docling-rag 0.47.3

Pluggable RAG subsystem for docling.rs: chunking, embeddings, vector store, and semantic search.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145

//! In-memory vector store. Never persisted; used by tests and quick evals.

use super::{top_k_by_cosine, VectorStore};
use crate::model::{Chunk, Document, Scored};
use crate::Result;
use async_trait::async_trait;
use std::sync::RwLock;

/// A store that keeps everything in process memory.
#[derive(Default)]
pub struct MemoryStore {
    docs: RwLock<Vec<Document>>,
    chunks: RwLock<Vec<Chunk>>,
}

impl MemoryStore {
    /// Create an empty in-memory store.
    pub fn new() -> Self {
        MemoryStore::default()
    }
}

#[async_trait]
impl VectorStore for MemoryStore {
    async fn migrate(&self) -> Result<()> {
        Ok(())
    }

    async fn upsert_document(&self, doc: &Document) -> Result<()> {
        let mut docs = self.docs.write().unwrap();
        if let Some(existing) = docs.iter_mut().find(|d| d.id == doc.id) {
            *existing = doc.clone();
        } else {
            docs.push(doc.clone());
        }
        Ok(())
    }

    async fn find_document_by_hash(&self, hash: &str) -> Result<Option<String>> {
        Ok(self
            .docs
            .read()
            .unwrap()
            .iter()
            .find(|d| d.hash == hash)
            .map(|d| d.id.clone()))
    }

    async fn insert_chunks(&self, chunks: &[Chunk]) -> Result<()> {
        self.chunks.write().unwrap().extend_from_slice(chunks);
        Ok(())
    }

    async fn vector_search(&self, query: &[f32], k: usize) -> Result<Vec<Scored>> {
        let chunks = self.chunks.read().unwrap();
        let candidates = chunks.iter().filter_map(|c| {
            c.embedding.as_ref().map(|e| {
                let mut bare = c.clone();
                bare.embedding = None;
                (bare, e.clone())
            })
        });
        Ok(top_k_by_cosine(query, candidates.collect::<Vec<_>>(), k))
    }

    async fn all_chunks(&self) -> Result<Vec<Chunk>> {
        Ok(self
            .chunks
            .read()
            .unwrap()
            .iter()
            .map(|c| {
                let mut bare = c.clone();
                bare.embedding = None;
                bare
            })
            .collect())
    }

    async fn count_chunks(&self) -> Result<usize> {
        Ok(self.chunks.read().unwrap().len())
    }

    async fn count_chunks_for(&self, doc_id: &str) -> Result<usize> {
        Ok(self
            .chunks
            .read()
            .unwrap()
            .iter()
            .filter(|c| c.doc_id == doc_id)
            .count())
    }

    async fn chunk_neighborhood(&self, doc_id: &str, ordinal: i64) -> Result<Vec<Chunk>> {
        let mut out: Vec<Chunk> = self
            .chunks
            .read()
            .unwrap()
            .iter()
            .filter(|c| c.doc_id == doc_id && (c.ordinal - ordinal).abs() <= 1)
            .cloned()
            .collect();
        out.sort_by_key(|c| c.ordinal);
        Ok(out)
    }

    async fn count_documents(&self) -> Result<usize> {
        Ok(self.docs.read().unwrap().len())
    }

    async fn list_documents(&self) -> Result<Vec<Document>> {
        Ok(self.docs.read().unwrap().clone())
    }

    async fn delete_document(&self, doc_id: &str) -> Result<()> {
        self.docs.write().unwrap().retain(|d| d.id != doc_id);
        self.chunks.write().unwrap().retain(|c| c.doc_id != doc_id);
        Ok(())
    }

    async fn delete_documents_by_source(&self, source_uri: &str) -> Result<()> {
        let ids: Vec<String> = {
            let docs = self.docs.read().unwrap();
            docs.iter()
                .filter(|d| d.source_uri == source_uri)
                .map(|d| d.id.clone())
                .collect()
        };
        self.docs
            .write()
            .unwrap()
            .retain(|d| d.source_uri != source_uri);
        self.chunks
            .write()
            .unwrap()
            .retain(|c| !ids.contains(&c.doc_id));
        Ok(())
    }

    async fn clear(&self) -> Result<()> {
        self.docs.write().unwrap().clear();
        self.chunks.write().unwrap().clear();
        Ok(())
    }
}