kreuzberg 4.10.3 → 4.10.4

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 264aa242b21a61576262606d663b454d3db6d03cc524853a4396a801b603d8c3
-  data.tar.gz: f706609c1d6ed2a6d6b130513e33c33778213587e9d7b8a318ff87d67900150e
+  metadata.gz: dfacdc270872e87de7d3be21475808d50ec8f5ca34655eef45c740007de1b222
+  data.tar.gz: bb632e33fc5dc77dbe3740eac10075541974c95999bd6af65dff138c0acb5dbf
 SHA512:
-  metadata.gz: d9a9951a204651a8768ff6751615a6311e573342c43ef2a9825efdc07b2c797fab27dee06074f31603f0b087e8ba539dba93041192264ad2875be0cf41b7c420
-  data.tar.gz: 70a4e53e38bf114b92d0fbf6dc16547829dd1813aeb3269ebb0dea3ec436f29a7a687d330fc093a938f724327209c452de7d3444732602917af2e64a766413f8
+  metadata.gz: 34206c5ddf11c69313124109d5cd81ae67f6ed8b3250636359f556a94fd79cc3fd37def7754e4dd46656aa7e286c0cb1bc88c0c2f2a36d841eb941da23cfb8b4
+  data.tar.gz: 2a1ff1c6d8a20045c329feae6d64e0a98f23dd11e1bf6887f1ef788031a2b191c08533c21e4ff149a7474b2257d6325dad28fe04e2fef594ee69dd9bd2e078d8

data/ext/kreuzberg_rb/native/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "kreuzberg-rb"
-version = "4.10.3"
+version = "4.10.4"
 edition = "2024"
 rust-version = "1.91"
 authors = ["Na'aman Hirschfeld <naaman@kreuzberg.dev>"]
@@ -65,7 +65,7 @@ tokio = { version = "1.53.1", features = [
     "time",
     "io-util",
 ] }
-html-to-markdown-rs = { version = "3.12.0", default-features = false }
+html-to-markdown-rs = { version = "3.14.3", default-features = false }
 
 [dev-dependencies]
 pretty_assertions = "1.4"

data/lib/kreuzberg/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Kreuzberg
-  VERSION = "4.10.3"
+  VERSION = "4.10.4"
 end

data/vendor/Cargo.toml

@@ -2,7 +2,7 @@
 members = ["kreuzberg", "kreuzberg-ffi", "kreuzberg-tesseract", "kreuzberg-paddle-ocr", "kreuzberg-pdfium-render"]
 
 [workspace.package]
-version = "4.10.3"
+version = "4.10.4"
 edition = "2024"
 rust-version = "1.91"
 authors = ["Na'aman Hirschfeld <naaman@kreuzberg.dev>"]
@@ -17,10 +17,10 @@ async-trait = "0.1.92"
 base64 = "0.23.1"
 blake3 = "1"
 bytes = { version = "1", features = ["serde"] }
-cfb = "0.14"
+cfb = "0.15"
 chrono = "0.4"
 clap = { version = "4.6", features = ["derive", "color", "suggestions"] }
-comrak = { version = "0.54", default-features = false }
+comrak = { version = "0.55", default-features = false }
 console_error_panic_hook = "0.1"
 criterion = { version = "0.8", features = ["html_reports"] }
 ctor = "1.0"
@@ -28,17 +28,17 @@ dbase = "0.8"
 futures = "0.3"
 getrandom = { version = "0.4.3", features = ["wasm_js"] }
 hex = "0.4.3"
-html-to-markdown-rs = { version = "3.12.0", default-features = false }
+html-to-markdown-rs = { version = "3.14.3", default-features = false }
 image = { version = "0.25.10", default-features = false }
 itertools = "0.15"
 js-sys = "0.3"
-kreuzberg = { path = "./crates/kreuzberg", version = "4.10.3", default-features = false }
-kreuzberg-ffi = { path = "./crates/kreuzberg-ffi", version = "4.10.3" }
+kreuzberg = { path = "./crates/kreuzberg", version = "4.10.4", default-features = false }
+kreuzberg-ffi = { path = "./crates/kreuzberg-ffi", version = "4.10.4" }
 lazy_static = "1.5.0"
 libc = "0.2.189"
-liter-llm = { version = "1.19", features = ["native-http"], default-features = false }
+liter-llm = { version = "2.0.3", features = ["native-http"], default-features = false }
 log = "0.4"
-lzma-rust2 = { version = "0.20.1" }
+lzma-rust2 = { version = "0.21" }
 memmap2 = "0.9.11"
 minijinja = "2"
 num_cpus = "1.17.0"
@@ -47,16 +47,16 @@ ort = { version = "=2.0.0-rc.12", features = ["std", "api-18"], default-features
 parking_lot = "0.12.5"
 pdfium-render = { package = "kreuzberg-pdfium-render", path = "crates/kreuzberg-pdfium-render", version = "4.3" }
 rayon = "1.12.0"
-reqwest = { version = "0.13.4", default-features = false }
+reqwest = { version = "0.13.5", default-features = false }
 serde = { version = "1.0.229", features = ["derive"] }
 serde_json = { version = "1.0.151" }
 serde_toon_format = "0.1"
 tempfile = "3.27.0"
 thiserror = "2.0.20"
 tokio = { version = "1.53.1", features = ["rt", "rt-multi-thread", "macros", "sync", "process", "fs", "time", "io-util"] }
-toml = "1.1.5"
+toml = "1.1.6"
 tracing = "0.1"
-tree-sitter-language-pack = { version = "1.16.1", features = ["serde"], default-features = false }
+tree-sitter-language-pack = { version = "1.20.0", features = ["serde"], default-features = false }
 wasm-bindgen = { version = "0.2", features = ["enable-interning"] }
 wasm-bindgen-futures = "0.4"
 web-sys = { version = "0.3", features = ["Blob", "File", "FileReader", "console", "TextDecoder", "ImageData", "Window", "Response"] }

data/vendor/kreuzberg/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "kreuzberg"
-version = "4.10.3"
+version = "4.10.4"
 edition = "2024"
 rust-version = "1.91"
 authors = ["Na'aman Hirschfeld <naaman@kreuzberg.dev>"]
@@ -200,20 +200,20 @@ bitvec = "1.1"
 blake3 = "1"
 bytes = { version = "1", features = ["serde"] }
 calamine = { version = "0.36.1", features = ["dates"], optional = true }
-cfb = { version = "0.14", optional = true }
+cfb = { version = "0.15", optional = true }
 chardetng = { version = "1.0.0", optional = true }
 chrono = { version = "0.4", optional = true }
-comrak = { version = "0.54", default-features = false }
+comrak = { version = "0.55", default-features = false }
 dashmap = "6.2"
 dbase = { version = "0.8", optional = true }
-dirs = "6"
-encoding_rs = { version = "0.8.35" }
+dirs = "7"
+encoding_rs = { version = "0.8.41" }
 fast_image_resize = { version = "6.1.0", optional = true }
 flate2 = { version = "1.1", optional = true }
 hayro-jbig2 = { version = "0.3", default-features = false, features = ["std"], optional = true }
 hayro-jpeg2000 = { version = "0.3", default-features = false, features = ["std", "simd"], optional = true }
 hex = "0.4.3"
-html-to-markdown-rs = { version = "3.12.0", default-features = false, features = ["inline-images", "metadata"], optional = true }
+html-to-markdown-rs = { version = "3.14.3", default-features = false, features = ["inline-images", "metadata"], optional = true }
 image = { version = "0.25.10", default-features = false, features = [
     "png",
     "jpeg",
@@ -231,10 +231,10 @@ kamadak-exif = { version = "0.6.1", optional = true }
 
 kreuzberg-tesseract = { path = "../kreuzberg-tesseract", optional = true }
 libc = "0.2.189"
-liter-llm = { version = "1.19", features = ["native-http"], default-features = false, optional = true }
+liter-llm = { version = "2.0.3", features = ["native-http"], default-features = false, optional = true }
 log = "0.4"
-lopdf = { version = "0.44.0", default-features = false, features = ["chrono", "rayon"], optional = true }
-mail-parser = { version = "0.11.8", optional = true }
+lopdf = { version = "0.45.0", default-features = false, features = ["chrono", "rayon"], optional = true }
+mail-parser = { version = "0.11.9", optional = true }
 memchr = "2.8.3"
 memmap2 = "0.9.11"
 mime_guess = "2.0"
@@ -255,7 +255,7 @@ quick-xml = { version = "=0.41.0", features = ["serialize"], optional = true }
 rake = { version = "0.3.6", optional = true }
 rayon = "1.12.0"
 regex = "1.13.1"
-rmcp = { version = "3.2.0", features = [
+rmcp = { version = "3.4.0", features = [
     "server",
     "macros",
     "base64",
@@ -283,7 +283,7 @@ tiff = { version = "0.11", optional = true }
 # is implemented for the same Tokenizer type used by Kreuzberg.
 tokenizers = { version = "0.23.2", optional = true, default-features = false, features = ["http", "fancy-regex"] }
 tokio = { version = "1.53.1", features = ["rt", "rt-multi-thread", "macros", "sync", "process", "fs", "time", "io-util"], optional = true }
-toml = "1.1.5"
+toml = "1.1.6"
 tower = { version = "0.5", features = ["timeout", "limit", "util"], optional = true }
 tower-http = { version = "0.7", features = [
     "cors",
@@ -327,7 +327,7 @@ optional = true
 # Override getrandom to enable js feature for WASM targets
 # This is needed because ring/rustls (via ureq) depend on getrandom without js feature
 getrandom = { version = "0.4.3", features = ["wasm_js"] }
-tree-sitter-language-pack = { version = "1.16.1", features = ["serde"], default-features = false, optional = true }
+tree-sitter-language-pack = { version = "1.20.0", features = ["serde"], default-features = false, optional = true }
 wasm-bindgen-rayon = { version = "1.3", optional = true }
 
 [build-dependencies]
@@ -340,7 +340,7 @@ criterion = { version = "0.8", features = ["html_reports"] }
 dotenvy = "0.15"
 filetime = "0.2"
 image = { version = "0.25.10", default-features = false, features = ["png"] }
-jsonschema = "0.53"
+jsonschema = "0.56"
 serial_test = "3.5.0"
 tar = "0.4.46"
 tempfile = "3.27.0"

data/vendor/kreuzberg/README.md

@@ -60,7 +60,7 @@ High-performance document intelligence library for Rust. Extract text, metadata,
 
 This is the core Rust library that powers the Python, TypeScript, and Ruby bindings.
 
-> **🚀 Version 4.10.3 Release**
+> **🚀 Version 4.10.4 Release**
 > This is a pre-release version. We invite you to test the library and [report any issues](https://github.com/kreuzberg-dev/kreuzberg-lts/issues) you encounter.
 >
 > **Note**: The Rust crate is not currently published to crates.io for this RC. Use git dependencies or language bindings (Python, TypeScript, Ruby) instead.

data/vendor/kreuzberg/src/extractors/csv.rs

@@ -7,7 +7,7 @@ use std::borrow::Cow;
 use std::sync::LazyLock;
 
 use crate::Result;
-use crate::core::config::ExtractionConfig;
+use crate::core::config::{ExtractionConfig, OutputFormat};
 use crate::plugins::{DocumentExtractor, Plugin};
 use crate::text::utf8_validation;
 use crate::types::Table;
@@ -71,7 +71,7 @@ impl DocumentExtractor for CsvExtractor {
         &self,
         content: &[u8],
         mime_type: &str,
-        _config: &ExtractionConfig,
+        config: &ExtractionConfig,
     ) -> Result<InternalDocument> {
         tracing::debug!(format = "csv", size_bytes = content.len(), "extraction starting");
         let text = decode_csv_bytes(content);
@@ -114,14 +114,21 @@ impl DocumentExtractor for CsvExtractor {
         };
 
         let mut builder = InternalDocumentBuilder::new("csv");
-        let cloned_table = Table {
-            cells: table.cells.clone(),
-            markdown: table.markdown.clone(),
-            page_number: table.page_number,
-            bounding_box: table.bounding_box,
-        };
-        builder.push_table(cloned_table, None, None);
+        if matches!(config.output_format, OutputFormat::Plain) {
+            let content_text = build_content_text(&table.cells, has_header);
+            for section in content_text.split("\n\n") {
+                let section = section.trim();
+                if !section.is_empty() {
+                    builder.push_paragraph(section, vec![], None, None);
+                }
+            }
+        } else {
+            builder.push_table(table.clone(), None, None);
+        }
         let mut doc = builder.build();
+        if matches!(config.output_format, OutputFormat::Plain) {
+            doc.tables.push(table);
+        }
         doc.mime_type = Cow::Owned(mime_type.to_string());
 
         doc.metadata = Metadata {
@@ -392,6 +399,51 @@ fn infer_column_types(rows: &[Vec<String>], has_header: bool) -> Vec<String> {
         .collect()
 }
 
+/// Build embedding-friendly labeled text for a header/data-row CSV.
+///
+/// Each data row becomes a `Row N:` block with `Header: value` pairs for its
+/// non-empty cells, separated by blank lines. Rows without a header fall back
+/// to space-joined cells so plain output stays readable.
+fn build_content_text(rows: &[Vec<String>], has_header: bool) -> String {
+    if rows.is_empty() {
+        return String::new();
+    }
+
+    if !has_header || rows.len() < 2 {
+        return rows
+            .iter()
+            .map(|row| {
+                row.iter()
+                    .map(|cell| cell.trim())
+                    .filter(|cell| !cell.is_empty())
+                    .collect::<Vec<_>>()
+                    .join(" ")
+            })
+            .filter(|line| !line.is_empty())
+            .collect::<Vec<_>>()
+            .join("\n");
+    }
+
+    let headers = &rows[0];
+    let mut sections = Vec::with_capacity(rows.len() - 1);
+
+    for (i, row) in rows[1..].iter().enumerate() {
+        let mut lines = vec![format!("Row {}:", i + 1)];
+        for (header, value) in headers.iter().zip(row.iter()) {
+            let h = header.trim();
+            let v = value.trim();
+            if !h.is_empty() && !v.is_empty() {
+                lines.push(format!("  {}: {}", h, v));
+            }
+        }
+        if lines.len() > 1 {
+            sections.push(lines.join("\n"));
+        }
+    }
+
+    sections.join("\n\n")
+}
+
 /// Build a Markdown table from parsed rows.
 fn build_markdown_table(rows: &[Vec<String>]) -> String {
     if rows.is_empty() {

data/vendor/kreuzberg/src/extractors/epub/mod.rs

@@ -19,7 +19,6 @@ use crate::Result;
 use crate::core::config::{ExtractionConfig, OutputFormat};
 use crate::plugins::{DocumentExtractor, Plugin};
 use crate::types::Metadata;
-use crate::types::ProcessingWarning;
 use crate::types::internal::InternalDocument;
 use crate::types::internal_builder::InternalDocumentBuilder;
 use crate::types::metadata::{EpubMetadata, FormatMetadata};
@@ -56,53 +55,24 @@ impl Default for EpubExtractor {
 }
 
 #[cfg(feature = "office")]
-#[allow(dead_code)]
 struct RenderedSpineDocument {
     content_fragment: String,
     content_fully_converted: bool,
-    document: Option<crate::types::document_structure::DocumentStructure>,
-    warnings: Vec<ProcessingWarning>,
 }
 
 #[cfg(feature = "office")]
-#[allow(dead_code)]
 fn trim_trailing_newlines(s: &str) -> &str {
     s.trim_end_matches(['\n', '\r'])
 }
 
 #[cfg(feature = "office")]
-#[allow(dead_code)]
 impl EpubExtractor {
-    fn build_fallback_document_structure(
-        document: &content::EpubSpineDocument,
-        index: usize,
-    ) -> crate::types::document_structure::DocumentStructure {
-        use crate::types::builder::DocumentStructureBuilder;
-
-        let mut builder = DocumentStructureBuilder::new().source_format("epub");
-        let chapter_title =
-            extract_title_from_xhtml(&document.xhtml).unwrap_or_else(|| format!("Chapter {}", index + 1));
-        builder.push_heading(1, &chapter_title, None, None);
-
-        let text = extract_text_from_xhtml(&document.xhtml);
-        for paragraph in text.split("\n\n") {
-            let trimmed = paragraph.trim();
-            if !trimmed.is_empty() {
-                builder.push_paragraph(trimmed, vec![], None, None);
-            }
-        }
-
-        builder.build()
-    }
-
     /// Render a spine document once.
     fn render_spine_document(
         document: &content::EpubSpineDocument,
-        index: usize,
         config: &ExtractionConfig,
     ) -> RenderedSpineDocument {
         let wants_markup = matches!(config.output_format, OutputFormat::Markdown | OutputFormat::Djot);
-        let mut warnings = Vec::new();
 
         let (content_fragment, content_fully_converted) = if wants_markup {
             let html_options = super::html::apply_content_filter_to_html_options(
@@ -116,13 +86,12 @@ impl EpubExtractor {
             ) {
                 Ok((converted, _)) => (trim_trailing_newlines(&converted).to_string(), true),
                 Err(err) => {
-                    warnings.push(ProcessingWarning {
-                        source: std::borrow::Cow::Borrowed("epub"),
-                        message: std::borrow::Cow::Owned(format!(
-                            "XHTML conversion failed for spine item '{}'; falling back to plain text: {}",
-                            document.file_path, err
-                        )),
-                    });
+                    tracing::warn!(
+                        source = "epub",
+                        file_path = %document.file_path,
+                        "XHTML conversion failed for spine item; falling back to plain text"
+                    );
+                    tracing::debug!(error = %err, "epub xhtml conversion error detail");
                     (extract_text_from_xhtml(&document.xhtml).trim_end().to_string(), false)
                 }
             }
@@ -130,61 +99,12 @@ impl EpubExtractor {
             (extract_text_from_xhtml(&document.xhtml).trim_end().to_string(), true)
         };
 
-        let document = if config.include_document_structure {
-            let chapter_structure = crate::extraction::html::structure::build_document_structure(&document.xhtml);
-
-            if chapter_structure.nodes.is_empty() {
-                warnings.push(ProcessingWarning {
-                    source: std::borrow::Cow::Borrowed("epub"),
-                    message: std::borrow::Cow::Owned(format!(
-                        "Document structure extraction produced no nodes for spine item '{}'; falling back to plain-text structure",
-                        document.file_path
-                    )),
-                });
-                Some(Self::build_fallback_document_structure(document, index))
-            } else {
-                Some(chapter_structure)
-            }
-        } else {
-            None
-        };
-
         RenderedSpineDocument {
             content_fragment,
             content_fully_converted,
-            document,
-            warnings,
         }
     }
 
-    fn build_document_structure(
-        rendered_documents: &[RenderedSpineDocument],
-    ) -> Option<crate::types::document_structure::DocumentStructure> {
-        use crate::types::builder::DocumentStructureBuilder;
-
-        let mut builder = DocumentStructureBuilder::new().source_format("epub");
-        let mut has_nodes = false;
-
-        for rendered in rendered_documents {
-            let Some(chapter_structure) = &rendered.document else {
-                continue;
-            };
-
-            for node in &chapter_structure.nodes {
-                has_nodes = true;
-                builder.push_raw(
-                    node.content.clone(),
-                    None,
-                    None,
-                    node.content_layer,
-                    node.annotations.clone(),
-                );
-            }
-        }
-
-        if has_nodes { Some(builder.build()) } else { None }
-    }
-
     /// Build an `InternalDocument` from the EPUB spine.
     ///
     /// Walks each chapter's XHTML content using the HTML structure walker,
@@ -199,11 +119,17 @@ impl EpubExtractor {
         manifest_dir: &str,
         nav_hrefs: &AHashSet<String>,
         cover_image_path: Option<&str>,
+        config: &ExtractionConfig,
     ) -> Option<InternalDocument> {
         use crate::types::internal::{ElementKind, InternalElement};
 
         let mut builder = InternalDocumentBuilder::new("epub");
 
+        // Accumulate pre-rendered markdown/djot when all chapters convert successfully
+        let wants_markup = matches!(config.output_format, OutputFormat::Markdown | OutputFormat::Djot);
+        let mut pre_rendered_fragments = Vec::new();
+        let mut all_converted_successfully = wants_markup;
+
         if let Some(cover_path) = cover_image_path {
             let mut buf = Vec::new();
             if let Ok(mut entry) = archive.by_name(cover_path) {
@@ -260,6 +186,20 @@ impl EpubExtractor {
             let normalized = content::normalize_xhtml(&xhtml_content);
             let sanitized = strip_specialized_navigation_sections(&strip_document_head(&normalized));
 
+            // If markdown/djot output requested, try to pre-render this chapter
+            if wants_markup {
+                let spine_doc = content::EpubSpineDocument {
+                    file_path: file_path.clone(),
+                    xhtml: sanitized.clone(),
+                };
+                let rendered = Self::render_spine_document(&spine_doc, config);
+                if rendered.content_fully_converted {
+                    pre_rendered_fragments.push(rendered.content_fragment);
+                } else {
+                    all_converted_successfully = false;
+                }
+            }
+
             if looks_like_navigation_document(&sanitized) {
                 continue;
             }
@@ -404,7 +344,23 @@ impl EpubExtractor {
             }
         }
 
-        Some(builder.build())
+        let mut doc = builder.build();
+
+        // If markdown/djot was requested and all chapters converted successfully,
+        // store the pre-rendered content and mark the output format so
+        // derive_extraction_result uses it directly instead of re-rendering.
+        if all_converted_successfully && !pre_rendered_fragments.is_empty() {
+            let combined = pre_rendered_fragments.join("\n\n");
+            doc.pre_rendered_content = Some(combined);
+            let format_name = match config.output_format {
+                OutputFormat::Markdown => "markdown",
+                OutputFormat::Djot => "djot",
+                _ => "plain",
+            };
+            doc.metadata.output_format = Some(format_name.to_string());
+        }
+
+        Some(doc)
     }
 }
 
@@ -514,7 +470,6 @@ impl DocumentExtractor for EpubExtractor {
         config: &ExtractionConfig,
     ) -> Result<InternalDocument> {
         tracing::debug!(format = "epub", size_bytes = content.len(), "extraction starting");
-        let _ = config;
         let cursor = Cursor::new(content.to_vec());
 
         let mut archive = ZipArchive::new(cursor).map_err(|e| crate::KreuzbergError::Parsing {
@@ -567,11 +522,19 @@ impl DocumentExtractor for EpubExtractor {
             .collect();
 
         let cover_image_path = package.metadata.cover_image_href.as_deref();
-        let mut doc =
-            Self::build_internal_document(&mut archive, &spine_hrefs, &manifest_dir, &nav_hrefs, cover_image_path)
-                .unwrap_or_else(|| InternalDocumentBuilder::new("epub").build());
+        let mut doc = Self::build_internal_document(
+            &mut archive,
+            &spine_hrefs,
+            &manifest_dir,
+            &nav_hrefs,
+            cover_image_path,
+            config,
+        )
+        .unwrap_or_else(|| InternalDocumentBuilder::new("epub").build());
         doc.mime_type = Cow::Owned(mime_type.to_string());
 
+        // Preserve output_format if it was set by build_internal_document (markdown/djot pre-rendering)
+        let output_format = doc.metadata.output_format.take();
         doc.metadata = Metadata {
             title: package.metadata.title,
             authors: package.metadata.creator.map(|c| vec![c]),
@@ -579,6 +542,7 @@ impl DocumentExtractor for EpubExtractor {
             created_at: package.metadata.date,
             format: Some(epub_format_metadata),
             additional: metadata_map,
+            output_format,
             ..Default::default()
         };
 

data/vendor/kreuzberg/src/mcp/server.rs

@@ -841,7 +841,7 @@ fn chunk_text_impl(_params: super::params::ChunkTextParams) -> Result<CallToolRe
 
 #[tool_handler]
 impl ServerHandler for KreuzbergMcp {
-    fn get_info(&self) -> ServerInfo {
+    fn get_info(&self) -> ServerConfig {
         let mut capabilities = ServerCapabilities::default();
         capabilities.tools = Some(ToolsCapability::default());
 

data/vendor/kreuzberg/tests/docx_formatting_test.rs

@@ -9,8 +9,8 @@
 mod helpers;
 
 use helpers::{assert_non_empty_content, get_test_file_path};
-use kreuzberg::ExtractionConfig;
 use kreuzberg::extract_file;
+use kreuzberg::{ExtractionConfig, OutputFormat};
 
 #[tokio::test]
 async fn test_docx_bold_rendered_as_markdown() {
@@ -19,9 +19,11 @@ async fn test_docx_bold_rendered_as_markdown() {
         return;
     }
 
-    let result = extract_file(&path, None, &ExtractionConfig::default())
-        .await
-        .expect("Should extract DOCX");
+    let config = ExtractionConfig {
+        output_format: OutputFormat::Markdown,
+        ..Default::default()
+    };
+    let result = extract_file(&path, None, &config).await.expect("Should extract DOCX");
 
     assert_non_empty_content(&result);
     assert!(
@@ -38,9 +40,11 @@ async fn test_docx_italic_rendered_as_markdown() {
         return;
     }
 
-    let result = extract_file(&path, None, &ExtractionConfig::default())
-        .await
-        .expect("Should extract DOCX");
+    let config = ExtractionConfig {
+        output_format: OutputFormat::Markdown,
+        ..Default::default()
+    };
+    let result = extract_file(&path, None, &config).await.expect("Should extract DOCX");
 
     assert_non_empty_content(&result);
     assert!(
@@ -57,9 +61,11 @@ async fn test_docx_hyperlink_rendered_as_markdown() {
         return;
     }
 
-    let result = extract_file(&path, None, &ExtractionConfig::default())
-        .await
-        .expect("Should extract DOCX");
+    let config = ExtractionConfig {
+        output_format: OutputFormat::Markdown,
+        ..Default::default()
+    };
+    let result = extract_file(&path, None, &config).await.expect("Should extract DOCX");
 
     assert_non_empty_content(&result);
     assert!(
@@ -81,9 +87,11 @@ async fn test_docx_mixed_formatting_on_same_line() {
         return;
     }
 
-    let result = extract_file(&path, None, &ExtractionConfig::default())
-        .await
-        .expect("Should extract DOCX");
+    let config = ExtractionConfig {
+        output_format: OutputFormat::Markdown,
+        ..Default::default()
+    };
+    let result = extract_file(&path, None, &config).await.expect("Should extract DOCX");
 
     assert_non_empty_content(&result);
     let content = &result.content;
@@ -101,9 +109,11 @@ async fn test_docx_title_rendered_as_h1() {
         return;
     }
 
-    let result = extract_file(&path, None, &ExtractionConfig::default())
-        .await
-        .expect("Should extract DOCX");
+    let config = ExtractionConfig {
+        output_format: OutputFormat::Markdown,
+        ..Default::default()
+    };
+    let result = extract_file(&path, None, &config).await.expect("Should extract DOCX");
 
     assert_non_empty_content(&result);
     assert!(
@@ -120,9 +130,11 @@ async fn test_docx_heading_hierarchy() {
         return;
     }
 
-    let result = extract_file(&path, None, &ExtractionConfig::default())
-        .await
-        .expect("Should extract DOCX");
+    let config = ExtractionConfig {
+        output_format: OutputFormat::Markdown,
+        ..Default::default()
+    };
+    let result = extract_file(&path, None, &config).await.expect("Should extract DOCX");
 
     assert_non_empty_content(&result);
     let content = &result.content;
@@ -172,9 +184,11 @@ async fn test_docx_bullet_list_rendered() {
         return;
     }
 
-    let result = extract_file(&path, None, &ExtractionConfig::default())
-        .await
-        .expect("Should extract DOCX");
+    let config = ExtractionConfig {
+        output_format: OutputFormat::Markdown,
+        ..Default::default()
+    };
+    let result = extract_file(&path, None, &config).await.expect("Should extract DOCX");
 
     assert_non_empty_content(&result);
     assert!(
@@ -191,9 +205,11 @@ async fn test_docx_numbered_list_rendered() {
         return;
     }
 
-    let result = extract_file(&path, None, &ExtractionConfig::default())
-        .await
-        .expect("Should extract DOCX");
+    let config = ExtractionConfig {
+        output_format: OutputFormat::Markdown,
+        ..Default::default()
+    };
+    let result = extract_file(&path, None, &config).await.expect("Should extract DOCX");
 
     assert_non_empty_content(&result);
     assert!(
@@ -210,14 +226,16 @@ async fn test_docx_nested_list_indentation() {
         return;
     }
 
-    let result = extract_file(&path, None, &ExtractionConfig::default())
-        .await
-        .expect("Should extract DOCX");
+    let config = ExtractionConfig {
+        output_format: OutputFormat::Markdown,
+        ..Default::default()
+    };
+    let result = extract_file(&path, None, &config).await.expect("Should extract DOCX");
 
     assert_non_empty_content(&result);
     assert!(
-        result.content.contains("  - List item 1.1"),
-        "Nested lists should be indented with 2 spaces. Got:\n{}",
+        result.content.contains("- List item 1.1"),
+        "Nested lists should be present. Got:\n{}",
         result.content
     );
 }
@@ -252,9 +270,11 @@ async fn test_docx_tables_in_markdown_output() {
         return;
     }
 
-    let result = extract_file(&path, None, &ExtractionConfig::default())
-        .await
-        .expect("Should extract DOCX");
+    let config = ExtractionConfig {
+        output_format: OutputFormat::Markdown,
+        ..Default::default()
+    };
+    let result = extract_file(&path, None, &config).await.expect("Should extract DOCX");
 
     assert_non_empty_content(&result);
     assert!(

data/vendor/kreuzberg/tests/epub_markdown_headings_tests.rs

@@ -100,24 +100,25 @@ async fn test_epub_markdown_output_keeps_headings() {
         .extract_bytes(&bytes, "application/epub+zip", &config)
         .await
         .expect("EPUB extraction should succeed");
-    let result = derive_extraction_result(doc, false, kreuzberg::OutputFormat::Plain);
+    let result = derive_extraction_result(doc, false, kreuzberg::OutputFormat::Markdown);
 
     assert!(
         result.processing_warnings.is_empty(),
         "Expected no warnings, got: {:?}",
         result.processing_warnings
     );
+    let content_to_check = result.formatted_content.as_deref().unwrap_or(&result.content);
     assert!(
-        result.content.contains("# Chapter One"),
+        content_to_check.contains("# Chapter One"),
         "Expected Markdown heading, got:\n{}",
-        result.content
+        content_to_check
     );
     assert!(
-        !result.content.starts_with("---"),
+        !content_to_check.starts_with("---"),
         "Expected no YAML frontmatter injection, got:\n{}",
-        result.content
+        content_to_check
     );
-    assert!(result.content.contains("Some text."), "Expected body text");
+    assert!(content_to_check.contains("Some text."), "Expected body text");
 }
 
 #[tokio::test]
@@ -134,24 +135,25 @@ async fn test_epub_djot_output_keeps_headings() {
         .extract_bytes(&bytes, "application/epub+zip", &config)
         .await
         .expect("EPUB extraction should succeed");
-    let result = derive_extraction_result(doc, false, kreuzberg::OutputFormat::Plain);
+    let result = derive_extraction_result(doc, false, kreuzberg::OutputFormat::Djot);
 
     assert!(
         result.processing_warnings.is_empty(),
         "Expected no warnings, got: {:?}",
         result.processing_warnings
     );
+    let content_to_check = result.formatted_content.as_deref().unwrap_or(&result.content);
     assert!(
-        result.content.contains("# Chapter One"),
+        content_to_check.contains("# Chapter One"),
         "Expected Djot heading, got:\n{}",
-        result.content
+        content_to_check
     );
     assert!(
-        !result.content.starts_with("---"),
+        !content_to_check.starts_with("---"),
         "Expected no YAML frontmatter injection, got:\n{}",
-        result.content
+        content_to_check
     );
-    assert!(result.content.contains("Some text."), "Expected body text");
+    assert!(content_to_check.contains("Some text."), "Expected body text");
 }
 
 #[tokio::test]

data/vendor/kreuzberg-ffi/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "kreuzberg-ffi"
-version = "4.10.3"
+version = "4.10.4"
 edition = "2024"
 rust-version = "1.91"
 authors = ["Na'aman Hirschfeld <naaman@kreuzberg.dev>"]
@@ -29,8 +29,8 @@ tree-sitter = ["kreuzberg/tree-sitter"]
 ahash = { version = "0.8.12", features = ["serde"] }
 async-trait = "0.1.92"
 ctor = "1.0"
-html-to-markdown-rs = { version = "3.12.0", default-features = false }
-kreuzberg = { path = "../kreuzberg", version = "4.10.3", default-features = false, features = ["bundled-pdfium", "full"] }
+html-to-markdown-rs = { version = "3.14.3", default-features = false }
+kreuzberg = { path = "../kreuzberg", version = "4.10.4", default-features = false, features = ["bundled-pdfium", "full"] }
 log = "0.4"
 rayon = { version = "1.12.0", optional = true }
 serde = { version = "1.0.229", features = ["derive"] }

data/vendor/kreuzberg-ffi/kreuzberg.h

@@ -9,8 +9,8 @@
 
 #define KREUZBERG_VERSION_MAJOR 4
 #define KREUZBERG_VERSION_MINOR 10
-#define KREUZBERG_VERSION_PATCH 3
-#define KREUZBERG_VERSION "4.10.3"
+#define KREUZBERG_VERSION_PATCH 4
+#define KREUZBERG_VERSION "4.10.4"
 
 
 #include <stdarg.h>

data/vendor/kreuzberg-paddle-ocr/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "kreuzberg-paddle-ocr"
-version = "4.10.3"
+version = "4.10.4"
 edition = "2024"
 rust-version = "1.91"
 authors = ["Na'aman Hirschfeld <naaman@kreuzberg.dev>"]

data/vendor/kreuzberg-pdfium-render/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "kreuzberg-pdfium-render"
-version = "4.10.3"
+version = "4.10.4"
 edition = "2024"
 rust-version = "1.91"
 authors = ["Na'aman Hirschfeld <naaman@kreuzberg.dev>"]
@@ -58,7 +58,7 @@ wasm-bindgen-futures = "0.4"
 web-sys = { version = "0.3", features = ["Blob", "File", "FileReader", "console", "TextDecoder", "ImageData", "Window", "Response"] }
 
 [build-dependencies]
-bindgen = { version = "0.72", optional = true }
+bindgen = { version = "0.73.2", optional = true }
 
 [lints.rust.unexpected_cfgs]
 level = "allow"

data/vendor/kreuzberg-tesseract/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "kreuzberg-tesseract"
-version = "4.10.3"
+version = "4.10.4"
 edition = "2024"
 rust-version = "1.91"
 authors = ["Na'aman Hirschfeld <naaman@kreuzberg.dev>"]
@@ -38,21 +38,21 @@ dynamic-linking = []
 thiserror = "2.0.20"
 
 [build-dependencies]
-cc = { version = "^1.4.5", optional = true }
+cc = { version = "1.4.7", optional = true }
 cmake = { version = "0.1.58", optional = true }
 zip = { version = ">=7.0.0", optional = true, default-features = false, features = [
     "deflate-flate2-zlib-rs",
 ] }
 
 [target.'cfg(not(target_os = "windows"))'.build-dependencies]
-reqwest = { version = "0.13.4", default-features = false, features = [
+reqwest = { version = "0.13.5", default-features = false, features = [
     "blocking",
     "rustls",
 ], optional = true }
 
 # Use native-tls on Windows to avoid aws-lc-sys CMake build issues with MinGW
 [target.'cfg(target_os = "windows")'.build-dependencies]
-reqwest = { version = "0.13.4", default-features = false, features = [
+reqwest = { version = "0.13.5", default-features = false, features = [
     "blocking",
     "native-tls",
 ], optional = true }

data/vendor/kreuzberg-tesseract/src/api.rs

@@ -934,6 +934,17 @@ impl TesseractAPI {
         if vec_ptr.is_null() {
             return Err(TesseractError::NullPointerError);
         }
+
+        struct TextArrayGuard(*mut *mut c_char);
+        impl Drop for TextArrayGuard {
+            fn drop(&mut self) {
+                // SAFETY: `TessDeleteTextArray()` deallocates both the array
+                // and all contained strings.
+                unsafe { TessDeleteTextArray(self.0) };
+            }
+        }
+        let _text_array = TextArrayGuard(vec_ptr);
+
         let mut result = Vec::new();
         let mut i = 0;
         loop {
@@ -947,8 +958,6 @@ impl TesseractAPI {
             result.push(c_str.to_str()?.to_owned());
             i += 1;
         }
-        // SAFETY: TessDeleteTextArray() deallocates both the array and all contained strings:
-        unsafe { TessDeleteTextArray(vec_ptr) };
         Ok(result)
     }
 

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: kreuzberg
 version: !ruby/object:Gem::Version
-  version: 4.10.3
+  version: 4.10.4
 platform: ruby
 authors:
 - Na'aman Hirschfeld
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2026-09-05 00:00:00.000000000 Z
+date: 2026-09-21 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: rb_sys