xz-rag 0.1.1

Multi-channel Retrieval-Augmented Generation engine
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

use async_trait::async_trait;
use std::sync::Arc;

use crate::error::RagError;
use crate::pipeline::channel::{ChannelConfig, ChannelType};
use crate::types::chunk::ChunkMetadata;
use crate::types::retrieval::{RetrievedChunk, StructuredFilter};

/// Trait for metadata-based search.
/// Returns: Vec of (chunk_id, score, metadata, content, document_id)
#[async_trait]
pub trait MetadataStore: Send + Sync {
    /// Search by metadata filters and query text.
    async fn search_by_metadata(
        &self,
        query: &str,
        filters: &[StructuredFilter],
        top_k: usize,
        namespace: Option<&str>,
    ) -> Result<Vec<(String, f32, ChunkMetadata, String, String)>, RagError>;
}

/// Metadata channel executor using structured filters.
pub struct MetadataChannelExecutor {
    store: Arc<dyn MetadataStore>,
}

impl MetadataChannelExecutor {
    /// Create a new metadata channel executor with the given store.
    pub fn new(store: Arc<dyn MetadataStore>) -> Self {
        Self { store }
    }

    /// Execute metadata-based search using structured filters.
    pub async fn execute(
        &self,
        query: &str,
        config: &ChannelConfig,
        global_filters: &[StructuredFilter],
        namespace: Option<&str>,
    ) -> Result<Vec<RetrievedChunk>, RagError> {
        let results = self
            .store
            .search_by_metadata(query, global_filters, config.top_k, namespace)
            .await
            .map_err(|e| RagError::Store(e.to_string()))?;

        let hits: Vec<RetrievedChunk> =
            results
                .into_iter()
                .filter(|(_, score, _, _, _)| {
                    if let Some(min_score) = config.min_score { *score >= min_score } else { true }
                })
                .map(|(id, score, metadata, content, document_id)| RetrievedChunk {
                    chunk_id: id,
                    document_id,
                    content,
                    score,
                    channel: ChannelType::Metadata.as_str().to_string(),
                    channel_score: score,
                    metadata,
                    embedding: None,
                })
                .collect();

        Ok(hits)
    }
}