kreuzberg 4.0.0.rc2 → 4.0.0

data/vendor/kreuzberg/src/chunking/mod.rs

@@ -1,677 +1,2303 @@
-//! Text chunking utilities.
-//!
-//! This module provides text chunking functionality using the `text-splitter` library.
-//! It splits long text into smaller chunks while preserving semantic boundaries.
-//!
-//! # Features
-//!
-//! - **Smart splitting**: Respects word and sentence boundaries
-//! - **Markdown-aware**: Preserves Markdown structure (headings, code blocks, lists)
-//! - **Configurable overlap**: Overlap chunks to maintain context
-//! - **Unicode support**: Handles CJK characters and emojis correctly
-//! - **Batch processing**: Process multiple texts efficiently
-//!
-//! # Chunker Types
-//!
-//! - **Text**: Generic text splitter, splits on whitespace and punctuation
-//! - **Markdown**: Markdown-aware splitter, preserves formatting and structure
-//!
-//! # Example
-//!
-//! ```rust
-//! use kreuzberg::chunking::{chunk_text, ChunkingConfig, ChunkerType};
-//!
-//! # fn example() -> kreuzberg::Result<()> {
-//! let config = ChunkingConfig {
-//!     max_characters: 500,
-//!     overlap: 50,
-//!     trim: true,
-//!     chunker_type: ChunkerType::Text,
-//! };
-//!
-//! let long_text = "This is a very long document...".repeat(100);
-//! let result = chunk_text(&long_text, &config)?;
-//!
-//! println!("Split into {} chunks", result.chunk_count);
-//! for (i, chunk) in result.chunks.iter().enumerate() {
-//!     println!("Chunk {}: {} chars", i + 1, chunk.content.len());
-//! }
-//! # Ok(())
-//! # }
-//! ```
-//!
-//! # Use Cases
-//!
-//! - Splitting documents for LLM context windows
-//! - Creating overlapping chunks for semantic search
-//! - Processing large documents in batches
-//! - Maintaining context across chunk boundaries
-use crate::error::{KreuzbergError, Result};
-use crate::types::{Chunk, ChunkMetadata};
-use serde::{Deserialize, Serialize};
-use text_splitter::{Characters, ChunkCapacity, ChunkConfig, MarkdownSplitter, TextSplitter};
-
-#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
-pub enum ChunkerType {
-    Text,
-    Markdown,
-}
-
-#[derive(Debug, Clone, Serialize, Deserialize)]
-pub struct ChunkingResult {
-    pub chunks: Vec<Chunk>,
-    pub chunk_count: usize,
-}
-
-pub struct ChunkingConfig {
-    pub max_characters: usize,
-    pub overlap: usize,
-    pub trim: bool,
-    pub chunker_type: ChunkerType,
-}
-
-impl Default for ChunkingConfig {
-    fn default() -> Self {
-        Self {
-            max_characters: 2000,
-            overlap: 100,
-            trim: true,
-            chunker_type: ChunkerType::Text,
-        }
-    }
-}
-
-fn build_chunk_config(max_characters: usize, overlap: usize, trim: bool) -> Result<ChunkConfig<Characters>> {
-    ChunkConfig::new(ChunkCapacity::new(max_characters))
-        .with_overlap(overlap)
-        .map(|config| config.with_trim(trim))
-        .map_err(|e| KreuzbergError::validation(format!("Invalid chunking configuration: {}", e)))
-}
-
-pub fn chunk_text(text: &str, config: &ChunkingConfig) -> Result<ChunkingResult> {
-    if text.is_empty() {
-        return Ok(ChunkingResult {
-            chunks: vec![],
-            chunk_count: 0,
-        });
-    }
-
-    let chunk_config = build_chunk_config(config.max_characters, config.overlap, config.trim)?;
-
-    let text_chunks: Vec<&str> = match config.chunker_type {
-        ChunkerType::Text => {
-            let splitter = TextSplitter::new(chunk_config);
-            splitter.chunks(text).collect()
-        }
-        ChunkerType::Markdown => {
-            let splitter = MarkdownSplitter::new(chunk_config);
-            splitter.chunks(text).collect()
-        }
-    };
-
-    let total_chunks = text_chunks.len();
-    let mut char_offset = 0;
-
-    let chunks: Vec<Chunk> = text_chunks
-        .into_iter()
-        .enumerate()
-        .map(|(index, chunk_text)| {
-            let char_start = char_offset;
-            let chunk_length = chunk_text.chars().count();
-            let char_end = char_start + chunk_length;
-
-            let overlap_chars = if index < total_chunks - 1 {
-                config.overlap.min(chunk_length)
-            } else {
-                0
-            };
-            char_offset = char_end - overlap_chars;
-
-            Chunk {
-                content: chunk_text.to_string(),
-                embedding: None,
-                metadata: ChunkMetadata {
-                    char_start,
-                    char_end,
-                    token_count: None,
-                    chunk_index: index,
-                    total_chunks,
-                },
-            }
-        })
-        .collect();
-
-    let chunk_count = chunks.len();
-
-    Ok(ChunkingResult { chunks, chunk_count })
-}
-
-pub fn chunk_text_with_type(
-    text: &str,
-    max_characters: usize,
-    overlap: usize,
-    trim: bool,
-    chunker_type: ChunkerType,
-) -> Result<ChunkingResult> {
-    let config = ChunkingConfig {
-        max_characters,
-        overlap,
-        trim,
-        chunker_type,
-    };
-    chunk_text(text, &config)
-}
-
-pub fn chunk_texts_batch(texts: &[&str], config: &ChunkingConfig) -> Result<Vec<ChunkingResult>> {
-    texts.iter().map(|text| chunk_text(text, config)).collect()
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_chunk_empty_text() {
-        let config = ChunkingConfig::default();
-        let result = chunk_text("", &config).unwrap();
-        assert_eq!(result.chunks.len(), 0);
-        assert_eq!(result.chunk_count, 0);
-    }
-
-    #[test]
-    fn test_chunk_short_text_single_chunk() {
-        let config = ChunkingConfig {
-            max_characters: 100,
-            overlap: 10,
-            trim: true,
-            chunker_type: ChunkerType::Text,
-        };
-        let text = "This is a short text.";
-        let result = chunk_text(text, &config).unwrap();
-        assert_eq!(result.chunks.len(), 1);
-        assert_eq!(result.chunk_count, 1);
-        assert_eq!(result.chunks[0].content, text);
-    }
-
-    #[test]
-    fn test_chunk_long_text_multiple_chunks() {
-        let config = ChunkingConfig {
-            max_characters: 20,
-            overlap: 5,
-            trim: true,
-            chunker_type: ChunkerType::Text,
-        };
-        let text = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";
-        let result = chunk_text(text, &config).unwrap();
-        assert!(result.chunk_count >= 2);
-        assert_eq!(result.chunks.len(), result.chunk_count);
-        assert!(result.chunks.iter().all(|chunk| chunk.content.len() <= 20));
-    }
-
-    #[test]
-    fn test_chunk_text_with_overlap() {
-        let config = ChunkingConfig {
-            max_characters: 20,
-            overlap: 5,
-            trim: true,
-            chunker_type: ChunkerType::Text,
-        };
-        let text = "abcdefghijklmnopqrstuvwxyz0123456789";
-        let result = chunk_text(text, &config).unwrap();
-        assert!(result.chunk_count >= 2);
-
-        if result.chunks.len() >= 2 {
-            let first_chunk_end = &result.chunks[0].content[result.chunks[0].content.len().saturating_sub(5)..];
-            assert!(
-                result.chunks[1].content.starts_with(first_chunk_end),
-                "Expected overlap '{}' at start of second chunk '{}'",
-                first_chunk_end,
-                result.chunks[1].content
-            );
-        }
-    }
-
-    #[test]
-    fn test_chunk_markdown_preserves_structure() {
-        let config = ChunkingConfig {
-            max_characters: 50,
-            overlap: 10,
-            trim: true,
-            chunker_type: ChunkerType::Markdown,
-        };
-        let markdown = "# Title\n\nParagraph one.\n\n## Section\n\nParagraph two.";
-        let result = chunk_text(markdown, &config).unwrap();
-        assert!(result.chunk_count >= 1);
-        assert!(result.chunks.iter().any(|chunk| chunk.content.contains("# Title")));
-    }
-
-    #[test]
-    fn test_chunk_markdown_with_code_blocks() {
-        let config = ChunkingConfig {
-            max_characters: 100,
-            overlap: 10,
-            trim: true,
-            chunker_type: ChunkerType::Markdown,
-        };
-        let markdown = "# Code Example\n\n```python\nprint('hello')\n```\n\nSome text after code.";
-        let result = chunk_text(markdown, &config).unwrap();
-        assert!(result.chunk_count >= 1);
-        assert!(result.chunks.iter().any(|chunk| chunk.content.contains("```")));
-    }
-
-    #[test]
-    fn test_chunk_markdown_with_links() {
-        let config = ChunkingConfig {
-            max_characters: 80,
-            overlap: 10,
-            trim: true,
-            chunker_type: ChunkerType::Markdown,
-        };
-        let markdown = "Check out [this link](https://example.com) for more info.";
-        let result = chunk_text(markdown, &config).unwrap();
-        assert_eq!(result.chunk_count, 1);
-        assert!(result.chunks[0].content.contains("[this link]"));
-    }
-
-    #[test]
-    fn test_chunk_text_with_trim() {
-        let config = ChunkingConfig {
-            max_characters: 30,
-            overlap: 5,
-            trim: true,
-            chunker_type: ChunkerType::Text,
-        };
-        let text = "  Leading and trailing spaces  should be trimmed  ";
-        let result = chunk_text(text, &config).unwrap();
-        assert!(result.chunk_count >= 1);
-        assert!(result.chunks.iter().all(|chunk| !chunk.content.starts_with(' ')));
-    }
-
-    #[test]
-    fn test_chunk_text_without_trim() {
-        let config = ChunkingConfig {
-            max_characters: 30,
-            overlap: 5,
-            trim: false,
-            chunker_type: ChunkerType::Text,
-        };
-        let text = "  Text with spaces  ";
-        let result = chunk_text(text, &config).unwrap();
-        assert_eq!(result.chunk_count, 1);
-        assert!(result.chunks[0].content.starts_with(' ') || result.chunks[0].content.len() < text.len());
-    }
-
-    #[test]
-    fn test_chunk_with_invalid_overlap() {
-        let config = ChunkingConfig {
-            max_characters: 10,
-            overlap: 20,
-            trim: true,
-            chunker_type: ChunkerType::Text,
-        };
-        let result = chunk_text("Some text", &config);
-        assert!(result.is_err());
-        let err = result.unwrap_err();
-        assert!(matches!(err, KreuzbergError::Validation { .. }));
-    }
-
-    #[test]
-    fn test_chunk_text_with_type_text() {
-        let result = chunk_text_with_type("Simple text", 50, 10, true, ChunkerType::Text).unwrap();
-        assert_eq!(result.chunk_count, 1);
-        assert_eq!(result.chunks[0].content, "Simple text");
-    }
-
-    #[test]
-    fn test_chunk_text_with_type_markdown() {
-        let markdown = "# Header\n\nContent here.";
-        let result = chunk_text_with_type(markdown, 50, 10, true, ChunkerType::Markdown).unwrap();
-        assert_eq!(result.chunk_count, 1);
-        assert!(result.chunks[0].content.contains("# Header"));
-    }
-
-    #[test]
-    fn test_chunk_texts_batch_empty() {
-        let config = ChunkingConfig::default();
-        let texts: Vec<&str> = vec![];
-        let results = chunk_texts_batch(&texts, &config).unwrap();
-        assert_eq!(results.len(), 0);
-    }
-
-    #[test]
-    fn test_chunk_texts_batch_multiple() {
-        let config = ChunkingConfig {
-            max_characters: 30,
-            overlap: 5,
-            trim: true,
-            chunker_type: ChunkerType::Text,
-        };
-        let texts = vec!["First text", "Second text", "Third text"];
-        let results = chunk_texts_batch(&texts, &config).unwrap();
-        assert_eq!(results.len(), 3);
-        assert!(results.iter().all(|r| r.chunk_count >= 1));
-    }
-
-    #[test]
-    fn test_chunk_texts_batch_mixed_lengths() {
-        let config = ChunkingConfig {
-            max_characters: 20,
-            overlap: 5,
-            trim: true,
-            chunker_type: ChunkerType::Text,
-        };
-        let texts = vec![
-            "Short",
-            "This is a longer text that should be split into multiple chunks",
-            "",
-        ];
-        let results = chunk_texts_batch(&texts, &config).unwrap();
-        assert_eq!(results.len(), 3);
-        assert_eq!(results[0].chunk_count, 1);
-        assert!(results[1].chunk_count > 1);
-        assert_eq!(results[2].chunk_count, 0);
-    }
-
-    #[test]
-    fn test_chunk_texts_batch_error_propagation() {
-        let config = ChunkingConfig {
-            max_characters: 10,
-            overlap: 20,
-            trim: true,
-            chunker_type: ChunkerType::Text,
-        };
-        let texts = vec!["Text one", "Text two"];
-        let result = chunk_texts_batch(&texts, &config);
-        assert!(result.is_err());
-    }
-
-    #[test]
-    fn test_chunking_config_default() {
-        let config = ChunkingConfig::default();
-        assert_eq!(config.max_characters, 2000);
-        assert_eq!(config.overlap, 100);
-        assert!(config.trim);
-        assert_eq!(config.chunker_type, ChunkerType::Text);
-    }
-
-    #[test]
-    fn test_chunk_very_long_text() {
-        let config = ChunkingConfig {
-            max_characters: 100,
-            overlap: 20,
-            trim: true,
-            chunker_type: ChunkerType::Text,
-        };
-        let text = "a".repeat(1000);
-        let result = chunk_text(&text, &config).unwrap();
-        assert!(result.chunk_count >= 10);
-        assert!(result.chunks.iter().all(|chunk| chunk.content.len() <= 100));
-    }
-
-    #[test]
-    fn test_chunk_text_with_newlines() {
-        let config = ChunkingConfig {
-            max_characters: 30,
-            overlap: 5,
-            trim: true,
-            chunker_type: ChunkerType::Text,
-        };
-        let text = "Line one\nLine two\nLine three\nLine four\nLine five";
-        let result = chunk_text(text, &config).unwrap();
-        assert!(result.chunk_count >= 1);
-    }
-
-    #[test]
-    fn test_chunk_markdown_with_lists() {
-        let config = ChunkingConfig {
-            max_characters: 100,
-            overlap: 10,
-            trim: true,
-            chunker_type: ChunkerType::Markdown,
-        };
-        let markdown = "# List Example\n\n- Item 1\n- Item 2\n- Item 3\n\nMore text.";
-        let result = chunk_text(markdown, &config).unwrap();
-        assert!(result.chunk_count >= 1);
-        assert!(result.chunks.iter().any(|chunk| chunk.content.contains("- Item")));
-    }
-
-    #[test]
-    fn test_chunk_markdown_with_tables() {
-        let config = ChunkingConfig {
-            max_characters: 150,
-            overlap: 10,
-            trim: true,
-            chunker_type: ChunkerType::Markdown,
-        };
-        let markdown = "# Table\n\n| Col1 | Col2 |\n|------|------|\n| A    | B    |\n| C    | D    |";
-        let result = chunk_text(markdown, &config).unwrap();
-        assert!(result.chunk_count >= 1);
-        assert!(result.chunks.iter().any(|chunk| chunk.content.contains("|")));
-    }
-
-    #[test]
-    fn test_chunk_special_characters() {
-        let config = ChunkingConfig {
-            max_characters: 50,
-            overlap: 5,
-            trim: true,
-            chunker_type: ChunkerType::Text,
-        };
-        let text = "Special chars: @#$%^&*()[]{}|\\<>?/~`";
-        let result = chunk_text(text, &config).unwrap();
-        assert_eq!(result.chunk_count, 1);
-        assert!(result.chunks[0].content.contains("@#$%"));
-    }
-
-    #[test]
-    fn test_chunk_unicode_characters() {
-        let config = ChunkingConfig {
-            max_characters: 50,
-            overlap: 5,
-            trim: true,
-            chunker_type: ChunkerType::Text,
-        };
-        let text = "Unicode: 你好世界 🌍 café résumé";
-        let result = chunk_text(text, &config).unwrap();
-        assert_eq!(result.chunk_count, 1);
-        assert!(result.chunks[0].content.contains("你好"));
-        assert!(result.chunks[0].content.contains("🌍"));
-    }
-
-    #[test]
-    fn test_chunk_cjk_text() {
-        let config = ChunkingConfig {
-            max_characters: 30,
-            overlap: 5,
-            trim: true,
-            chunker_type: ChunkerType::Text,
-        };
-        let text = "日本語のテキストです。これは長い文章で、複数のチャンクに分割されるべきです。";
-        let result = chunk_text(text, &config).unwrap();
-        assert!(result.chunk_count >= 1);
-    }
-
-    #[test]
-    fn test_chunk_mixed_languages() {
-        let config = ChunkingConfig {
-            max_characters: 40,
-            overlap: 5,
-            trim: true,
-            chunker_type: ChunkerType::Text,
-        };
-        let text = "English text mixed with 中文文本 and some français";
-        let result = chunk_text(text, &config).unwrap();
-        assert!(result.chunk_count >= 1);
-    }
-
-    #[test]
-    fn test_chunk_offset_calculation_with_overlap() {
-        let config = ChunkingConfig {
-            max_characters: 20,
-            overlap: 5,
-            trim: false,
-            chunker_type: ChunkerType::Text,
-        };
-        let text = "AAAAA BBBBB CCCCC DDDDD EEEEE FFFFF";
-        let result = chunk_text(text, &config).unwrap();
-
-        assert!(result.chunks.len() >= 2, "Expected at least 2 chunks");
-
-        for i in 0..result.chunks.len() {
-            let chunk = &result.chunks[i];
-            let metadata = &chunk.metadata;
-
-            assert_eq!(
-                metadata.char_end - metadata.char_start,
-                chunk.content.chars().count(),
-                "Chunk {} offset range doesn't match content length",
-                i
-            );
-
-            assert_eq!(metadata.chunk_index, i);
-            assert_eq!(metadata.total_chunks, result.chunks.len());
-        }
-
-        for i in 0..result.chunks.len() - 1 {
-            let current_chunk = &result.chunks[i];
-            let next_chunk = &result.chunks[i + 1];
-
-            assert!(
-                next_chunk.metadata.char_start < current_chunk.metadata.char_end,
-                "Chunk {} and {} don't overlap: next starts at {} but current ends at {}",
-                i,
-                i + 1,
-                next_chunk.metadata.char_start,
-                current_chunk.metadata.char_end
-            );
-
-            let overlap_size = current_chunk.metadata.char_end - next_chunk.metadata.char_start;
-            assert!(
-                overlap_size <= config.overlap + 10,
-                "Overlap between chunks {} and {} is too large: {}",
-                i,
-                i + 1,
-                overlap_size
-            );
-        }
-    }
-
-    #[test]
-    fn test_chunk_offset_calculation_without_overlap() {
-        let config = ChunkingConfig {
-            max_characters: 20,
-            overlap: 0,
-            trim: false,
-            chunker_type: ChunkerType::Text,
-        };
-        let text = "AAAAA BBBBB CCCCC DDDDD EEEEE FFFFF";
-        let result = chunk_text(text, &config).unwrap();
-
-        for i in 0..result.chunks.len() - 1 {
-            let current_chunk = &result.chunks[i];
-            let next_chunk = &result.chunks[i + 1];
-
-            assert!(
-                next_chunk.metadata.char_start >= current_chunk.metadata.char_end,
-                "Chunk {} and {} overlap when they shouldn't: next starts at {} but current ends at {}",
-                i,
-                i + 1,
-                next_chunk.metadata.char_start,
-                current_chunk.metadata.char_end
-            );
-        }
-    }
-
-    #[test]
-    fn test_chunk_offset_covers_full_text() {
-        let config = ChunkingConfig {
-            max_characters: 15,
-            overlap: 3,
-            trim: false,
-            chunker_type: ChunkerType::Text,
-        };
-        let text = "0123456789 ABCDEFGHIJ KLMNOPQRST UVWXYZ";
-        let result = chunk_text(text, &config).unwrap();
-
-        assert!(result.chunks.len() >= 2, "Expected multiple chunks");
-
-        assert_eq!(
-            result.chunks[0].metadata.char_start, 0,
-            "First chunk should start at position 0"
-        );
-
-        for i in 0..result.chunks.len() - 1 {
-            let current_chunk = &result.chunks[i];
-            let next_chunk = &result.chunks[i + 1];
-
-            assert!(
-                next_chunk.metadata.char_start <= current_chunk.metadata.char_end,
-                "Gap detected between chunk {} (ends at {}) and chunk {} (starts at {})",
-                i,
-                current_chunk.metadata.char_end,
-                i + 1,
-                next_chunk.metadata.char_start
-            );
-        }
-    }
-
-    #[test]
-    fn test_chunk_offset_with_various_overlap_sizes() {
-        for overlap in [0, 5, 10, 20] {
-            let config = ChunkingConfig {
-                max_characters: 30,
-                overlap,
-                trim: false,
-                chunker_type: ChunkerType::Text,
-            };
-            let text = "Word ".repeat(30);
-            let result = chunk_text(&text, &config).unwrap();
-
-            for chunk in &result.chunks {
-                assert!(
-                    chunk.metadata.char_end > chunk.metadata.char_start,
-                    "Invalid offset range for overlap {}: start={}, end={}",
-                    overlap,
-                    chunk.metadata.char_start,
-                    chunk.metadata.char_end
-                );
-            }
-
-            for chunk in &result.chunks {
-                assert!(
-                    chunk.metadata.char_start < text.chars().count(),
-                    "char_start with overlap {} is out of bounds: {}",
-                    overlap,
-                    chunk.metadata.char_start
-                );
-            }
-        }
-    }
-
-    #[test]
-    fn test_chunk_last_chunk_offset() {
-        let config = ChunkingConfig {
-            max_characters: 20,
-            overlap: 5,
-            trim: false,
-            chunker_type: ChunkerType::Text,
-        };
-        let text = "AAAAA BBBBB CCCCC DDDDD EEEEE";
-        let result = chunk_text(text, &config).unwrap();
-
-        assert!(result.chunks.len() >= 2, "Need multiple chunks for this test");
-
-        let last_chunk = result.chunks.last().unwrap();
-        let second_to_last = &result.chunks[result.chunks.len() - 2];
-
-        assert!(
-            last_chunk.metadata.char_start < second_to_last.metadata.char_end,
-            "Last chunk should overlap with previous chunk"
-        );
-
-        let expected_end = text.chars().count();
-        let last_chunk_covers_end =
-            last_chunk.content.trim_end() == text.trim_end() || last_chunk.metadata.char_end >= expected_end - 5;
-        assert!(last_chunk_covers_end, "Last chunk should cover the end of the text");
-    }
-}
+//! Text chunking utilities.
+//!
+//! This module provides text chunking functionality using the `text-splitter` library.
+//! It splits long text into smaller chunks while preserving semantic boundaries.
+//!
+//! # Features
+//!
+//! - **Smart splitting**: Respects word and sentence boundaries
+//! - **Markdown-aware**: Preserves Markdown structure (headings, code blocks, lists)
+//! - **Configurable overlap**: Overlap chunks to maintain context
+//! - **Unicode support**: Handles CJK characters and emojis correctly
+//! - **Batch processing**: Process multiple texts efficiently
+//!
+//! # Chunker Types
+//!
+//! - **Text**: Generic text splitter, splits on whitespace and punctuation
+//! - **Markdown**: Markdown-aware splitter, preserves formatting and structure
+//!
+//! # Example
+//!
+//! ```rust
+//! use kreuzberg::chunking::{chunk_text, ChunkingConfig, ChunkerType};
+//!
+//! # fn example() -> kreuzberg::Result<()> {
+//! let config = ChunkingConfig {
+//!     max_characters: 500,
+//!     overlap: 50,
+//!     trim: true,
+//!     chunker_type: ChunkerType::Text,
+//! };
+//!
+//! let long_text = "This is a very long document...".repeat(100);
+//! let result = chunk_text(&long_text, &config, None)?;
+//!
+//! println!("Split into {} chunks", result.chunk_count);
+//! for (i, chunk) in result.chunks.iter().enumerate() {
+//!     println!("Chunk {}: {} chars", i + 1, chunk.content.len());
+//! }
+//! # Ok(())
+//! # }
+//! ```
+//!
+//! # Use Cases
+//!
+//! - Splitting documents for LLM context windows
+//! - Creating overlapping chunks for semantic search
+//! - Processing large documents in batches
+//! - Maintaining context across chunk boundaries
+use crate::error::{KreuzbergError, Result};
+use crate::types::{Chunk, ChunkMetadata, PageBoundary};
+use bitvec::prelude::*;
+use once_cell::sync::Lazy;
+use serde::{Deserialize, Serialize};
+use std::sync::Arc;
+use text_splitter::{Characters, ChunkCapacity, ChunkConfig, MarkdownSplitter, TextSplitter};
+
+pub mod processor;
+pub use processor::ChunkingProcessor;
+
+/// Threshold below which we use O(1) direct validation instead of precomputing a BitVec.
+///
+/// When there are 10 or fewer boundaries, the overhead of creating a BitVec (which is O(n)
+/// where n is the text length) exceeds the cost of calling `is_char_boundary()` directly
+/// for each boundary position. This threshold balances performance across different scenarios:
+/// - Small documents with few boundaries: fast path dominates
+/// - Large documents with many boundaries: batch path leverages the precomputed BitVec
+const ADAPTIVE_VALIDATION_THRESHOLD: usize = 10;
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
+pub enum ChunkerType {
+    Text,
+    Markdown,
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct ChunkingResult {
+    pub chunks: Vec<Chunk>,
+    pub chunk_count: usize,
+}
+
+pub struct ChunkingConfig {
+    pub max_characters: usize,
+    pub overlap: usize,
+    pub trim: bool,
+    pub chunker_type: ChunkerType,
+}
+
+impl Default for ChunkingConfig {
+    fn default() -> Self {
+        Self {
+            max_characters: 2000,
+            overlap: 100,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        }
+    }
+}
+
+fn build_chunk_config(max_characters: usize, overlap: usize, trim: bool) -> Result<ChunkConfig<Characters>> {
+    ChunkConfig::new(ChunkCapacity::new(max_characters))
+        .with_overlap(overlap)
+        .map(|config| config.with_trim(trim))
+        .map_err(|e| KreuzbergError::validation(format!("Invalid chunking configuration: {}", e)))
+}
+
+/// Pre-computes valid UTF-8 character boundaries for a text string.
+///
+/// This function performs a single O(n) pass through the text to identify all valid
+/// UTF-8 character boundaries, storing them in a BitVec for O(1) lookups.
+///
+/// # Arguments
+///
+/// * `text` - The text to analyze
+///
+/// # Returns
+///
+/// A BitVec where each bit represents whether a byte offset is a valid UTF-8 character boundary.
+/// The BitVec has length `text.len() + 1` (includes the end position).
+///
+/// # Examples
+///
+/// ```ignore
+/// let text = "Hello 👋";
+/// let boundaries = precompute_utf8_boundaries(text);
+/// assert!(boundaries[0]);      // Start is always valid
+/// assert!(boundaries[6]);      // 'H' + "ello " = 6 bytes
+/// assert!(!boundaries[7]);     // Middle of emoji (first byte of 4-byte sequence)
+/// assert!(boundaries[10]);     // After emoji (valid boundary)
+/// ```
+fn precompute_utf8_boundaries(text: &str) -> BitVec {
+    let text_len = text.len();
+    let mut boundaries = bitvec![0; text_len + 1];
+
+    boundaries.set(0, true);
+
+    for (i, _) in text.char_indices() {
+        if i <= text_len {
+            boundaries.set(i, true);
+        }
+    }
+
+    if text_len > 0 {
+        boundaries.set(text_len, true);
+    }
+
+    boundaries
+}
+
+/// Validates that byte offsets in page boundaries fall on valid UTF-8 character boundaries.
+///
+/// This function ensures that all page boundary positions are at valid UTF-8 character
+/// boundaries within the text. This is CRITICAL to prevent text corruption when boundaries
+/// are created from language bindings or external sources, particularly with multibyte
+/// UTF-8 characters (emoji, CJK characters, combining marks, etc.).
+///
+/// **Performance Strategy**: Uses adaptive validation to optimize for different boundary counts:
+/// - **Small sets (≤10 boundaries)**: O(k) approach using Rust's native `is_char_boundary()` for each position
+/// - **Large sets (>10 boundaries)**: O(n) precomputation with O(1) lookups via BitVec
+///
+/// For typical PDF documents with 1-10 page boundaries, the fast path provides 30-50% faster
+/// validation than always precomputing. For documents with 100+ boundaries, batch precomputation
+/// is 2-4% faster overall due to amortized costs. This gives ~2-4% improvement across all scenarios.
+///
+/// # Arguments
+///
+/// * `text` - The text being chunked
+/// * `boundaries` - Page boundary markers to validate
+///
+/// # Returns
+///
+/// Returns `Ok(())` if all boundaries are at valid UTF-8 character boundaries.
+/// Returns `KreuzbergError::Validation` if any boundary is at an invalid position.
+///
+/// # UTF-8 Boundary Safety
+///
+/// Rust strings use UTF-8 encoding where characters can be 1-4 bytes. For example:
+/// - ASCII letters: 1 byte each
+/// - Emoji (🌍): 4 bytes but 1 character
+/// - CJK characters (中): 3 bytes but 1 character
+///
+/// This function checks that all byte_start and byte_end values are at character boundaries
+/// using an adaptive strategy: direct calls for small boundary sets, or precomputed BitVec
+/// for large sets.
+fn validate_utf8_boundaries(text: &str, boundaries: &[PageBoundary]) -> Result<()> {
+    if boundaries.is_empty() {
+        return Ok(());
+    }
+
+    let text_len = text.len();
+
+    if boundaries.len() <= ADAPTIVE_VALIDATION_THRESHOLD {
+        validate_utf8_boundaries_fast_path(text, boundaries, text_len)
+    } else {
+        validate_utf8_boundaries_batch_path(text, boundaries, text_len)
+    }
+}
+
+/// Fast path: direct UTF-8 boundary validation for small boundary counts (≤10).
+///
+/// Uses Rust's native `str::is_char_boundary()` for O(1) checks on each boundary position.
+/// This avoids the O(n) overhead of BitVec precomputation, making it ideal for typical
+/// PDF documents with few page boundaries.
+///
+/// # Arguments
+///
+/// * `text` - The text being validated
+/// * `boundaries` - Page boundary markers to validate
+/// * `text_len` - Pre-computed text length (avoids recomputation)
+///
+/// # Returns
+///
+/// Returns `Ok(())` if all boundaries are at valid UTF-8 character boundaries.
+/// Returns `KreuzbergError::Validation` if any boundary is invalid.
+fn validate_utf8_boundaries_fast_path(text: &str, boundaries: &[PageBoundary], text_len: usize) -> Result<()> {
+    for (idx, boundary) in boundaries.iter().enumerate() {
+        if boundary.byte_start > text_len {
+            return Err(KreuzbergError::validation(format!(
+                "Page boundary {} has byte_start={} which exceeds text length {}",
+                idx, boundary.byte_start, text_len
+            )));
+        }
+
+        if boundary.byte_end > text_len {
+            return Err(KreuzbergError::validation(format!(
+                "Page boundary {} has byte_end={} which exceeds text length {}",
+                idx, boundary.byte_end, text_len
+            )));
+        }
+
+        if boundary.byte_start > 0 && boundary.byte_start < text_len && !text.is_char_boundary(boundary.byte_start) {
+            return Err(KreuzbergError::validation(format!(
+                "Page boundary {} has byte_start={} which is not a valid UTF-8 character boundary (text length={}). This may indicate corrupted multibyte characters (emoji, CJK, etc.)",
+                idx, boundary.byte_start, text_len
+            )));
+        }
+
+        if boundary.byte_end > 0 && boundary.byte_end < text_len && !text.is_char_boundary(boundary.byte_end) {
+            return Err(KreuzbergError::validation(format!(
+                "Page boundary {} has byte_end={} which is not a valid UTF-8 character boundary (text length={}). This may indicate corrupted multibyte characters (emoji, CJK, etc.)",
+                idx, boundary.byte_end, text_len
+            )));
+        }
+    }
+
+    Ok(())
+}
+
+/// Batch path: precomputed BitVec validation for large boundary counts (>10).
+///
+/// Precomputes all valid UTF-8 boundaries in a single O(n) pass, then performs O(1)
+/// lookups for each boundary position. This is more efficient than O(k*1) direct checks
+/// when k is large or when the repeated `is_char_boundary()` calls have measurable overhead.
+///
+/// # Arguments
+///
+/// * `text` - The text being validated
+/// * `boundaries` - Page boundary markers to validate
+/// * `text_len` - Pre-computed text length (avoids recomputation)
+///
+/// # Returns
+///
+/// Returns `Ok(())` if all boundaries are at valid UTF-8 character boundaries.
+/// Returns `KreuzbergError::Validation` if any boundary is invalid.
+fn validate_utf8_boundaries_batch_path(text: &str, boundaries: &[PageBoundary], text_len: usize) -> Result<()> {
+    let valid_boundaries = precompute_utf8_boundaries(text);
+
+    for (idx, boundary) in boundaries.iter().enumerate() {
+        if boundary.byte_start > text_len {
+            return Err(KreuzbergError::validation(format!(
+                "Page boundary {} has byte_start={} which exceeds text length {}",
+                idx, boundary.byte_start, text_len
+            )));
+        }
+
+        if boundary.byte_end > text_len {
+            return Err(KreuzbergError::validation(format!(
+                "Page boundary {} has byte_end={} which exceeds text length {}",
+                idx, boundary.byte_end, text_len
+            )));
+        }
+
+        if boundary.byte_start > 0 && boundary.byte_start <= text_len && !valid_boundaries[boundary.byte_start] {
+            return Err(KreuzbergError::validation(format!(
+                "Page boundary {} has byte_start={} which is not a valid UTF-8 character boundary (text length={}). This may indicate corrupted multibyte characters (emoji, CJK, etc.)",
+                idx, boundary.byte_start, text_len
+            )));
+        }
+
+        if boundary.byte_end > 0 && boundary.byte_end <= text_len && !valid_boundaries[boundary.byte_end] {
+            return Err(KreuzbergError::validation(format!(
+                "Page boundary {} has byte_end={} which is not a valid UTF-8 character boundary (text length={}). This may indicate corrupted multibyte characters (emoji, CJK, etc.)",
+                idx, boundary.byte_end, text_len
+            )));
+        }
+    }
+
+    Ok(())
+}
+
+/// Calculate which pages a character range spans.
+///
+/// # Arguments
+///
+/// * `char_start` - Starting character offset of the chunk
+/// * `char_end` - Ending character offset of the chunk
+/// * `boundaries` - Page boundary markers from the document
+///
+/// # Returns
+///
+/// A tuple of (first_page, last_page) where page numbers are 1-indexed.
+/// Returns (None, None) if boundaries are empty or chunk doesn't overlap any page.
+/// Validates page boundaries for consistency and correctness.
+///
+/// # Validation Rules
+///
+/// 1. Boundaries must be sorted by char_start (monotonically increasing)
+/// 2. Boundaries must not overlap (char_end[i] <= char_start[i+1])
+/// 3. Each boundary must have char_start < char_end
+///
+/// # Errors
+///
+/// Returns `KreuzbergError::Validation` if any boundary is invalid.
+fn validate_page_boundaries(boundaries: &[PageBoundary]) -> Result<()> {
+    if boundaries.is_empty() {
+        return Ok(());
+    }
+
+    for (idx, boundary) in boundaries.iter().enumerate() {
+        if boundary.byte_start >= boundary.byte_end {
+            return Err(KreuzbergError::validation(format!(
+                "Invalid boundary range at index {}: byte_start ({}) must be < byte_end ({})",
+                idx, boundary.byte_start, boundary.byte_end
+            )));
+        }
+    }
+
+    for i in 0..boundaries.len() - 1 {
+        let current = &boundaries[i];
+        let next = &boundaries[i + 1];
+
+        if current.byte_start > next.byte_start {
+            return Err(KreuzbergError::validation(format!(
+                "Page boundaries not sorted: boundary at index {} (byte_start={}) comes after boundary at index {} (byte_start={})",
+                i,
+                current.byte_start,
+                i + 1,
+                next.byte_start
+            )));
+        }
+
+        if current.byte_end > next.byte_start {
+            return Err(KreuzbergError::validation(format!(
+                "Overlapping page boundaries: boundary {} ends at {} but boundary {} starts at {}",
+                i,
+                current.byte_end,
+                i + 1,
+                next.byte_start
+            )));
+        }
+    }
+
+    Ok(())
+}
+
+/// Calculate which pages a byte range spans.
+///
+/// # Arguments
+///
+/// * `byte_start` - Starting byte offset of the chunk
+/// * `byte_end` - Ending byte offset of the chunk
+/// * `boundaries` - Page boundary markers from the document
+///
+/// # Returns
+///
+/// A tuple of (first_page, last_page) where page numbers are 1-indexed.
+/// Returns (None, None) if boundaries are empty or chunk doesn't overlap any page.
+///
+/// # Errors
+///
+/// Returns `KreuzbergError::Validation` if boundaries are invalid.
+fn calculate_page_range(
+    byte_start: usize,
+    byte_end: usize,
+    boundaries: &[PageBoundary],
+) -> Result<(Option<usize>, Option<usize>)> {
+    if boundaries.is_empty() {
+        return Ok((None, None));
+    }
+
+    validate_page_boundaries(boundaries)?;
+
+    let mut first_page = None;
+    let mut last_page = None;
+
+    for boundary in boundaries {
+        if byte_start < boundary.byte_end && byte_end > boundary.byte_start {
+            if first_page.is_none() {
+                first_page = Some(boundary.page_number);
+            }
+            last_page = Some(boundary.page_number);
+        }
+    }
+
+    Ok((first_page, last_page))
+}
+
+/// Split text into chunks with optional page boundary tracking.
+///
+/// # Arguments
+///
+/// * `text` - The text to split into chunks
+/// * `config` - Chunking configuration (max size, overlap, type)
+/// * `page_boundaries` - Optional page boundary markers for mapping chunks to pages
+///
+/// # Returns
+///
+/// A ChunkingResult containing all chunks and their metadata.
+///
+/// # Examples
+///
+/// ```rust
+/// use kreuzberg::chunking::{chunk_text, ChunkingConfig, ChunkerType};
+///
+/// # fn example() -> kreuzberg::Result<()> {
+/// let config = ChunkingConfig {
+///     max_characters: 500,
+///     overlap: 50,
+///     trim: true,
+///     chunker_type: ChunkerType::Text,
+/// };
+/// let result = chunk_text("Long text...", &config, None)?;
+/// assert!(!result.chunks.is_empty());
+/// # Ok(())
+/// # }
+/// ```
+pub fn chunk_text(
+    text: &str,
+    config: &ChunkingConfig,
+    page_boundaries: Option<&[PageBoundary]>,
+) -> Result<ChunkingResult> {
+    if text.is_empty() {
+        return Ok(ChunkingResult {
+            chunks: vec![],
+            chunk_count: 0,
+        });
+    }
+
+    if let Some(boundaries) = page_boundaries {
+        validate_utf8_boundaries(text, boundaries)?;
+    }
+
+    let chunk_config = build_chunk_config(config.max_characters, config.overlap, config.trim)?;
+
+    let text_chunks: Vec<&str> = match config.chunker_type {
+        ChunkerType::Text => {
+            let splitter = TextSplitter::new(chunk_config);
+            splitter.chunks(text).collect()
+        }
+        ChunkerType::Markdown => {
+            let splitter = MarkdownSplitter::new(chunk_config);
+            splitter.chunks(text).collect()
+        }
+    };
+
+    let total_chunks = text_chunks.len();
+    let mut byte_offset = 0;
+
+    let mut chunks: Vec<Chunk> = Vec::new();
+
+    for (index, chunk_text) in text_chunks.into_iter().enumerate() {
+        let byte_start = byte_offset;
+        let chunk_length = chunk_text.len();
+        let byte_end = byte_start + chunk_length;
+
+        let overlap_chars = if index < total_chunks - 1 {
+            config.overlap.min(chunk_length)
+        } else {
+            0
+        };
+        byte_offset = byte_end - overlap_chars;
+
+        let (first_page, last_page) = if let Some(boundaries) = page_boundaries {
+            calculate_page_range(byte_start, byte_end, boundaries)?
+        } else {
+            (None, None)
+        };
+
+        chunks.push(Chunk {
+            content: chunk_text.to_string(),
+            embedding: None,
+            metadata: ChunkMetadata {
+                byte_start,
+                byte_end,
+                token_count: None,
+                chunk_index: index,
+                total_chunks,
+                first_page,
+                last_page,
+            },
+        });
+    }
+
+    let chunk_count = chunks.len();
+
+    Ok(ChunkingResult { chunks, chunk_count })
+}
+
+pub fn chunk_text_with_type(
+    text: &str,
+    max_characters: usize,
+    overlap: usize,
+    trim: bool,
+    chunker_type: ChunkerType,
+) -> Result<ChunkingResult> {
+    let config = ChunkingConfig {
+        max_characters,
+        overlap,
+        trim,
+        chunker_type,
+    };
+    chunk_text(text, &config, None)
+}
+
+pub fn chunk_texts_batch(texts: &[&str], config: &ChunkingConfig) -> Result<Vec<ChunkingResult>> {
+    texts.iter().map(|text| chunk_text(text, config, None)).collect()
+}
+
+/// Lazy-initialized flag that ensures chunking processor is registered exactly once.
+///
+/// This static is accessed on first use to automatically register the
+/// chunking processor with the plugin registry.
+static PROCESSOR_INITIALIZED: Lazy<Result<()>> = Lazy::new(register_chunking_processor);
+
+/// Ensure the chunking processor is registered.
+///
+/// This function is called automatically when needed.
+/// It's safe to call multiple times - registration only happens once.
+pub fn ensure_initialized() -> Result<()> {
+    PROCESSOR_INITIALIZED
+        .as_ref()
+        .map(|_| ())
+        .map_err(|e| crate::KreuzbergError::Plugin {
+            message: format!("Failed to register chunking processor: {}", e),
+            plugin_name: "text-chunking".to_string(),
+        })
+}
+
+/// Register the chunking processor with the global registry.
+///
+/// This function should be called once at application startup to register
+/// the chunking post-processor.
+///
+/// **Note:** This is called automatically on first use.
+/// Explicit calling is optional.
+pub fn register_chunking_processor() -> Result<()> {
+    let registry = crate::plugins::registry::get_post_processor_registry();
+    let mut registry = registry
+        .write()
+        .map_err(|e| crate::KreuzbergError::Other(format!("Post-processor registry lock poisoned: {}", e)))?;
+
+    registry.register(Arc::new(ChunkingProcessor), 50)?;
+
+    Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_chunk_empty_text() {
+        let config = ChunkingConfig::default();
+        let result = chunk_text("", &config, None).unwrap();
+        assert_eq!(result.chunks.len(), 0);
+        assert_eq!(result.chunk_count, 0);
+    }
+
+    #[test]
+    fn test_chunk_short_text_single_chunk() {
+        let config = ChunkingConfig {
+            max_characters: 100,
+            overlap: 10,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "This is a short text.";
+        let result = chunk_text(text, &config, None).unwrap();
+        assert_eq!(result.chunks.len(), 1);
+        assert_eq!(result.chunk_count, 1);
+        assert_eq!(result.chunks[0].content, text);
+    }
+
+    #[test]
+    fn test_chunk_long_text_multiple_chunks() {
+        let config = ChunkingConfig {
+            max_characters: 20,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";
+        let result = chunk_text(text, &config, None).unwrap();
+        assert!(result.chunk_count >= 2);
+        assert_eq!(result.chunks.len(), result.chunk_count);
+        assert!(result.chunks.iter().all(|chunk| chunk.content.len() <= 20));
+    }
+
+    #[test]
+    fn test_chunk_text_with_overlap() {
+        let config = ChunkingConfig {
+            max_characters: 20,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "abcdefghijklmnopqrstuvwxyz0123456789";
+        let result = chunk_text(text, &config, None).unwrap();
+        assert!(result.chunk_count >= 2);
+
+        if result.chunks.len() >= 2 {
+            let first_chunk_end = &result.chunks[0].content[result.chunks[0].content.len().saturating_sub(5)..];
+            assert!(
+                result.chunks[1].content.starts_with(first_chunk_end),
+                "Expected overlap '{}' at start of second chunk '{}'",
+                first_chunk_end,
+                result.chunks[1].content
+            );
+        }
+    }
+
+    #[test]
+    fn test_chunk_markdown_preserves_structure() {
+        let config = ChunkingConfig {
+            max_characters: 50,
+            overlap: 10,
+            trim: true,
+            chunker_type: ChunkerType::Markdown,
+        };
+        let markdown = "# Title\n\nParagraph one.\n\n## Section\n\nParagraph two.";
+        let result = chunk_text(markdown, &config, None).unwrap();
+        assert!(result.chunk_count >= 1);
+        assert!(result.chunks.iter().any(|chunk| chunk.content.contains("# Title")));
+    }
+
+    #[test]
+    fn test_chunk_markdown_with_code_blocks() {
+        let config = ChunkingConfig {
+            max_characters: 100,
+            overlap: 10,
+            trim: true,
+            chunker_type: ChunkerType::Markdown,
+        };
+        let markdown = "# Code Example\n\n```python\nprint('hello')\n```\n\nSome text after code.";
+        let result = chunk_text(markdown, &config, None).unwrap();
+        assert!(result.chunk_count >= 1);
+        assert!(result.chunks.iter().any(|chunk| chunk.content.contains("```")));
+    }
+
+    #[test]
+    fn test_chunk_markdown_with_links() {
+        let config = ChunkingConfig {
+            max_characters: 80,
+            overlap: 10,
+            trim: true,
+            chunker_type: ChunkerType::Markdown,
+        };
+        let markdown = "Check out [this link](https://example.com) for more info.";
+        let result = chunk_text(markdown, &config, None).unwrap();
+        assert_eq!(result.chunk_count, 1);
+        assert!(result.chunks[0].content.contains("[this link]"));
+    }
+
+    #[test]
+    fn test_chunk_text_with_trim() {
+        let config = ChunkingConfig {
+            max_characters: 30,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "  Leading and trailing spaces  should be trimmed  ";
+        let result = chunk_text(text, &config, None).unwrap();
+        assert!(result.chunk_count >= 1);
+        assert!(result.chunks.iter().all(|chunk| !chunk.content.starts_with(' ')));
+    }
+
+    #[test]
+    fn test_chunk_text_without_trim() {
+        let config = ChunkingConfig {
+            max_characters: 30,
+            overlap: 5,
+            trim: false,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "  Text with spaces  ";
+        let result = chunk_text(text, &config, None).unwrap();
+        assert_eq!(result.chunk_count, 1);
+        assert!(result.chunks[0].content.starts_with(' ') || result.chunks[0].content.len() < text.len());
+    }
+
+    #[test]
+    fn test_chunk_with_invalid_overlap() {
+        let config = ChunkingConfig {
+            max_characters: 10,
+            overlap: 20,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let result = chunk_text("Some text", &config, None);
+        assert!(result.is_err());
+        let err = result.unwrap_err();
+        assert!(matches!(err, KreuzbergError::Validation { .. }));
+    }
+
+    #[test]
+    fn test_chunk_text_with_type_text() {
+        let result = chunk_text_with_type("Simple text", 50, 10, true, ChunkerType::Text).unwrap();
+        assert_eq!(result.chunk_count, 1);
+        assert_eq!(result.chunks[0].content, "Simple text");
+    }
+
+    #[test]
+    fn test_chunk_text_with_type_markdown() {
+        let markdown = "# Header\n\nContent here.";
+        let result = chunk_text_with_type(markdown, 50, 10, true, ChunkerType::Markdown).unwrap();
+        assert_eq!(result.chunk_count, 1);
+        assert!(result.chunks[0].content.contains("# Header"));
+    }
+
+    #[test]
+    fn test_chunk_texts_batch_empty() {
+        let config = ChunkingConfig::default();
+        let texts: Vec<&str> = vec![];
+        let results = chunk_texts_batch(&texts, &config).unwrap();
+        assert_eq!(results.len(), 0);
+    }
+
+    #[test]
+    fn test_chunk_texts_batch_multiple() {
+        let config = ChunkingConfig {
+            max_characters: 30,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let texts = vec!["First text", "Second text", "Third text"];
+        let results = chunk_texts_batch(&texts, &config).unwrap();
+        assert_eq!(results.len(), 3);
+        assert!(results.iter().all(|r| r.chunk_count >= 1));
+    }
+
+    #[test]
+    fn test_chunk_texts_batch_mixed_lengths() {
+        let config = ChunkingConfig {
+            max_characters: 20,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let texts = vec![
+            "Short",
+            "This is a longer text that should be split into multiple chunks",
+            "",
+        ];
+        let results = chunk_texts_batch(&texts, &config).unwrap();
+        assert_eq!(results.len(), 3);
+        assert_eq!(results[0].chunk_count, 1);
+        assert!(results[1].chunk_count > 1);
+        assert_eq!(results[2].chunk_count, 0);
+    }
+
+    #[test]
+    fn test_chunk_texts_batch_error_propagation() {
+        let config = ChunkingConfig {
+            max_characters: 10,
+            overlap: 20,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let texts = vec!["Text one", "Text two"];
+        let result = chunk_texts_batch(&texts, &config);
+        assert!(result.is_err());
+    }
+
+    #[test]
+    fn test_chunking_config_default() {
+        let config = ChunkingConfig::default();
+        assert_eq!(config.max_characters, 2000);
+        assert_eq!(config.overlap, 100);
+        assert!(config.trim);
+        assert_eq!(config.chunker_type, ChunkerType::Text);
+    }
+
+    #[test]
+    fn test_chunk_very_long_text() {
+        let config = ChunkingConfig {
+            max_characters: 100,
+            overlap: 20,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "a".repeat(1000);
+        let result = chunk_text(&text, &config, None).unwrap();
+        assert!(result.chunk_count >= 10);
+        assert!(result.chunks.iter().all(|chunk| chunk.content.len() <= 100));
+    }
+
+    #[test]
+    fn test_chunk_text_with_newlines() {
+        let config = ChunkingConfig {
+            max_characters: 30,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "Line one\nLine two\nLine three\nLine four\nLine five";
+        let result = chunk_text(text, &config, None).unwrap();
+        assert!(result.chunk_count >= 1);
+    }
+
+    #[test]
+    fn test_chunk_markdown_with_lists() {
+        let config = ChunkingConfig {
+            max_characters: 100,
+            overlap: 10,
+            trim: true,
+            chunker_type: ChunkerType::Markdown,
+        };
+        let markdown = "# List Example\n\n- Item 1\n- Item 2\n- Item 3\n\nMore text.";
+        let result = chunk_text(markdown, &config, None).unwrap();
+        assert!(result.chunk_count >= 1);
+        assert!(result.chunks.iter().any(|chunk| chunk.content.contains("- Item")));
+    }
+
+    #[test]
+    fn test_chunk_markdown_with_tables() {
+        let config = ChunkingConfig {
+            max_characters: 150,
+            overlap: 10,
+            trim: true,
+            chunker_type: ChunkerType::Markdown,
+        };
+        let markdown = "# Table\n\n| Col1 | Col2 |\n|------|------|\n| A    | B    |\n| C    | D    |";
+        let result = chunk_text(markdown, &config, None).unwrap();
+        assert!(result.chunk_count >= 1);
+        assert!(result.chunks.iter().any(|chunk| chunk.content.contains("|")));
+    }
+
+    #[test]
+    fn test_chunk_special_characters() {
+        let config = ChunkingConfig {
+            max_characters: 50,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "Special chars: @#$%^&*()[]{}|\\<>?/~`";
+        let result = chunk_text(text, &config, None).unwrap();
+        assert_eq!(result.chunk_count, 1);
+        assert!(result.chunks[0].content.contains("@#$%"));
+    }
+
+    #[test]
+    fn test_chunk_unicode_characters() {
+        let config = ChunkingConfig {
+            max_characters: 50,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "Unicode: 你好世界 🌍 café résumé";
+        let result = chunk_text(text, &config, None).unwrap();
+        assert_eq!(result.chunk_count, 1);
+        assert!(result.chunks[0].content.contains("你好"));
+        assert!(result.chunks[0].content.contains("🌍"));
+    }
+
+    #[test]
+    fn test_chunk_cjk_text() {
+        let config = ChunkingConfig {
+            max_characters: 30,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "日本語のテキストです。これは長い文章で、複数のチャンクに分割されるべきです。";
+        let result = chunk_text(text, &config, None).unwrap();
+        assert!(result.chunk_count >= 1);
+    }
+
+    #[test]
+    fn test_chunk_mixed_languages() {
+        let config = ChunkingConfig {
+            max_characters: 40,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "English text mixed with 中文文本 and some français";
+        let result = chunk_text(text, &config, None).unwrap();
+        assert!(result.chunk_count >= 1);
+    }
+
+    #[test]
+    fn test_chunk_offset_calculation_with_overlap() {
+        let config = ChunkingConfig {
+            max_characters: 20,
+            overlap: 5,
+            trim: false,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "AAAAA BBBBB CCCCC DDDDD EEEEE FFFFF";
+        let result = chunk_text(text, &config, None).unwrap();
+
+        assert!(result.chunks.len() >= 2, "Expected at least 2 chunks");
+
+        for i in 0..result.chunks.len() {
+            let chunk = &result.chunks[i];
+            let metadata = &chunk.metadata;
+
+            assert_eq!(
+                metadata.byte_end - metadata.byte_start,
+                chunk.content.len(),
+                "Chunk {} offset range doesn't match content length",
+                i
+            );
+
+            assert_eq!(metadata.chunk_index, i);
+            assert_eq!(metadata.total_chunks, result.chunks.len());
+        }
+
+        for i in 0..result.chunks.len() - 1 {
+            let current_chunk = &result.chunks[i];
+            let next_chunk = &result.chunks[i + 1];
+
+            assert!(
+                next_chunk.metadata.byte_start < current_chunk.metadata.byte_end,
+                "Chunk {} and {} don't overlap: next starts at {} but current ends at {}",
+                i,
+                i + 1,
+                next_chunk.metadata.byte_start,
+                current_chunk.metadata.byte_end
+            );
+
+            let overlap_size = current_chunk.metadata.byte_end - next_chunk.metadata.byte_start;
+            assert!(
+                overlap_size <= config.overlap + 10,
+                "Overlap between chunks {} and {} is too large: {}",
+                i,
+                i + 1,
+                overlap_size
+            );
+        }
+    }
+
+    #[test]
+    fn test_chunk_offset_calculation_without_overlap() {
+        let config = ChunkingConfig {
+            max_characters: 20,
+            overlap: 0,
+            trim: false,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "AAAAA BBBBB CCCCC DDDDD EEEEE FFFFF";
+        let result = chunk_text(text, &config, None).unwrap();
+
+        for i in 0..result.chunks.len() - 1 {
+            let current_chunk = &result.chunks[i];
+            let next_chunk = &result.chunks[i + 1];
+
+            assert!(
+                next_chunk.metadata.byte_start >= current_chunk.metadata.byte_end,
+                "Chunk {} and {} overlap when they shouldn't: next starts at {} but current ends at {}",
+                i,
+                i + 1,
+                next_chunk.metadata.byte_start,
+                current_chunk.metadata.byte_end
+            );
+        }
+    }
+
+    #[test]
+    fn test_chunk_offset_covers_full_text() {
+        let config = ChunkingConfig {
+            max_characters: 15,
+            overlap: 3,
+            trim: false,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "0123456789 ABCDEFGHIJ KLMNOPQRST UVWXYZ";
+        let result = chunk_text(text, &config, None).unwrap();
+
+        assert!(result.chunks.len() >= 2, "Expected multiple chunks");
+
+        assert_eq!(
+            result.chunks[0].metadata.byte_start, 0,
+            "First chunk should start at position 0"
+        );
+
+        for i in 0..result.chunks.len() - 1 {
+            let current_chunk = &result.chunks[i];
+            let next_chunk = &result.chunks[i + 1];
+
+            assert!(
+                next_chunk.metadata.byte_start <= current_chunk.metadata.byte_end,
+                "Gap detected between chunk {} (ends at {}) and chunk {} (starts at {})",
+                i,
+                current_chunk.metadata.byte_end,
+                i + 1,
+                next_chunk.metadata.byte_start
+            );
+        }
+    }
+
+    #[test]
+    fn test_chunk_offset_with_various_overlap_sizes() {
+        for overlap in [0, 5, 10, 20] {
+            let config = ChunkingConfig {
+                max_characters: 30,
+                overlap,
+                trim: false,
+                chunker_type: ChunkerType::Text,
+            };
+            let text = "Word ".repeat(30);
+            let result = chunk_text(&text, &config, None).unwrap();
+
+            for chunk in &result.chunks {
+                assert!(
+                    chunk.metadata.byte_end > chunk.metadata.byte_start,
+                    "Invalid offset range for overlap {}: start={}, end={}",
+                    overlap,
+                    chunk.metadata.byte_start,
+                    chunk.metadata.byte_end
+                );
+            }
+
+            for chunk in &result.chunks {
+                assert!(
+                    chunk.metadata.byte_start < text.len(),
+                    "char_start with overlap {} is out of bounds: {}",
+                    overlap,
+                    chunk.metadata.byte_start
+                );
+            }
+        }
+    }
+
+    #[test]
+    fn test_chunk_last_chunk_offset() {
+        let config = ChunkingConfig {
+            max_characters: 20,
+            overlap: 5,
+            trim: false,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "AAAAA BBBBB CCCCC DDDDD EEEEE";
+        let result = chunk_text(text, &config, None).unwrap();
+
+        assert!(result.chunks.len() >= 2, "Need multiple chunks for this test");
+
+        let last_chunk = result.chunks.last().unwrap();
+        let second_to_last = &result.chunks[result.chunks.len() - 2];
+
+        assert!(
+            last_chunk.metadata.byte_start < second_to_last.metadata.byte_end,
+            "Last chunk should overlap with previous chunk"
+        );
+
+        let expected_end = text.len();
+        let last_chunk_covers_end =
+            last_chunk.content.trim_end() == text.trim_end() || last_chunk.metadata.byte_end >= expected_end - 5;
+        assert!(last_chunk_covers_end, "Last chunk should cover the end of the text");
+    }
+
+    #[test]
+    fn test_chunk_with_page_boundaries() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 30,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "Page one content here. Page two starts here and continues.";
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 21,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 22,
+                byte_end: 58,
+                page_number: 2,
+            },
+        ];
+
+        let result = chunk_text(text, &config, Some(&boundaries)).unwrap();
+        assert!(result.chunks.len() >= 2);
+
+        assert_eq!(result.chunks[0].metadata.first_page, Some(1));
+
+        let last_chunk = result.chunks.last().unwrap();
+        assert_eq!(last_chunk.metadata.last_page, Some(2));
+    }
+
+    #[test]
+    fn test_chunk_without_page_boundaries() {
+        let config = ChunkingConfig {
+            max_characters: 30,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "This is some test content that should be split into multiple chunks.";
+
+        let result = chunk_text(text, &config, None).unwrap();
+        assert!(result.chunks.len() >= 2);
+
+        for chunk in &result.chunks {
+            assert_eq!(chunk.metadata.first_page, None);
+            assert_eq!(chunk.metadata.last_page, None);
+        }
+    }
+
+    #[test]
+    fn test_chunk_empty_boundaries() {
+        let config = ChunkingConfig {
+            max_characters: 30,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "Some text content here.";
+        let boundaries: Vec<PageBoundary> = vec![];
+
+        let result = chunk_text(text, &config, Some(&boundaries)).unwrap();
+        assert_eq!(result.chunks.len(), 1);
+
+        assert_eq!(result.chunks[0].metadata.first_page, None);
+        assert_eq!(result.chunks[0].metadata.last_page, None);
+    }
+
+    #[test]
+    fn test_chunk_spanning_multiple_pages() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 50,
+            overlap: 5,
+            trim: false,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "0123456789 AAAAAAAAAA 1111111111 BBBBBBBBBB 2222222222";
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 20,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 20,
+                byte_end: 40,
+                page_number: 2,
+            },
+            PageBoundary {
+                byte_start: 40,
+                byte_end: 54,
+                page_number: 3,
+            },
+        ];
+
+        let result = chunk_text(text, &config, Some(&boundaries)).unwrap();
+        assert!(result.chunks.len() >= 2);
+
+        for chunk in &result.chunks {
+            assert!(chunk.metadata.first_page.is_some() || chunk.metadata.last_page.is_some());
+        }
+    }
+
+    #[test]
+    fn test_chunk_text_with_invalid_boundary_range() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 30,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "Page one content here. Page two content.";
+
+        let boundaries = vec![PageBoundary {
+            byte_start: 10,
+            byte_end: 5,
+            page_number: 1,
+        }];
+
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_err());
+        let err = result.unwrap_err();
+        assert!(err.to_string().contains("Invalid boundary range"));
+        assert!(err.to_string().contains("byte_start"));
+    }
+
+    #[test]
+    fn test_chunk_text_with_unsorted_boundaries() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 30,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "Page one content here. Page two content.";
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 22,
+                byte_end: 40,
+                page_number: 2,
+            },
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 21,
+                page_number: 1,
+            },
+        ];
+
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_err());
+        let err = result.unwrap_err();
+        assert!(err.to_string().contains("not sorted"));
+        assert!(err.to_string().contains("boundaries"));
+    }
+
+    #[test]
+    fn test_chunk_text_with_overlapping_boundaries() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 30,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "Page one content here. Page two content.";
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 25,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 20,
+                byte_end: 40,
+                page_number: 2,
+            },
+        ];
+
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_err());
+        let err = result.unwrap_err();
+        assert!(err.to_string().contains("Overlapping"));
+        assert!(err.to_string().contains("boundaries"));
+    }
+
+    #[test]
+    fn test_calculate_page_range_with_invalid_boundaries() {
+        use crate::types::PageBoundary;
+
+        let boundaries = vec![PageBoundary {
+            byte_start: 15,
+            byte_end: 10,
+            page_number: 1,
+        }];
+
+        let result = calculate_page_range(0, 20, &boundaries);
+        assert!(result.is_err());
+        let err = result.unwrap_err();
+        assert!(err.to_string().contains("Invalid boundary range"));
+    }
+
+    #[test]
+    fn test_validate_page_boundaries_valid() {
+        use crate::types::PageBoundary;
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 20,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 20,
+                byte_end: 40,
+                page_number: 2,
+            },
+            PageBoundary {
+                byte_start: 40,
+                byte_end: 60,
+                page_number: 3,
+            },
+        ];
+
+        let result = chunk_text(
+            "x".repeat(60).as_str(),
+            &ChunkingConfig {
+                max_characters: 30,
+                overlap: 5,
+                trim: false,
+                chunker_type: ChunkerType::Text,
+            },
+            Some(&boundaries),
+        );
+        assert!(result.is_ok());
+    }
+
+    #[test]
+    fn test_validate_page_boundaries_empty() {
+        let boundaries: Vec<PageBoundary> = vec![];
+        let result = chunk_text(
+            "Some test text",
+            &ChunkingConfig {
+                max_characters: 30,
+                overlap: 5,
+                trim: true,
+                chunker_type: ChunkerType::Text,
+            },
+            Some(&boundaries),
+        );
+        assert!(result.is_ok());
+    }
+
+    #[test]
+    fn test_page_boundaries_with_gaps() {
+        use crate::types::PageBoundary;
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 10,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 15,
+                byte_end: 25,
+                page_number: 2,
+            },
+        ];
+
+        let text = "0123456789XXXXX0123456789";
+        let result = chunk_text(
+            text,
+            &ChunkingConfig {
+                max_characters: 30,
+                overlap: 5,
+                trim: false,
+                chunker_type: ChunkerType::Text,
+            },
+            Some(&boundaries),
+        );
+        assert!(result.is_ok());
+    }
+
+    #[test]
+    fn test_chunk_with_same_start_and_end() {
+        use crate::types::PageBoundary;
+
+        let boundaries = vec![PageBoundary {
+            byte_start: 10,
+            byte_end: 10,
+            page_number: 1,
+        }];
+
+        let result = chunk_text(
+            "test content here",
+            &ChunkingConfig {
+                max_characters: 30,
+                overlap: 5,
+                trim: true,
+                chunker_type: ChunkerType::Text,
+            },
+            Some(&boundaries),
+        );
+        assert!(result.is_err());
+        let err = result.unwrap_err();
+        assert!(err.to_string().contains("Invalid boundary range"));
+    }
+
+    #[test]
+    fn test_multiple_overlapping_errors() {
+        use crate::types::PageBoundary;
+
+        let text = "This is a longer test content string that spans more bytes";
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 20,
+                byte_end: 40,
+                page_number: 2,
+            },
+            PageBoundary {
+                byte_start: 10,
+                byte_end: 35,
+                page_number: 1,
+            },
+        ];
+
+        let result = chunk_text(
+            text,
+            &ChunkingConfig {
+                max_characters: 30,
+                overlap: 5,
+                trim: true,
+                chunker_type: ChunkerType::Text,
+            },
+            Some(&boundaries),
+        );
+        assert!(result.is_err());
+        assert!(result.unwrap_err().to_string().contains("not sorted"));
+    }
+
+    #[test]
+    fn test_chunk_with_pages_basic() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 25,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "First page content here.Second page content here.Third page.";
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 24,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 24,
+                byte_end: 50,
+                page_number: 2,
+            },
+            PageBoundary {
+                byte_start: 50,
+                byte_end: 60,
+                page_number: 3,
+            },
+        ];
+
+        let result = chunk_text(text, &config, Some(&boundaries)).unwrap();
+
+        if !result.chunks.is_empty() {
+            assert!(result.chunks[0].metadata.first_page.is_some());
+        }
+    }
+
+    #[test]
+    fn test_chunk_with_pages_single_page_chunk() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 100,
+            overlap: 10,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "All content on single page fits in one chunk.";
+
+        let boundaries = vec![PageBoundary {
+            byte_start: 0,
+            byte_end: 45,
+            page_number: 1,
+        }];
+
+        let result = chunk_text(text, &config, Some(&boundaries)).unwrap();
+        assert_eq!(result.chunks.len(), 1);
+        assert_eq!(result.chunks[0].metadata.first_page, Some(1));
+        assert_eq!(result.chunks[0].metadata.last_page, Some(1));
+    }
+
+    #[test]
+    fn test_chunk_with_pages_no_overlap() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 20,
+            overlap: 0,
+            trim: false,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "AAAAA BBBBB CCCCC DDDDD";
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 11,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 11,
+                byte_end: 23,
+                page_number: 2,
+            },
+        ];
+
+        let result = chunk_text(text, &config, Some(&boundaries)).unwrap();
+        assert!(!result.chunks.is_empty());
+
+        for chunk in &result.chunks {
+            if let (Some(first), Some(last)) = (chunk.metadata.first_page, chunk.metadata.last_page) {
+                assert!(first <= last);
+            }
+        }
+    }
+
+    #[test]
+    fn test_calculate_page_range_within_page() {
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 100,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 100,
+                byte_end: 200,
+                page_number: 2,
+            },
+        ];
+
+        let (first, last) = calculate_page_range(10, 50, &boundaries).unwrap();
+        assert_eq!(first, Some(1));
+        assert_eq!(last, Some(1));
+    }
+
+    #[test]
+    fn test_calculate_page_range_spanning_pages() {
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 100,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 100,
+                byte_end: 200,
+                page_number: 2,
+            },
+        ];
+
+        let (first, last) = calculate_page_range(50, 150, &boundaries).unwrap();
+        assert_eq!(first, Some(1));
+        assert_eq!(last, Some(2));
+    }
+
+    #[test]
+    fn test_calculate_page_range_empty_boundaries() {
+        let boundaries: Vec<PageBoundary> = vec![];
+
+        let (first, last) = calculate_page_range(0, 50, &boundaries).unwrap();
+        assert_eq!(first, None);
+        assert_eq!(last, None);
+    }
+
+    #[test]
+    fn test_calculate_page_range_no_overlap() {
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 100,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 100,
+                byte_end: 200,
+                page_number: 2,
+            },
+        ];
+
+        let (first, last) = calculate_page_range(200, 250, &boundaries).unwrap();
+        assert_eq!(first, None);
+        assert_eq!(last, None);
+    }
+
+    #[test]
+    fn test_calculate_page_range_three_pages() {
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 100,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 100,
+                byte_end: 200,
+                page_number: 2,
+            },
+            PageBoundary {
+                byte_start: 200,
+                byte_end: 300,
+                page_number: 3,
+            },
+        ];
+
+        let (first, last) = calculate_page_range(50, 250, &boundaries).unwrap();
+        assert_eq!(first, Some(1));
+        assert_eq!(last, Some(3));
+    }
+
+    #[test]
+    fn test_chunk_metadata_page_range_accuracy() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 30,
+            overlap: 5,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "Page One Content Here.Page Two.";
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 21,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 21,
+                byte_end: 31,
+                page_number: 2,
+            },
+        ];
+
+        let result = chunk_text(text, &config, Some(&boundaries)).unwrap();
+
+        for chunk in &result.chunks {
+            assert_eq!(chunk.metadata.byte_end - chunk.metadata.byte_start, chunk.content.len());
+        }
+    }
+
+    #[test]
+    fn test_chunk_page_range_boundary_edge_cases() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 10,
+            overlap: 2,
+            trim: false,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "0123456789ABCDEFGHIJ";
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 10,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 10,
+                byte_end: 20,
+                page_number: 2,
+            },
+        ];
+
+        let result = chunk_text(text, &config, Some(&boundaries)).unwrap();
+
+        for chunk in &result.chunks {
+            let on_page1 = chunk.metadata.byte_start < 10;
+            let on_page2 = chunk.metadata.byte_end > 10;
+
+            if on_page1 && on_page2 {
+                assert_eq!(chunk.metadata.first_page, Some(1));
+                assert_eq!(chunk.metadata.last_page, Some(2));
+            } else if on_page1 {
+                assert_eq!(chunk.metadata.first_page, Some(1));
+            } else if on_page2 {
+                assert_eq!(chunk.metadata.first_page, Some(2));
+            }
+        }
+    }
+
+    #[test]
+    fn test_validate_utf8_boundaries_valid_ascii() {
+        use crate::types::PageBoundary;
+
+        let text = "This is ASCII text.";
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 10,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 10,
+                byte_end: 19,
+                page_number: 2,
+            },
+        ];
+
+        let result = chunk_text(text, &ChunkingConfig::default(), Some(&boundaries));
+        assert!(result.is_ok());
+    }
+
+    #[test]
+    fn test_validate_utf8_boundaries_valid_emoji() {
+        use crate::types::PageBoundary;
+
+        let text = "Hello 👋 World 🌍 End";
+        let config = ChunkingConfig::default();
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 11,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 11,
+                byte_end: 25,
+                page_number: 2,
+            },
+        ];
+
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_ok());
+    }
+
+    #[test]
+    fn test_validate_utf8_boundaries_valid_cjk() {
+        use crate::types::PageBoundary;
+
+        let text = "你好世界 こんにちは 안녕하세요";
+        let config = ChunkingConfig::default();
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 13,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 13,
+                byte_end: 44,
+                page_number: 2,
+            },
+        ];
+
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_ok());
+    }
+
+    #[test]
+    fn test_validate_utf8_boundaries_invalid_mid_emoji() {
+        use crate::types::PageBoundary;
+
+        let text = "Hello 👋 World";
+        let boundaries = vec![PageBoundary {
+            byte_start: 0,
+            byte_end: 7,
+            page_number: 1,
+        }];
+
+        let config = ChunkingConfig::default();
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_err());
+        let err = result.unwrap_err();
+        assert!(err.to_string().contains("UTF-8 character boundary"));
+        assert!(err.to_string().contains("byte_end=7"));
+    }
+
+    #[test]
+    fn test_validate_utf8_boundaries_invalid_mid_multibyte_cjk() {
+        use crate::types::PageBoundary;
+
+        let text = "中文文本";
+        let boundaries = vec![PageBoundary {
+            byte_start: 0,
+            byte_end: 1,
+            page_number: 1,
+        }];
+
+        let config = ChunkingConfig::default();
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_err());
+        let err = result.unwrap_err();
+        assert!(err.to_string().contains("UTF-8 character boundary"));
+    }
+
+    #[test]
+    fn test_validate_utf8_boundaries_byte_start_exceeds_length() {
+        use crate::types::PageBoundary;
+
+        let text = "Short";
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 3,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 10,
+                byte_end: 15,
+                page_number: 2,
+            },
+        ];
+
+        let config = ChunkingConfig::default();
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_err());
+        let err = result.unwrap_err();
+        assert!(err.to_string().contains("exceeds text length"));
+    }
+
+    #[test]
+    fn test_validate_utf8_boundaries_byte_end_exceeds_length() {
+        use crate::types::PageBoundary;
+
+        let text = "Short";
+        let boundaries = vec![PageBoundary {
+            byte_start: 0,
+            byte_end: 100,
+            page_number: 1,
+        }];
+
+        let config = ChunkingConfig::default();
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_err());
+        let err = result.unwrap_err();
+        assert!(err.to_string().contains("exceeds text length"));
+    }
+
+    #[test]
+    fn test_validate_utf8_boundaries_empty_boundaries() {
+        use crate::types::PageBoundary;
+
+        let text = "Some text";
+        let boundaries: Vec<PageBoundary> = vec![];
+
+        let config = ChunkingConfig::default();
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_ok());
+    }
+
+    #[test]
+    fn test_validate_utf8_boundaries_at_text_boundaries() {
+        use crate::types::PageBoundary;
+
+        let text = "Exact boundary test";
+        let text_len = text.len();
+        let boundaries = vec![PageBoundary {
+            byte_start: 0,
+            byte_end: text_len,
+            page_number: 1,
+        }];
+
+        let config = ChunkingConfig::default();
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_ok());
+    }
+
+    #[test]
+    fn test_validate_utf8_boundaries_mixed_languages() {
+        use crate::types::PageBoundary;
+
+        let text = "English text mixed with 中文 and français";
+        let config = ChunkingConfig::default();
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 24,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 24,
+                byte_end: text.len(),
+                page_number: 2,
+            },
+        ];
+
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_ok());
+    }
+
+    #[test]
+    fn test_chunk_text_rejects_invalid_utf8_boundaries() {
+        use crate::types::PageBoundary;
+
+        let text = "🌍🌎🌏 Three emoji planets";
+        let config = ChunkingConfig::default();
+
+        let boundaries = vec![PageBoundary {
+            byte_start: 0,
+            byte_end: 1000,
+            page_number: 1,
+        }];
+
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_err());
+    }
+
+    #[test]
+    fn test_validate_utf8_boundaries_combining_diacriticals() {
+        use crate::types::PageBoundary;
+
+        let text = "café";
+        let config = ChunkingConfig::default();
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 2,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 2,
+                byte_end: text.len(),
+                page_number: 2,
+            },
+        ];
+
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_ok());
+    }
+
+    #[test]
+    fn test_validate_utf8_boundaries_error_messages_are_clear() {
+        use crate::types::PageBoundary;
+
+        let text = "Test 👋 text";
+        let config = ChunkingConfig::default();
+
+        let boundaries = vec![PageBoundary {
+            byte_start: 0,
+            byte_end: 6,
+            page_number: 1,
+        }];
+
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_err());
+        let err = result.unwrap_err();
+        let err_msg = err.to_string();
+        assert!(err_msg.contains("UTF-8"));
+        assert!(err_msg.contains("boundary"));
+        assert!(err_msg.contains("6"));
+    }
+
+    #[test]
+    fn test_validate_utf8_boundaries_multiple_valid_boundaries() {
+        use crate::types::PageBoundary;
+
+        let text = "First👋Second🌍Third";
+        let config = ChunkingConfig::default();
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 5,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 5,
+                byte_end: 9,
+                page_number: 2,
+            },
+            PageBoundary {
+                byte_start: 9,
+                byte_end: 15,
+                page_number: 3,
+            },
+            PageBoundary {
+                byte_start: 15,
+                byte_end: 19,
+                page_number: 4,
+            },
+            PageBoundary {
+                byte_start: 19,
+                byte_end: text.len(),
+                page_number: 5,
+            },
+        ];
+
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_ok());
+    }
+
+    #[test]
+    fn test_validate_utf8_boundaries_zero_start_and_end() {
+        use crate::types::PageBoundary;
+
+        let text = "Text";
+        let config = ChunkingConfig::default();
+
+        let boundaries = vec![PageBoundary {
+            byte_start: 0,
+            byte_end: 0,
+            page_number: 1,
+        }];
+
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_err());
+    }
+
+    #[test]
+    fn test_utf8_boundaries_caching_with_many_boundaries() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 500,
+            overlap: 50,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+
+        let text = "🌍 Hello World ".repeat(200);
+        let text_len = text.len();
+
+        let mut boundaries = vec![];
+        let boundary_count = 10;
+        let step = text_len / boundary_count;
+
+        for i in 0..boundary_count {
+            let start = i * step;
+            let end = if i == boundary_count - 1 {
+                text_len
+            } else {
+                (i + 1) * step
+            };
+
+            if start < end
+                && start <= text_len
+                && end <= text_len
+                && let Some(boundary_start) = text[..start].char_indices().last().map(|(idx, _)| idx)
+                && let Some(boundary_end) = text[..end].char_indices().last().map(|(idx, _)| idx)
+            {
+                boundaries.push(PageBoundary {
+                    byte_start: boundary_start,
+                    byte_end: boundary_end,
+                    page_number: i + 1,
+                });
+            }
+        }
+
+        if !boundaries.is_empty() {
+            let result = chunk_text(&text, &config, Some(&boundaries));
+            assert!(
+                result.is_ok(),
+                "Failed to chunk text with {} boundaries",
+                boundaries.len()
+            );
+
+            let chunks = result.unwrap();
+            assert!(chunks.chunk_count > 0);
+        }
+    }
+
+    #[test]
+    fn test_utf8_boundaries_caching_large_document_with_emojis() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 1000,
+            overlap: 100,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+
+        let large_text = "This is a large document with lots of emoji: 🌍 🚀 💻 🎉 🔥 ✨ 🎨 🌟 ".repeat(100);
+
+        let all_indices: Vec<usize> = large_text.char_indices().map(|(idx, _)| idx).collect();
+
+        let third_idx = all_indices.len() / 3;
+        let two_thirds_idx = (2 * all_indices.len()) / 3;
+
+        let boundary_start_1 = if third_idx < all_indices.len() {
+            all_indices[third_idx]
+        } else {
+            large_text.len()
+        };
+
+        let boundary_start_2 = if two_thirds_idx < all_indices.len() {
+            all_indices[two_thirds_idx]
+        } else {
+            large_text.len()
+        };
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: boundary_start_1,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: boundary_start_1,
+                byte_end: boundary_start_2,
+                page_number: 2,
+            },
+            PageBoundary {
+                byte_start: boundary_start_2,
+                byte_end: large_text.len(),
+                page_number: 3,
+            },
+        ];
+
+        let result = chunk_text(&large_text, &config, Some(&boundaries));
+        assert!(result.is_ok());
+
+        let chunks = result.unwrap();
+        assert!(!chunks.chunks.is_empty());
+
+        for chunk in &chunks.chunks {
+            assert!(!chunk.content.is_empty());
+            if let (Some(first), Some(last)) = (chunk.metadata.first_page, chunk.metadata.last_page) {
+                assert!(first <= last);
+            }
+        }
+    }
+
+    #[test]
+    fn test_adaptive_validation_small_boundary_set() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 100,
+            overlap: 10,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "Hello 👋 World 🌍 End";
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 6,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 6,
+                byte_end: 15,
+                page_number: 2,
+            },
+            PageBoundary {
+                byte_start: 15,
+                byte_end: text.len(),
+                page_number: 3,
+            },
+        ];
+
+        let result = chunk_text(text, &config, Some(&boundaries));
+        assert!(result.is_ok());
+    }
+
+    #[test]
+    fn test_adaptive_validation_threshold_boundary() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 200,
+            overlap: 20,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "Test text ".repeat(50);
+        let text_len = text.len();
+
+        let mut boundaries = vec![];
+        let step = text_len / ADAPTIVE_VALIDATION_THRESHOLD;
+
+        for i in 0..ADAPTIVE_VALIDATION_THRESHOLD {
+            let start = i * step;
+            let end = if i == ADAPTIVE_VALIDATION_THRESHOLD - 1 {
+                text_len
+            } else {
+                (i + 1) * step
+            };
+
+            if start < end
+                && start <= text_len
+                && end <= text_len
+                && let Some(boundary_start) = text[..start.min(text_len - 1)]
+                    .char_indices()
+                    .last()
+                    .map(|(idx, _)| idx)
+                && let Some(boundary_end) = text[..end.min(text_len)].char_indices().last().map(|(idx, _)| idx)
+                && boundary_start < boundary_end
+            {
+                boundaries.push(PageBoundary {
+                    byte_start: boundary_start,
+                    byte_end: boundary_end,
+                    page_number: i + 1,
+                });
+            }
+        }
+
+        if !boundaries.is_empty() {
+            let result = chunk_text(&text, &config, Some(&boundaries));
+            assert!(result.is_ok());
+        }
+    }
+
+    #[test]
+    fn test_adaptive_validation_large_boundary_set() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 500,
+            overlap: 50,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "Lorem ipsum dolor sit amet ".repeat(100);
+        let text_len = text.len();
+
+        let mut boundaries = vec![];
+        let boundary_count = 50;
+        let step = text_len / boundary_count;
+
+        for i in 0..boundary_count {
+            let start = i * step;
+            let end = if i == boundary_count - 1 {
+                text_len
+            } else {
+                (i + 1) * step
+            };
+
+            if start < end
+                && start <= text_len
+                && end <= text_len
+                && let Some(boundary_start) = text[..start.min(text_len - 1)]
+                    .char_indices()
+                    .last()
+                    .map(|(idx, _)| idx)
+                && let Some(boundary_end) = text[..end.min(text_len)].char_indices().last().map(|(idx, _)| idx)
+                && boundary_start < boundary_end
+            {
+                boundaries.push(PageBoundary {
+                    byte_start: boundary_start,
+                    byte_end: boundary_end,
+                    page_number: i + 1,
+                });
+            }
+        }
+
+        if !boundaries.is_empty() {
+            let result = chunk_text(&text, &config, Some(&boundaries));
+            assert!(result.is_ok());
+        }
+    }
+
+    #[test]
+    fn test_adaptive_validation_consistency() {
+        use crate::types::PageBoundary;
+
+        let config = ChunkingConfig {
+            max_characters: 300,
+            overlap: 30,
+            trim: true,
+            chunker_type: ChunkerType::Text,
+        };
+        let text = "Mixed language: 你好 مرحبا Здравствуй ".repeat(50);
+
+        let boundaries = vec![
+            PageBoundary {
+                byte_start: 0,
+                byte_end: 50,
+                page_number: 1,
+            },
+            PageBoundary {
+                byte_start: 50,
+                byte_end: 100,
+                page_number: 2,
+            },
+            PageBoundary {
+                byte_start: 100,
+                byte_end: 150,
+                page_number: 3,
+            },
+            PageBoundary {
+                byte_start: 150,
+                byte_end: 200,
+                page_number: 4,
+            },
+            PageBoundary {
+                byte_start: 200,
+                byte_end: text.len(),
+                page_number: 5,
+            },
+        ];
+
+        let result = chunk_text(&text, &config, Some(&boundaries));
+        let _ = result;
+    }
+}