tt-retrieval 0.1.0

RAG / context compression: chunking, embedding, HNSW retrieval, <retrievable> tag substitution.
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

//! Data shapes shared across the crate.

use serde::{Deserialize, Serialize};
use uuid::Uuid;

#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct Corpus {
    pub org_id: Uuid,
    pub name: String,
    pub description: Option<String>,
    pub created_at: String, // ISO-8601
}

#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct Document {
    pub doc_id: Uuid,
    pub corpus: String,
    pub org_id: Uuid,
    pub source_path: String,
    pub bytes_indexed: u64,
    pub chunks: u32,
}

#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct Chunk {
    pub id: Uuid,
    pub org_id: Uuid,
    pub corpus: String,
    pub doc_id: Uuid,
    pub chunk_idx: u32,
    pub text: String,
    pub embedding: Vec<f32>, // 1536-dim
    pub embedding_model: String,
    pub metadata: serde_json::Value,
}

#[derive(Debug, Clone, Serialize)]
pub struct RetrievalResult {
    pub chunk_id: Uuid,
    pub doc_id: Uuid,
    pub chunk_idx: u32,
    pub text: String,
    pub similarity: f32,
}

#[derive(Debug, Clone, PartialEq)]
pub struct RetrievableTag {
    pub corpus: String,
    pub k: u32,
    /// Per-tag similarity floor override. `None` means use the default from the
    /// substitution caller (see `substitute::DEFAULT_MIN_SIMILARITY`).
    pub min_similarity: Option<f32>,
    /// Span in the original message text (start_byte_idx, end_byte_idx_exclusive).
    pub span: (usize, usize),
}