objectpath::ObjectPath};
+use pile_value::value::PileValue;
 use serde::Deserialize;
 use std::{sync::Arc, time::Instant};
 use tracing::debug;
 use utoipa::ToSchema;

-use crate::{Datasets, PileValue, extract::MetaExtractor};
+use crate::Datasets;

 #[derive(Deserialize, ToSchema)]
 pub struct FieldQuery {
@@ -61,10 +62,8 @@ pub async fn get_field(
 		return StatusCode::NOT_FOUND.into_response();
 	};

-	let extractor = MetaExtractor::new(&item);
-	let root: PileValue = PileValue::ObjectExtractor(Arc::new(extractor));
-
-	let value = match root.query(&path).await {
+	let item = PileValue::Item(item);
+	let value = match item.query(&path).await {
 		Ok(Some(v)) => v,
 		Ok(None) => return StatusCode::NOT_FOUND.into_response(),
 		Err(e) => return (StatusCode::INTERNAL_SERVER_ERROR, format!("{e:?}")).into_response(),
--- a/crates/pile-dataset/src/serve/item.rs
+++ b/crates/pile-dataset/src/serve/item.rs
@@ -1,15 +1,19 @@
 use axum::{
+	body::Body,
 	extract::{Query, State},
-	http::{StatusCode, header},
+	http::{HeaderMap, StatusCode, header},
 	response::{IntoResponse, Response},
 };
 use pile_config::Label;
+use pile_value::value::{AsyncReader, AsyncSeekReader};
 use serde::Deserialize;
-use std::{sync::Arc, time::Instant};
+use std::{io::SeekFrom, sync::Arc, time::Instant};
+use tokio::sync::mpsc;
+use tokio_stream::wrappers::ReceiverStream;
 use tracing::debug;
 use utoipa::ToSchema;

-use crate::{AsyncReader, Datasets};
+use crate::Datasets;

 #[derive(Deserialize, ToSchema)]
 pub struct ItemQuery {
@@ -17,6 +21,24 @@ pub struct ItemQuery {
 	key: String,
 }

+/// Parse a `Range: bytes=...` header value.
+/// Returns `(start, end)` where either may be `None` (suffix form has `None` start).
+fn parse_byte_range(s: &str) -> Option<(Option<u64>, Option<u64>)> {
+	let spec = s.strip_prefix("bytes=")?;
+	if spec.contains(',') {
+		return None; // multiple ranges not supported
+	}
+	if let Some(suffix) = spec.strip_prefix('-') {
+		return Some((None, Some(suffix.parse().ok()?)));
+	}
+	let mut parts = spec.splitn(2, '-');
+	let start: u64 = parts.next()?.parse().ok()?;
+	let end = parts
+		.next()
+		.and_then(|e| if e.is_empty() { None } else { e.parse().ok() });
+	Some((Some(start), end))
+}
+
 /// Fetch the raw bytes of an item by source and key
 #[utoipa::path(
 	get,
@@ -27,14 +49,17 @@ pub struct ItemQuery {
 	),
 	responses(
 		(status = 200, description = "Raw item bytes"),
+		(status = 206, description = "Partial content"),
 		(status = 400, description = "Invalid source label"),
 		(status = 404, description = "Item not found"),
+		(status = 416, description = "Range not satisfiable"),
 		(status = 500, description = "Internal server error"),
 	)
 )]
 pub async fn item_get(
 	State(state): State<Arc<Datasets>>,
 	Query(params): Query<ItemQuery>,
+	headers: HeaderMap,
 ) -> Response {
 	let start = Instant::now();
 	debug!(
@@ -59,6 +84,43 @@ pub async fn item_get(
 		Err(e) => return (StatusCode::INTERNAL_SERVER_ERROR, format!("{e:?}")).into_response(),
 	};

+	let total = match reader.seek(SeekFrom::End(0)).await {
+		Ok(n) => n,
+		Err(e) => return (StatusCode::INTERNAL_SERVER_ERROR, format!("{e:?}")).into_response(),
+	};
+
+	let range = headers
+		.get(header::RANGE)
+		.and_then(|v| v.to_str().ok())
+		.and_then(parse_byte_range);
+
+	// Resolve (byte_start, byte_end, content_length, is_range)
+	let (byte_start, byte_end, length, is_range) = match range {
+		Some((Some(s), e)) => {
+			let e = e
+				.unwrap_or(total.saturating_sub(1))
+				.min(total.saturating_sub(1));
+			if s >= total || s > e {
+				return (
+					StatusCode::RANGE_NOT_SATISFIABLE,
+					[(header::CONTENT_RANGE, format!("bytes */{total}"))],
+				)
+					.into_response();
+			}
+			(s, e, e - s + 1, true)
+		}
+		Some((None, Some(suffix))) => {
+			let s = total.saturating_sub(suffix);
+			let e = total.saturating_sub(1);
+			(s, e, total.saturating_sub(s), true)
+		}
+		_ => (0, total.saturating_sub(1), total, false),
+	};
+
+	if let Err(e) = reader.seek(SeekFrom::Start(byte_start)).await {
+		return (StatusCode::INTERNAL_SERVER_ERROR, format!("{e:?}")).into_response();
+	}
+
 	debug!(
 		message = "Served /item",
 		source = params.source,
@@ -66,8 +128,54 @@ pub async fn item_get(
 		time_ms = start.elapsed().as_millis()
 	);

-	match reader.read_to_end().await {
-		Ok(bytes) => (StatusCode::OK, [(header::CONTENT_TYPE, mime)], bytes).into_response(),
-		Err(e) => (StatusCode::INTERNAL_SERVER_ERROR, format!("{e:?}")).into_response(),
+	let (tx, rx) = mpsc::channel::<Result<Vec<u8>, std::io::Error>>(8);
+
+	tokio::spawn(async move {
+		let mut buf = vec![0u8; 65536];
+		let mut remaining = length;
+		loop {
+			if remaining == 0 {
+				break;
+			}
+			let to_read = (buf.len() as u64).min(remaining) as usize;
+			match reader.read(&mut buf[..to_read]).await {
+				Ok(0) => break,
+				Ok(n) => {
+					remaining -= n as u64;
+					if tx.send(Ok(buf[..n].to_vec())).await.is_err() {
+						break;
+					}
+				}
+				Err(e) => {
+					let _ = tx.send(Err(e)).await;
+					break;
+				}
+			}
+		}
+	});
+
+	let body = Body::from_stream(ReceiverStream::new(rx));
+	let status = if is_range {
+		StatusCode::PARTIAL_CONTENT
+	} else {
+		StatusCode::OK
+	};
+
+	let mut builder = axum::http::Response::builder()
+		.status(status)
+		.header(header::CONTENT_TYPE, mime)
+		.header(header::ACCEPT_RANGES, "bytes")
+		.header(header::CONTENT_LENGTH, length);
+
+	if is_range {
+		builder = builder.header(
+			header::CONTENT_RANGE,
+			format!("bytes {byte_start}-{byte_end}/{total}"),
+		);
 	}
+
+	builder
+		.body(body)
+		.map(IntoResponse::into_response)
+		.unwrap_or_else(|e| (StatusCode::INTERNAL_SERVER_ERROR, format!("{e:?}")).into_response())
 }
--- a/crates/pile-dataset/src/source/mod.rs
+++ b/crates/pile-dataset/src/source/mod.rs
@@ -1,5 +0,0 @@
-mod dir;
-pub use dir::*;
-
-mod s3;
-pub use s3::*;
--- a/crates/pile-value/Cargo.toml
+++ b/crates/pile-value/Cargo.toml
@@ -0,0 +1,36 @@
+[package]
+name = "pile-value"
+version = { workspace = true }
+rust-version = { workspace = true }
+edition = { workspace = true }
+
+[lints]
+workspace = true
+
+[dependencies]
+pile-config = { workspace = true }
+pile-flac = { workspace = true }
+
+serde_json = { workspace = true }
+walkdir = { workspace = true }
+tracing = { workspace = true }
+chrono = { workspace = true }
+toml = { workspace = true }
+smartstring = { workspace = true }
+blake3 = { workspace = true }
+epub = { workspace = true }
+kamadak-exif = { workspace = true }
+pdf = { workspace = true }
+pdfium-render = { workspace = true, optional = true }
+image = { workspace = true, optional = true }
+id3 = { workspace = true }
+tokio = { workspace = true }
+tokio-stream = { workspace = true }
+async-trait = { workspace = true }
+aws-sdk-s3 = { workspace = true }
+mime = { workspace = true }
+mime_guess = { workspace = true }
+
+[features]
+default = []
+pdfium = ["dep:pdfium-render", "dep:image"]
--- a/crates/pile-dataset/build.rs
+++ b/crates/pile-dataset/build.rs
@@ -23,6 +23,13 @@ fn main() {
 		.expect("unexpected OUT_DIR structure")
 		.to_path_buf();

+	// If PDFIUM_LIB_DIR is set (e.g. by Nix), use the pre-installed library directly.
+	if let Ok(lib_dir) = env::var("PDFIUM_LIB_DIR") {
+		println!("cargo:rustc-link-search=native={lib_dir}");
+		println!("cargo:rustc-link-lib=dylib=pdfium");
+		return;
+	}
+
 	let lib_path = profile_dir.join("libpdfium.so");

 	if !lib_path.exists() {
--- a/crates/pile-value/src/extract/item/epub/epub_meta.rs
+++ b/crates/pile-value/src/extract/item/epub/epub_meta.rs
@@ -6,7 +6,10 @@ use std::{
 };
 use tracing::trace;

-use crate::{Item, PileValue, SyncReadBridge, extract::ObjectExtractor};
+use crate::{
+	extract::traits::ObjectExtractor,
+	value::{Item, PileValue, SyncReadBridge},
+};

 pub struct EpubMetaExtractor {
 	item: Item,
--- a/crates/pile-value/src/extract/item/epub/epub_text.rs
+++ b/crates/pile-value/src/extract/item/epub/epub_text.rs
@@ -6,7 +6,10 @@ use std::{
 };
 use tracing::debug;

-use crate::{Item, PileValue, SyncReadBridge, extract::ObjectExtractor};
+use crate::{
+	extract::traits::ObjectExtractor,
+	value::{Item, PileValue, SyncReadBridge},
+};

 pub struct EpubTextExtractor {
 	item: Item,
--- a/crates/pile-value/src/extract/item/epub/mod.rs
+++ b/crates/pile-value/src/extract/item/epub/mod.rs
@@ -7,7 +7,10 @@ pub use epub_meta::*;
 mod epub_text;
 pub use epub_text::*;

-use crate::{Item, PileValue, extract::ObjectExtractor};
+use crate::{
+	extract::traits::ObjectExtractor,
+	value::{Item, PileValue},
+};

 pub struct EpubExtractor {
 	text: Arc<EpubTextExtractor>,
--- a/crates/pile-value/src/extract/item/exif.rs
+++ b/crates/pile-value/src/extract/item/exif.rs
@@ -6,7 +6,10 @@ use std::{
 };
 use tracing::trace;

-use crate::{Item, PileValue, SyncReadBridge, extract::ObjectExtractor};
+use crate::{
+	extract::traits::ObjectExtractor,
+	value::{Item, PileValue, SyncReadBridge},
+};

 pub struct ExifExtractor {
 	item: Item,
--- a/crates/pile-value/src/extract/item/flac.rs
+++ b/crates/pile-value/src/extract/item/flac.rs
@@ -8,8 +8,8 @@ use std::{
 };

 use crate::{
-	Item, PileValue, SyncReadBridge,
-	extract::{ListExtractor, ObjectExtractor},
+	extract::traits::{ListExtractor, ObjectExtractor},
+	value::{Item, PileValue, SyncReadBridge},
 };

 pub struct FlacImagesExtractor {
--- a/crates/pile-value/src/extract/item/fs.rs
+++ b/crates/pile-value/src/extract/item/fs.rs
@@ -5,7 +5,10 @@ use std::{
 	sync::{Arc, OnceLock},
 };

-use crate::{Item, PileValue, extract::ObjectExtractor};
+use crate::{
+	extract::traits::ObjectExtractor,
+	value::{Item, PileValue},
+};

 pub struct FsExtractor {
 	item: Item,
--- a/crates/pile-value/src/extract/item/id3.rs
+++ b/crates/pile-value/src/extract/item/id3.rs
@@ -7,7 +7,10 @@ use std::{
 	sync::{Arc, OnceLock},
 };

-use crate::{Item, PileValue, SyncReadBridge, extract::ObjectExtractor};
+use crate::{
+	extract::traits::ObjectExtractor,
+	value::{Item, PileValue, SyncReadBridge},
+};

 pub struct Id3Extractor {
 	item: Item,
--- a/crates/pile-value/src/extract/item/mod.rs
+++ b/crates/pile-value/src/extract/item/mod.rs
@@ -0,0 +1,99 @@
+mod flac;
+use std::{collections::HashMap, sync::Arc};
+
+pub use flac::*;
+
+mod id3;
+pub use id3::*;
+
+mod fs;
+pub use fs::*;
+
+mod epub;
+pub use epub::*;
+
+mod exif;
+pub use exif::*;
+
+mod pdf;
+pub use pdf::*;
+
+mod toml;
+use pile_config::Label;
+pub use toml::*;
+
+mod sidecar;
+pub use sidecar::*;
+
+use crate::{
+	extract::{misc::MapExtractor, traits::ObjectExtractor},
+	value::{Item, PileValue},
+};
+
+pub struct ItemExtractor {
+	inner: MapExtractor,
+}
+
+impl ItemExtractor {
+	#[expect(clippy::unwrap_used)]
+	pub fn new(item: &Item) -> Self {
+		let inner = MapExtractor {
+			inner: HashMap::from([
+				(
+					Label::new("flac").unwrap(),
+					PileValue::ObjectExtractor(Arc::new(FlacExtractor::new(item))),
+				),
+				(
+					Label::new("id3").unwrap(),
+					PileValue::ObjectExtractor(Arc::new(Id3Extractor::new(item))),
+				),
+				(
+					Label::new("fs").unwrap(),
+					PileValue::ObjectExtractor(Arc::new(FsExtractor::new(item))),
+				),
+				(
+					Label::new("epub").unwrap(),
+					PileValue::ObjectExtractor(Arc::new(EpubExtractor::new(item))),
+				),
+				(
+					Label::new("exif").unwrap(),
+					PileValue::ObjectExtractor(Arc::new(ExifExtractor::new(item))),
+				),
+				(
+					Label::new("pdf").unwrap(),
+					PileValue::ObjectExtractor(Arc::new(PdfExtractor::new(item))),
+				),
+				(
+					Label::new("toml").unwrap(),
+					PileValue::ObjectExtractor(Arc::new(TomlExtractor::new(item))),
+				),
+				(
+					Label::new("sidecar").unwrap(),
+					PileValue::ObjectExtractor(Arc::new(SidecarExtractor::new(item))),
+				),
+			]),
+		};
+
+		Self { inner }
+	}
+}
+
+#[async_trait::async_trait]
+impl ObjectExtractor for ItemExtractor {
+	async fn field(&self, name: &pile_config::Label) -> Result<Option<PileValue>, std::io::Error> {
+		self.inner.field(name).await
+	}
+
+	#[expect(clippy::unwrap_used)]
+	async fn fields(&self) -> Result<Vec<Label>, std::io::Error> {
+		return Ok(vec![
+			Label::new("flac").unwrap(),
+			Label::new("id3").unwrap(),
+			Label::new("fs").unwrap(),
+			Label::new("epub").unwrap(),
+			Label::new("exif").unwrap(),
+			Label::new("pdf").unwrap(),
+			Label::new("sidecar").unwrap(),
+		]);
+	}
+}
--- a/crates/pile-value/src/extract/item/pdf/mod.rs
+++ b/crates/pile-value/src/extract/item/pdf/mod.rs
@@ -1,11 +1,6 @@
 use pile_config::Label;
 use std::sync::Arc;

-#[cfg(feature = "pdfium")]
-mod pdf_cover;
-#[cfg(feature = "pdfium")]
-pub use pdf_cover::*;
-
 #[cfg(feature = "pdfium")]
 mod pdf_pages;
 #[cfg(feature = "pdfium")]
@@ -17,14 +12,15 @@ pub use pdf_meta::*;
 mod pdf_text;
 pub use pdf_text::*;

-use crate::{Item, PileValue, extract::ObjectExtractor};
+use crate::{
+	extract::traits::ObjectExtractor,
+	value::{Item, PileValue},
+};

 pub struct PdfExtractor {
 	text: Arc<PdfTextExtractor>,
 	meta: Arc<PdfMetaExtractor>,
 	#[cfg(feature = "pdfium")]
-	cover: Arc<PdfCoverExtractor>,
-	#[cfg(feature = "pdfium")]
 	pages: Arc<PdfPagesExtractor>,
 }

@@ -34,8 +30,6 @@ impl PdfExtractor {
 			text: Arc::new(PdfTextExtractor::new(item)),
 			meta: Arc::new(PdfMetaExtractor::new(item)),
 			#[cfg(feature = "pdfium")]
-			cover: Arc::new(PdfCoverExtractor::new(item)),
-			#[cfg(feature = "pdfium")]
 			pages: Arc::new(PdfPagesExtractor::new(item)),
 		}
 	}
@@ -48,8 +42,6 @@ impl ObjectExtractor for PdfExtractor {
 			"text" => self.text.field(name).await,
 			"meta" => Ok(Some(PileValue::ObjectExtractor(self.meta.clone()))),
 			#[cfg(feature = "pdfium")]
-			"cover" => self.cover.field(name).await,
-			#[cfg(feature = "pdfium")]
 			"pages" => Ok(Some(PileValue::ListExtractor(self.pages.clone()))),
 			_ => Ok(None),
 		}
--- a/crates/pile-value/src/extract/item/pdf/pdf_meta.rs
+++ b/crates/pile-value/src/extract/item/pdf/pdf_meta.rs
@@ -8,8 +8,10 @@ use std::{
 };
 use tracing::trace;

-use crate::extract::ObjectExtractor;
-use crate::{Item, PileValue, SyncReadBridge};
+use crate::{
+	extract::traits::ObjectExtractor,
+	value::{Item, PileValue, SyncReadBridge},
+};

 pub struct PdfMetaExtractor {
 	item: Item,
--- a/crates/pile-value/src/extract/item/pdf/pdf_pages.rs
+++ b/crates/pile-value/src/extract/item/pdf/pdf_pages.rs
@@ -6,7 +6,10 @@ use std::{
 };
 use tracing::trace;

-use crate::{Item, PileValue, SyncReadBridge, extract::ListExtractor};
+use crate::{
+	extract::traits::ListExtractor,
+	value::{Item, PileValue, SyncReadBridge},
+};

 pub struct PdfPagesExtractor {
 	item: Item,
--- a/crates/pile-value/src/extract/item/pdf/pdf_text.rs
+++ b/crates/pile-value/src/extract/item/pdf/pdf_text.rs
@@ -8,8 +8,10 @@ use std::{
 };
 use tracing::trace;

-use crate::extract::ObjectExtractor;
-use crate::{Item, PileValue, SyncReadBridge};
+use crate::{
+	extract::traits::ObjectExtractor,
+	value::{Item, PileValue, SyncReadBridge},
+};

 pub struct PdfTextExtractor {
 	item: Item,
--- a/crates/pile-value/src/extract/item/sidecar.rs
+++ b/crates/pile-value/src/extract/item/sidecar.rs
@@ -1,9 +1,10 @@
 use pile_config::Label;
 use std::sync::OnceLock;

+use super::TomlExtractor;
 use crate::{
-	Item, PileValue,
-	extract::{ObjectExtractor, TomlExtractor},
+	extract::traits::ObjectExtractor,
+	value::{Item, PileValue},
 };

 pub struct SidecarExtractor {
--- a/crates/pile-value/src/extract/item/toml.rs
+++ b/crates/pile-value/src/extract/item/toml.rs
@@ -4,7 +4,10 @@ use std::{
 	sync::{Arc, OnceLock},
 };

-use crate::{AsyncReader, Item, PileValue, extract::ObjectExtractor};
+use crate::{
+	extract::traits::ObjectExtractor,
+	value::{AsyncReader, Item, PileValue},
+};

 fn toml_to_pile(value: toml::Value) -> PileValue {
 	match value {
--- a/crates/pile-value/src/extract/misc/list.rs
+++ b/crates/pile-value/src/extract/misc/list.rs
@@ -0,0 +1,24 @@
+use std::sync::Arc;
+
+use crate::{extract::traits::ListExtractor, value::PileValue};
+
+pub struct ArrayExtractor {
+	inner: Arc<Vec<PileValue>>,
+}
+
+impl ArrayExtractor {
+	pub fn new(inner: Arc<Vec<PileValue>>) -> Self {
+		Self { inner }
+	}
+}
+
+#[async_trait::async_trait]
+impl ListExtractor for ArrayExtractor {
+	async fn get(&self, idx: usize) -> Result<Option<PileValue>, std::io::Error> {
+		Ok(self.inner.get(idx).cloned())
+	}
+
+	async fn len(&self) -> Result<usize, std::io::Error> {
+		Ok(self.inner.len())
+	}
+}
--- a/crates/pile-value/src/extract/misc/map.rs
+++ b/crates/pile-value/src/extract/misc/map.rs
@@ -1,10 +1,11 @@
 use pile_config::Label;
 use std::collections::HashMap;

-use crate::{PileValue, extract::ObjectExtractor};
+use crate::{extract::traits::ObjectExtractor, value::PileValue};

+#[derive(Default)]
 pub struct MapExtractor {
-	pub(crate) inner: HashMap<Label, PileValue>,
+	pub inner: HashMap<Label, PileValue>,
 }

 #[async_trait::async_trait]
--- a/crates/pile-value/src/extract/misc/mod.rs
+++ b/crates/pile-value/src/extract/misc/mod.rs
@@ -0,0 +1,8 @@
+mod list;
+pub use list::*;
+
+mod vec;
+pub use vec::*;
+
+mod map;
+pub use map::*;
--- a/crates/pile-value/src/extract/misc/vec.rs
+++ b/crates/pile-value/src/extract/misc/vec.rs
@@ -0,0 +1,17 @@
+use crate::{extract::traits::ListExtractor, value::PileValue};
+
+#[derive(Default)]
+pub struct VecExtractor {
+	pub inner: Vec<PileValue>,
+}
+
+#[async_trait::async_trait]
+impl ListExtractor for VecExtractor {
+	async fn get(&self, idx: usize) -> Result<Option<PileValue>, std::io::Error> {
+		Ok(self.inner.get(idx).cloned())
+	}
+
+	async fn len(&self) -> Result<usize, std::io::Error> {
+		Ok(self.inner.len())
+	}
+}
--- a/crates/pile-value/src/extract/mod.rs
+++ b/crates/pile-value/src/extract/mod.rs
@@ -0,0 +1,4 @@
+pub mod item;
+pub mod misc;
+pub mod string;
+pub mod traits;
--- a/crates/pile-value/src/extract/string.rs
+++ b/crates/pile-value/src/extract/string.rs
@@ -0,0 +1,221 @@
+use pile_config::Label;
+use smartstring::{LazyCompact, SmartString};
+use std::sync::Arc;
+
+use crate::{extract::traits::ObjectExtractor, value::PileValue};
+
+fn parse_name(s: &str) -> (&str, Option<&str>) {
+	match s.find('(') {
+		None => (s, None),
+		Some(i) => {
+			let name = &s[..i];
+			let rest = &s[i + 1..];
+			match rest.strip_suffix(')') {
+				Some(args) => (name, Some(args)),
+				None => (name, None),
+			}
+		}
+	}
+}
+
+pub struct StringExtractor {
+	item: Arc<SmartString<LazyCompact>>,
+}
+
+impl StringExtractor {
+	pub fn new(item: &Arc<SmartString<LazyCompact>>) -> Self {
+		Self { item: item.clone() }
+	}
+}
+
+#[async_trait::async_trait]
+impl ObjectExtractor for StringExtractor {
+	async fn field(&self, name: &Label) -> Result<Option<PileValue>, std::io::Error> {
+		let (name, args) = parse_name(name.as_str());
+		Ok(match (name, args) {
+			("trim", None) => Some(PileValue::String(Arc::new(
+				self.item.as_str().trim().into(),
+			))),
+
+			("upper", None) => Some(PileValue::String(Arc::new(
+				self.item.as_str().to_lowercase().into(),
+			))),
+
+			("lower", None) => Some(PileValue::String(Arc::new(
+				self.item.as_str().to_uppercase().into(),
+			))),
+
+			("nonempty", None) => Some(match self.item.is_empty() {
+				true => PileValue::Null,
+				false => PileValue::String(self.item.clone()),
+			}),
+
+			("trimprefix", Some(prefix)) => Some(PileValue::String(Arc::new(
+				self.item
+					.as_str()
+					.strip_prefix(prefix)
+					.unwrap_or(self.item.as_str())
+					.into(),
+			))),
+
+			("trimsuffix", Some(suffix)) => Some(PileValue::String(Arc::new(
+				self.item
+					.as_str()
+					.strip_suffix(suffix)
+					.unwrap_or(self.item.as_str())
+					.into(),
+			))),
+
+			("split", Some(by)) => Some(PileValue::Array(Arc::new(
+				self.item
+					.as_str()
+					.split(by)
+					.map(|s| PileValue::String(Arc::new(s.into())))
+					.collect(),
+			))),
+
+			_ => None,
+		})
+	}
+
+	#[expect(clippy::unwrap_used)]
+	async fn fields(&self) -> Result<Vec<Label>, std::io::Error> {
+		return Ok(vec![
+			Label::new("trim").unwrap(),
+			Label::new("upper").unwrap(),
+			Label::new("lower").unwrap(),
+			Label::new("nonempty").unwrap(),
+		]);
+	}
+}
+
+#[cfg(test)]
+mod tests {
+	use super::*;
+
+	fn extractor(s: &str) -> StringExtractor {
+		StringExtractor::new(&Arc::new(s.into()))
+	}
+
+	#[expect(clippy::unwrap_used)]
+	async fn field(ext: &StringExtractor, name: &str) -> Option<PileValue> {
+		ext.field(&Label::new(name).unwrap()).await.unwrap()
+	}
+
+	fn string(v: Option<PileValue>) -> Option<String> {
+		match v? {
+			PileValue::String(s) => Some(s.as_str().to_owned()),
+			_ => panic!("expected string"),
+		}
+	}
+
+	fn array(v: Option<PileValue>) -> Vec<String> {
+		match v.expect("expected Some") {
+			PileValue::Array(arr) => arr
+				.iter()
+				.map(|v| match v {
+					PileValue::String(s) => s.as_str().to_owned(),
+					_ => panic!("expected string element"),
+				})
+				.collect(),
+			_ => panic!("expected array"),
+		}
+	}
+
+	#[tokio::test]
+	async fn trim() {
+		assert_eq!(
+			string(field(&extractor("  hi  "), "trim").await),
+			Some("hi".into())
+		);
+	}
+
+	#[tokio::test]
+	async fn trim_no_args() {
+		assert!(field(&extractor("x"), "trim(foo)").await.is_none());
+	}
+
+	#[tokio::test]
+	async fn nonempty_with_content() {
+		assert!(matches!(
+			field(&extractor("hello"), "nonempty").await,
+			Some(PileValue::String(_))
+		));
+	}
+
+	#[tokio::test]
+	async fn nonempty_empty_string() {
+		assert!(matches!(
+			field(&extractor(""), "nonempty").await,
+			Some(PileValue::Null)
+		));
+	}
+
+	#[tokio::test]
+	async fn trimprefix_present() {
+		assert_eq!(
+			string(field(&extractor("foobar"), "trimprefix(foo)").await),
+			Some("bar".into())
+		);
+	}
+
+	#[tokio::test]
+	async fn trimprefix_absent() {
+		assert_eq!(
+			string(field(&extractor("foobar"), "trimprefix(baz)").await),
+			Some("foobar".into())
+		);
+	}
+
+	#[tokio::test]
+	async fn trimprefix_no_args() {
+		assert!(field(&extractor("foobar"), "trimprefix").await.is_none());
+	}
+
+	#[tokio::test]
+	async fn trimsuffix_present() {
+		assert_eq!(
+			string(field(&extractor("foobar"), "trimsuffix(bar)").await),
+			Some("foo".into())
+		);
+	}
+
+	#[tokio::test]
+	async fn trimsuffix_absent() {
+		assert_eq!(
+			string(field(&extractor("foobar"), "trimsuffix(baz)").await),
+			Some("foobar".into())
+		);
+	}
+
+	#[tokio::test]
+	async fn split_basic() {
+		assert_eq!(
+			array(field(&extractor("a,b,c"), "split(,)").await),
+			vec!["a", "b", "c"]
+		);
+	}
+
+	#[tokio::test]
+	async fn split_no_match() {
+		assert_eq!(
+			array(field(&extractor("abc"), "split(,)").await),
+			vec!["abc"]
+		);
+	}
+
+	#[tokio::test]
+	async fn split_no_args() {
+		assert!(field(&extractor("abc"), "split").await.is_none());
+	}
+
+	#[tokio::test]
+	async fn split_unclosed_paren() {
+		assert!(field(&extractor("abc"), "split(,").await.is_none());
+	}
+
+	#[tokio::test]
+	async fn unknown_field() {
+		assert!(field(&extractor("abc"), "bogus").await.is_none());
+	}
+}
--- a/crates/pile-value/src/extract/traits.rs
+++ b/crates/pile-value/src/extract/traits.rs
@@ -0,0 +1,68 @@
+/// An attachment that extracts metadata from an [Item].
+///
+/// Metadata is exposed as an immutable map of {label: value},
+/// much like a json object.
+#[async_trait::async_trait]
+pub trait ObjectExtractor: Send + Sync {
+	/// Get the field at `name` from `item`.
+	/// - returns `None` if `name` is not a valid field
+	/// - returns `Some(Null)` if `name` is not available
+	async fn field(
+		&self,
+		name: &pile_config::Label,
+	) -> Result<Option<crate::value::PileValue>, std::io::Error>;
+
+	/// Return all fields in this extractor.
+	/// `Self::field` must return [Some] for all these keys
+	/// and [None] for all others.
+	async fn fields(&self) -> Result<Vec<pile_config::Label>, std::io::Error>;
+
+	/// Convert this to a JSON value.
+	async fn to_json(&self) -> Result<serde_json::Value, std::io::Error> {
+		let keys = self.fields().await?;
+		let mut map = serde_json::Map::new();
+		for k in &keys {
+			let v = match self.field(k).await? {
+				Some(x) => x,
+				None => continue,
+			};
+			map.insert(k.to_string(), Box::pin(v.to_json()).await?);
+		}
+
+		Ok(serde_json::Value::Object(map))
+	}
+}
+
+/// An attachment that extracts metadata from an [Item].
+///
+/// Metadata is exposed as an immutable list of values.
+#[async_trait::async_trait]
+pub trait ListExtractor: Send + Sync {
+	/// Get the item at index `idx`.
+	/// Indices start at zero, and must be consecutive.
+	/// - returns `None` if `idx` is out of range
+	/// - returns `Some(Null)` if `None` is at `idx`
+	async fn get(&self, idx: usize) -> Result<Option<crate::value::PileValue>, std::io::Error>;
+
+	async fn len(&self) -> Result<usize, std::io::Error>;
+
+	async fn is_empty(&self) -> Result<bool, std::io::Error> {
+		Ok(self.len().await? == 0)
+	}
+
+	/// Convert this list to a JSON value.
+	async fn to_json(&self) -> Result<serde_json::Value, std::io::Error> {
+		let len = self.len().await?;
+		let mut list = Vec::with_capacity(len);
+		for i in 0..len {
+			#[expect(clippy::expect_used)]
+			let v = self
+				.get(i)
+				.await?
+				.expect("value must be present according to length");
+			list.push(Box::pin(v.to_json()).await?);
+		}
+
+		Ok(serde_json::Value::Array(list))
+	}
+}
--- a/crates/pile-value/src/lib.rs
+++ b/crates/pile-value/src/lib.rs
@@ -0,0 +1,3 @@
+pub mod extract;
+pub mod source;
+pub mod value;
--- a/crates/pile-dataset/src/source/dir.rs
+++ b/crates/pile-dataset/src/source/dir.rs
@@ -4,7 +4,10 @@ use std::{path::PathBuf, sync::Arc};
 use tokio_stream::wrappers::ReceiverStream;
 use walkdir::WalkDir;

-use crate::{DataSource, Item, path_ts_latest};
+use crate::{
+	source::{DataSource, misc::path_ts_latest},
+	value::Item,
+};

 #[derive(Debug)]
 pub struct DirDataSource {
--- a/crates/pile-value/src/source/misc.rs
+++ b/crates/pile-value/src/source/misc.rs
--- a/crates/pile-value/src/source/mod.rs
+++ b/crates/pile-value/src/source/mod.rs
@@ -1,15 +1,24 @@
+mod dir;
+pub use dir::*;
+
+mod s3;
+pub use s3::*;
+
+pub mod misc;
+
 use chrono::{DateTime, Utc};
 use tokio_stream::wrappers::ReceiverStream;

-use crate::Item;
-
 /// A read-only set of [Item]s.
 pub trait DataSource {
 	/// Get an item from this datasource
-	fn get(&self, key: &str) -> impl Future<Output = Result<Option<Item>, std::io::Error>> + Send;
+	fn get(
+		&self,
+		key: &str,
+	) -> impl Future<Output = Result<Option<crate::value::Item>, std::io::Error>> + Send;

 	/// Iterate over all items in this source in an arbitrary order
-	fn iter(&self) -> ReceiverStream<Result<Item, std::io::Error>>;
+	fn iter(&self) -> ReceiverStream<Result<crate::value::Item, std::io::Error>>;

 	/// Return the time of the latest change to the data in this source
 	fn latest_change(
--- a/crates/pile-dataset/src/source/s3.rs
+++ b/crates/pile-dataset/src/source/s3.rs
@@ -5,7 +5,7 @@ use smartstring::{LazyCompact, SmartString};
 use std::sync::Arc;
 use tokio_stream::wrappers::ReceiverStream;

-use crate::{DataSource, Item};
+use crate::{source::DataSource, value::Item};

 #[derive(Debug)]
 pub struct S3DataSource {
--- a/crates/pile-dataset/src/source/s3reader.rs
+++ b/crates/pile-dataset/src/source/s3reader.rs
--- a/crates/pile-value/src/value/item.rs
+++ b/crates/pile-value/src/value/item.rs
@@ -0,0 +1,105 @@
+use mime::Mime;
+use smartstring::{LazyCompact, SmartString};
+use std::{fs::File, path::PathBuf, sync::Arc};
+
+use crate::{
+	source::{DirDataSource, S3DataSource},
+	value::{ItemReader, S3Reader},
+};
+
+//
+// MARK: item
+//
+
+/// A cheaply-cloneable pointer to an item in a dataset
+#[derive(Debug, Clone)]
+pub enum Item {
+	File {
+		source: Arc<DirDataSource>,
+		mime: Mime,
+
+		path: PathBuf,
+		sidecar: Option<Box<Item>>,
+	},
+
+	S3 {
+		source: Arc<S3DataSource>,
+		mime: Mime,
+
+		key: SmartString<LazyCompact>,
+		sidecar: Option<Box<Item>>,
+	},
+}
+
+impl Item {
+	/// Open the item for reading. For S3, performs a HEAD request to determine
+	/// the object size.
+	pub async fn read(&self) -> Result<ItemReader, std::io::Error> {
+		Ok(match self {
+			Self::File { path, .. } => ItemReader::File(File::open(path)?),
+
+			Self::S3 { source, key, .. } => {
+				let head = source
+					.client
+					.head_object()
+					.bucket(source.bucket.as_str())
+					.key(key.as_str())
+					.send()
+					.await
+					.map_err(std::io::Error::other)?;
+
+				let size = head.content_length().unwrap_or(0) as u64;
+
+				ItemReader::S3(S3Reader {
+					client: source.client.clone(),
+					bucket: source.bucket.clone(),
+					key: key.to_owned(),
+					cursor: 0,
+					size,
+				})
+			}
+		})
+	}
+
+	pub fn source_name(&self) -> &pile_config::Label {
+		match self {
+			Self::File { source, .. } => &source.name,
+			Self::S3 { source, .. } => &source.name,
+		}
+	}
+
+	#[expect(clippy::expect_used)]
+	pub fn key(&self) -> SmartString<LazyCompact> {
+		match self {
+			Self::File { path, .. } => path.to_str().expect("path is not utf-8").into(),
+			Self::S3 { key, .. } => key.clone(),
+		}
+	}
+
+	pub fn hash(&self) -> Result<blake3::Hash, std::io::Error> {
+		match self {
+			Self::File { path, .. } => {
+				let mut hasher = blake3::Hasher::new();
+				let mut file = std::fs::File::open(path)?;
+				std::io::copy(&mut file, &mut hasher)?;
+				return Ok(hasher.finalize());
+			}
+
+			Self::S3 { .. } => todo!(),
+		}
+	}
+
+	pub fn mime(&self) -> &Mime {
+		match self {
+			Self::File { mime, .. } => mime,
+			Self::S3 { mime, .. } => mime,
+		}
+	}
+
+	pub fn sidecar(&self) -> Option<&Self> {
+		match self {
+			Self::File { sidecar, .. } => sidecar.as_ref().map(|x| &**x),
+			Self::S3 { sidecar, .. } => sidecar.as_ref().map(|x| &**x),
+		}
+	}
+}
--- a/crates/pile-value/src/value/mod.rs
+++ b/crates/pile-value/src/value/mod.rs
@@ -0,0 +1,9 @@
+mod item;
+pub use item::*;
+
+mod readers;
+pub use readers::*;
+
+#[expect(clippy::module_inception)]
+mod value;
+pub use value::*;
--- a/crates/pile-value/src/value/readers.rs
+++ b/crates/pile-value/src/value/readers.rs
@@ -1,114 +1,13 @@
-use mime::Mime;
 use smartstring::{LazyCompact, SmartString};
 use std::{
 	fs::File,
 	io::{Read, Seek, SeekFrom},
-	path::PathBuf,
 	sync::Arc,
 };
 use tokio::runtime::Handle;

-use crate::source::{DirDataSource, S3DataSource};
-
 //
-// MARK: item
-//
-
-/// A cheaply-clonable pointer to an item in a dataset
-#[derive(Debug, Clone)]
-pub enum Item {
-	File {
-		source: Arc<DirDataSource>,
-		mime: Mime,
-
-		path: PathBuf,
-		sidecar: Option<Box<Item>>,
-	},
-
-	S3 {
-		source: Arc<S3DataSource>,
-		mime: Mime,
-
-		key: SmartString<LazyCompact>,
-		sidecar: Option<Box<Item>>,
-	},
-}
-
-impl Item {
-	/// Open the item for reading. For S3, performs a HEAD request to determine
-	/// the object size.
-	pub async fn read(&self) -> Result<ItemReader, std::io::Error> {
-		Ok(match self {
-			Self::File { path, .. } => ItemReader::File(File::open(path)?),
-
-			Self::S3 { source, key, .. } => {
-				let head = source
-					.client
-					.head_object()
-					.bucket(source.bucket.as_str())
-					.key(key.as_str())
-					.send()
-					.await
-					.map_err(std::io::Error::other)?;
-
-				let size = head.content_length().unwrap_or(0) as u64;
-
-				ItemReader::S3(S3Reader {
-					client: source.client.clone(),
-					bucket: source.bucket.clone(),
-					key: key.to_owned(),
-					cursor: 0,
-					size,
-				})
-			}
-		})
-	}
-
-	pub fn source_name(&self) -> &pile_config::Label {
-		match self {
-			Self::File { source, .. } => &source.name,
-			Self::S3 { source, .. } => &source.name,
-		}
-	}
-
-	#[expect(clippy::expect_used)]
-	pub fn key(&self) -> SmartString<LazyCompact> {
-		match self {
-			Self::File { path, .. } => path.to_str().expect("path is not utf-8").into(),
-			Self::S3 { key, .. } => key.clone(),
-		}
-	}
-
-	pub fn hash(&self) -> Result<blake3::Hash, std::io::Error> {
-		match self {
-			Self::File { path, .. } => {
-				let mut hasher = blake3::Hasher::new();
-				let mut file = std::fs::File::open(path)?;
-				std::io::copy(&mut file, &mut hasher)?;
-				return Ok(hasher.finalize());
-			}
-
-			Self::S3 { .. } => todo!(),
-		}
-	}
-
-	pub fn mime(&self) -> &Mime {
-		match self {
-			Self::File { mime, .. } => mime,
-			Self::S3 { mime, .. } => mime,
-		}
-	}
-
-	pub fn sidecar(&self) -> Option<&Self> {
-		match self {
-			Self::File { sidecar, .. } => sidecar.as_ref().map(|x| &**x),
-			Self::S3 { sidecar, .. } => sidecar.as_ref().map(|x| &**x),
-		}
-	}
-}
-
-//
-// MARK: reader
+// MARK: traits
 //

 pub trait AsyncReader: Send {
@@ -210,11 +109,11 @@ impl AsyncSeekReader for ItemReader {
 //

 pub struct S3Reader {
-	client: Arc<aws_sdk_s3::Client>,
-	bucket: SmartString<LazyCompact>,
-	key: SmartString<LazyCompact>,
-	cursor: u64,
-	size: u64,
+	pub client: Arc<aws_sdk_s3::Client>,
+	pub bucket: SmartString<LazyCompact>,
+	pub key: SmartString<LazyCompact>,
+	pub cursor: u64,
+	pub size: u64,
 }

 impl AsyncReader for S3Reader {
--- a/crates/pile-value/src/value/value.rs
+++ b/crates/pile-value/src/value/value.rs
@@ -4,9 +4,17 @@ use serde_json::{Map, Value};
 use smartstring::{LazyCompact, SmartString};
 use std::sync::Arc;

-use crate::extract::{ListExtractor, ObjectExtractor};
+use crate::{
+	extract::{
+		item::ItemExtractor,
+		misc::{ArrayExtractor, MapExtractor, VecExtractor},
+		string::StringExtractor,
+		traits::{ListExtractor, ObjectExtractor},
+	},
+	value::Item,
+};

-/// An immutable, cheaply-clonable, lazily-computed value.
+/// An immutable, cheaply-cloneable, lazily-computed value.
 /// Very similar to [serde_json::Value].
 pub enum PileValue {
 	Null,
@@ -30,6 +38,9 @@ pub enum PileValue {

 	/// A lazily-computed array
 	ListExtractor(Arc<dyn ListExtractor>),
+
+	/// An pointer to an item in this dataset
+	Item(Item),
 }

 impl Clone for PileValue {
@@ -46,11 +57,40 @@ impl Clone for PileValue {
 				mime: mime.clone(),
 				bytes: bytes.clone(),
 			},
+			Self::Item(i) => Self::Item(i.clone()),
 		}
 	}
 }

 impl PileValue {
+	pub fn object_extractor(&self) -> Arc<dyn ObjectExtractor> {
+		match self {
+			Self::Null => Arc::new(MapExtractor::default()),
+			Self::U64(_) => Arc::new(MapExtractor::default()),
+			Self::I64(_) => Arc::new(MapExtractor::default()),
+			Self::Array(_) => Arc::new(MapExtractor::default()),
+			Self::String(s) => Arc::new(StringExtractor::new(s)),
+			Self::Blob { .. } => Arc::new(MapExtractor::default()),
+			Self::ListExtractor(_) => Arc::new(MapExtractor::default()),
+			Self::ObjectExtractor(e) => e.clone(),
+			Self::Item(i) => Arc::new(ItemExtractor::new(i)),
+		}
+	}
+
+	pub fn list_extractor(&self) -> Arc<dyn ListExtractor> {
+		match self {
+			Self::Null => Arc::new(VecExtractor::default()),
+			Self::U64(_) => Arc::new(VecExtractor::default()),
+			Self::I64(_) => Arc::new(VecExtractor::default()),
+			Self::Array(a) => Arc::new(ArrayExtractor::new(a.clone())),
+			Self::String(_) => Arc::new(VecExtractor::default()),
+			Self::Blob { .. } => Arc::new(VecExtractor::default()),
+			Self::ListExtractor(e) => e.clone(),
+			Self::ObjectExtractor(_) => Arc::new(VecExtractor::default()),
+			Self::Item(_) => Arc::new(VecExtractor::default()),
+		}
+	}
+
 	pub async fn query(&self, query: &ObjectPath) -> Result<Option<Self>, std::io::Error> {
 		let mut out: Option<PileValue> = Some(self.clone());

@@ -58,50 +98,41 @@ impl PileValue {
 			match s {
 				PathSegment::Root => out = Some(self.clone()),
 				PathSegment::Field(field) => {
-					out = match out {
-						None => return Ok(None),
-						Some(Self::Null) => None,
-						Some(Self::U64(_)) => None,
-						Some(Self::I64(_)) => None,
-						Some(Self::Array(_)) => None,
-						Some(Self::String(_)) => None,
-						Some(Self::Blob { .. }) => None,
-						Some(Self::ListExtractor(_)) => None,
-						Some(Self::ObjectExtractor(e)) => e.field(field).await?,
-					}
+					let e = match out.map(|x| x.object_extractor()) {
+						Some(e) => e,
+						None => {
+							out = None;
+							continue;
+						}
+					};
+
+					out = e.field(field).await?;
 				}

 				PathSegment::Index(idx) => {
-					out = match &out {
-						None => return Ok(None),
-						Some(Self::Null) => None,
-						Some(Self::U64(_)) => None,
-						Some(Self::I64(_)) => None,
-						Some(Self::Blob { .. }) => None,
-						Some(Self::Array(v)) => {
-							let idx = if *idx >= 0 {
-								usize::try_from(*idx).ok()
-							} else {
-								usize::try_from(v.len() as i64 - idx).ok()
-							};
-
-							idx.and_then(|idx| v.get(idx)).cloned()
+					let e = match out.map(|x| x.list_extractor()) {
+						Some(e) => e,
+						None => {
+							out = None;
+							continue;
 						}
-						Some(Self::String(_)) => None,
-						Some(Self::ObjectExtractor(_)) => None,
-						Some(Self::ListExtractor(e)) => {
-							let idx = if *idx >= 0 {
-								usize::try_from(*idx).ok()
-							} else {
-								usize::try_from(e.len().await? as i64 - idx).ok()
-							};
+					};

-							match idx {
-								Some(idx) => e.get(idx).await?,
-								None => None,
-							}
+					let idx = if *idx >= 0 {
+						usize::try_from(*idx).ok()
+					} else {
+						usize::try_from(e.len().await? as i64 - idx).ok()
+					};
+
+					let idx = match idx {
+						Some(idx) => idx,
+						None => {
+							out = None;
+							continue;
 						}
-					}
+					};
+
+					out = e.get(idx).await?;
 				}
 			}
 		}
@@ -127,7 +158,8 @@ impl PileValue {
 			Self::Array(x) => (!x.is_empty()).then(|| Value::Number(1u64.into())),
 			Self::ListExtractor(x) => (!x.is_empty().await?).then(|| Value::Number(1u64.into())),

-			Self::ObjectExtractor(e) => {
+			Self::ObjectExtractor(_) | Self::Item(_) => {
+				let e = self.object_extractor();
 				let keys = e.fields().await?;
 				let mut map = Map::new();
 				for k in &keys {
@@ -160,22 +192,27 @@ impl PileValue {
 			Self::Null => Value::Null,
 			Self::U64(x) => Value::Number((*x).into()),
 			Self::I64(x) => Value::Number((*x).into()),
+			Self::String(x) => Value::String(x.to_string()),

-			// TODO: replace with something meaningful
+			// TODO: replace with something meaningful?
 			Self::Blob { mime, bytes } => {
 				Value::String(format!("<Blob ({mime}, {} bytes)>", bytes.len()))
 			}
-			Self::String(x) => Value::String(x.to_string()),

-			Self::Array(x) => {
+			#[expect(clippy::expect_used)]
+			Self::Array(_) | Self::ListExtractor(_) => {
+				let e = self.list_extractor();
+				let len = e.len().await?;
 				let mut arr = Vec::new();
-				for item in &**x {
-					arr.push(Box::pin(item.to_json()).await?);
+				for i in 0..len {
+					let v = e.get(i).await?.expect("item must be present");
+					arr.push(Box::pin(v.to_json()).await?);
 				}
 				Value::Array(arr)
 			}

-			Self::ObjectExtractor(e) => {
+			Self::ObjectExtractor(_) | Self::Item(_) => {
+				let e = self.object_extractor();
 				let keys = e.fields().await?;
 				let mut map = Map::new();
 				for k in &keys {
@@ -187,8 +224,6 @@ impl PileValue {
 				}
 				Value::Object(map)
 			}
-
-			Self::ListExtractor(e) => e.to_json().await?,
 		})
 	}
 }
--- a/crates/pile/Cargo.toml
+++ b/crates/pile/Cargo.toml
@@ -10,6 +10,7 @@ workspace = true
 [dependencies]
 pile-toolbox = { workspace = true }
 pile-dataset = { workspace = true, features = ["axum", "pdfium"] }
+pile-value = { workspace = true, features = ["pdfium"] }
 pile-config = { workspace = true }

 tracing = { workspace = true }
--- a/crates/pile/src/command/annotate.rs
+++ b/crates/pile/src/command/annotate.rs
@@ -1,10 +1,12 @@
 use anyhow::{Context, Result};
 use clap::Args;
 use pile_config::{Label, Source};
-use pile_dataset::index::DbFtsIndex;
-use pile_dataset::source::DirDataSource;
-use pile_dataset::{DataSource, Datasets, Item, PileValue, extract::MetaExtractor};
+use pile_dataset::{Datasets, index::DbFtsIndex};
 use pile_toolbox::cancelabletask::{CancelFlag, CancelableTaskError};
+use pile_value::{
+	source::{DataSource, DirDataSource},
+	value::{Item, PileValue},
+};
 use std::{path::PathBuf, sync::Arc};
 use tokio_stream::StreamExt;
 use tracing::{info, warn};
@@ -72,11 +74,9 @@ impl CliCmd for AnnotateCommand {
 							continue;
 						};

-						let meta = MetaExtractor::new(&item);
-						let extractor = PileValue::ObjectExtractor(Arc::new(meta));
-
+						let item = PileValue::Item(item.clone());
 						let Some(value) =
-							index.get_field(&extractor, &field).await.with_context(|| {
+							index.get_field(&item, &field).await.with_context(|| {
 								format!("while extracting field from {}", path.display())
 							})?
 						else {
--- a/crates/pile/src/command/fields.rs
+++ b/crates/pile/src/command/fields.rs
@@ -1,9 +1,10 @@
 use anyhow::{Context, Result};
 use clap::Args;
-use pile_dataset::{Datasets, PileValue, extract::MetaExtractor};
+use pile_dataset::Datasets;
 use pile_toolbox::cancelabletask::{CancelFlag, CancelableTaskError};
+use pile_value::value::PileValue;
 use serde_json::{Map, Value};
-use std::{path::PathBuf, sync::Arc, time::Instant};
+use std::{path::PathBuf, time::Instant};
 use tokio::task::JoinSet;
 use tokio_stream::StreamExt;
 use tracing::info;
@@ -93,9 +94,8 @@ impl CliCmd for FieldsCommand {
 							item_result.with_context(|| format!("while reading source {name}"))?;
 						let name = name.clone();
 						join_set.spawn(async move {
-							let meta = MetaExtractor::new(&item);
-							let value = PileValue::ObjectExtractor(Arc::new(meta));
-							let result = value.count_fields().await.with_context(|| {
+							let item = PileValue::Item(item);
+							let result = item.count_fields().await.with_context(|| {
 								format!("while counting fields in source {name}")
 							})?;
 							Ok(result.and_then(|v| {
--- a/crates/pile/src/command/list.rs
+++ b/crates/pile/src/command/list.rs
@@ -1,8 +1,9 @@
 use anyhow::{Context, Result};
 use clap::Args;
 use pile_config::objectpath::ObjectPath;
-use pile_dataset::{Datasets, PileValue, extract::MetaExtractor};
+use pile_dataset::Datasets;
 use pile_toolbox::cancelabletask::{CancelFlag, CancelableTaskError};
+use pile_value::value::PileValue;
 use std::{path::PathBuf, str::FromStr, sync::Arc};
 use tokio::task::JoinSet;
 use tokio_stream::StreamExt;
@@ -79,9 +80,8 @@ impl CliCmd for ListCommand {
 						let invert = self.invert;

 						join_set.spawn(async move {
-							let meta = MetaExtractor::new(&item);
-							let root = PileValue::ObjectExtractor(Arc::new(meta));
-							let value = root.query(&path).await?;
+							let item = PileValue::Item(item);
+							let value = item.query(&path).await?;

 							let is_present =
 								matches!(value, Some(v) if !matches!(v, PileValue::Null));
--- a/crates/pile/src/command/probe.rs
+++ b/crates/pile/src/command/probe.rs
@@ -1,9 +1,10 @@
 use anyhow::{Context, Result};
 use clap::Args;
 use pile_config::{Label, objectpath::ObjectPath};
-use pile_dataset::{Datasets, PileValue, extract::MetaExtractor};
+use pile_dataset::Datasets;
 use pile_toolbox::cancelabletask::{CancelFlag, CancelableTaskError};
-use std::{path::PathBuf, sync::Arc};
+use pile_value::value::PileValue;
+use std::path::PathBuf;

 use crate::{CliCmd, GlobalContext};

@@ -54,9 +55,8 @@ impl CliCmd for ProbeCommand {
 				anyhow::anyhow!("{:?} not found in source {:?}", self.key, self.source)
 			})?;

-			let value = PileValue::ObjectExtractor(Arc::new(MetaExtractor::new(&item)));
-			value
-				.to_json()
+			let item = PileValue::Item(item);
+			item.to_json()
 				.await
 				.with_context(|| format!("while extracting {}", self.key))?
 		};
--- a/default.nix
+++ b/default.nix
@@ -0,0 +1,52 @@
+let
+  rustOverlay = import (
+    builtins.fetchTarball {
+      url = "https://github.com/oxalica/rust-overlay/archive/master.tar.gz";
+      sha256 = "0qgrkgc695a7gja83dngxrcx4gdg9056gvg5325i5yyjxg0ni6c9";
+    }
+  );
+  pkgsDefault = import <nixpkgs> { overlays = [ rustOverlay ]; };
+  rustToolchain = pkgsDefault.rust-bin.fromRustupToolchainFile ./rust-toolchain.toml;
+  rustPlatformDefault = pkgsDefault.makeRustPlatform {
+    cargo = rustToolchain;
+    rustc = rustToolchain;
+  };
+in
+{
+  pkgs ? pkgsDefault,
+  pileRustPlatform ? rustPlatformDefault,
+}:
+
+pileRustPlatform.buildRustPackage {
+  pname = "pile";
+  version = "0.0.2";
+  src = ./.;
+
+  cargoLock.lockFile = ./Cargo.lock;
+
+  PDFIUM_LIB_DIR = "${pkgs.pdfium-binaries}/lib";
+
+  buildInputs = [
+    pkgs.pdfium-binaries
+    pkgs.openssl
+  ]
+  ++ pkgs.lib.optionals pkgs.stdenv.isDarwin [
+    pkgs.darwin.apple_sdk.frameworks.Security
+    pkgs.darwin.apple_sdk.frameworks.SystemConfiguration
+  ];
+
+  nativeBuildInputs = [
+    pkgs.pkg-config
+    pkgs.makeWrapper
+  ];
+
+  postInstall = ''
+    wrapProgram $out/bin/pile \
+      --prefix LD_LIBRARY_PATH : ${pkgs.lib.makeLibraryPath [ pkgs.pdfium-binaries ]}
+  '';
+
+  meta = {
+    description = "pile - flexible file indexing";
+    mainProgram = "pile";
+  };
+}
--- a/flake.nix
+++ b/flake.nix
@@ -0,0 +1,40 @@
+{
+  description = "pile - personal data indexer";
+
+  inputs = {
+    nixpkgs.url = "github:NixOS/nixpkgs/nixos-unstable";
+    rust-overlay = {
+      url = "github:oxalica/rust-overlay";
+      inputs.nixpkgs.follows = "nixpkgs";
+    };
+    flake-utils.url = "github:numtide/flake-utils";
+  };
+
+  outputs = { self, nixpkgs, rust-overlay, flake-utils }:
+    flake-utils.lib.eachDefaultSystem (system:
+      let
+        overlays = [ (import rust-overlay) ];
+        pkgs = import nixpkgs { inherit system overlays; };
+
+        rustToolchain = pkgs.rust-bin.fromRustupToolchainFile ./rust-toolchain.toml;
+
+        rustPlatform = pkgs.makeRustPlatform {
+          cargo = rustToolchain;
+          rustc = rustToolchain;
+        };
+
+      in {
+        packages.default = import ./default.nix { inherit pkgs; pileRustPlatform = rustPlatform; };
+
+        devShells.default = pkgs.mkShell {
+          buildInputs = [
+            rustToolchain
+            pkgs.pdfium-binaries
+            pkgs.openssl
+            pkgs.pkg-config
+          ];
+
+          LD_LIBRARY_PATH = pkgs.lib.makeLibraryPath [ pkgs.pdfium-binaries ];
+        };
+      });
+}
--- a/rust-toolchain.toml
+++ b/rust-toolchain.toml
@@ -0,0 +1,2 @@
+[toolchain]
+channel = "1.94.0"
Author	SHA1	Message	Date
rm-dr	82dcdbaa27	Bump versions Some checks failed CI / Build and test (all features) (push) Waiting to run Details CI / Typos (push) Successful in 41s Details CI / Clippy (push) Failing after 1m28s Details CI / Build and test (push) Has been cancelled Details	2026-03-11 12:22:02 -07:00
rm-dr	42f186d77f	More string extractors	2026-03-11 12:22:02 -07:00
rm-dr	b36b62150c	Add nix files	2026-03-11 12:21:58 -07:00
rm-dr	c03fac0e37	`/item` range requests	2026-03-11 11:05:32 -07:00
rm-dr	c9d99e8719	Stream items in `/item`	2026-03-11 11:05:32 -07:00
rm-dr	4546a85bd3	Extractor rewrite	2026-03-11 11:05:30 -07:00