vipune 0.2.2

A minimal memory layer for AI agents
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
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278

//! Tests for search operations.

use crate::memory::MemoryStore;
use crate::sqlite::Database;

#[test]
fn test_search_basic() {
    use tempfile::TempDir;
    let dir = TempDir::new().unwrap();
    let path = dir.path().join("test.db");
    std::mem::forget(dir);

    let db = Database::open(&path).unwrap();
    let embedding_match = vec![1.0f32; 384];
    let embedding_other = vec![0.0f32; 384];

    let id_match = db
        .insert("test-project", "matching content", &embedding_match, None)
        .unwrap();
    db.insert("test-project", "other content", &embedding_other, None)
        .unwrap();

    let results = db.search("test-project", &embedding_match, 5).unwrap();
    assert!(!results.is_empty());
    assert_eq!(results[0].id, id_match);
    assert!(results[0].similarity.unwrap() > 0.9);
}

#[test]
fn test_search_sorting() {
    use tempfile::TempDir;
    let dir = TempDir::new().unwrap();
    let path = dir.path().join("test.db");
    std::mem::forget(dir);

    let db = Database::open(&path).unwrap();
    let embedding_high = vec![1.0f32; 384];
    let embedding_medium = vec![0.5f32; 384];
    let embedding_low = vec![0.0f32; 384];

    db.insert("test-project", "low", &embedding_low, None)
        .unwrap();
    db.insert("test-project", "high", &embedding_high, None)
        .unwrap();
    db.insert("test-project", "medium", &embedding_medium, None)
        .unwrap();

    let results = db.search("test-project", &embedding_high, 10).unwrap();
    assert!(results[0].similarity.unwrap() >= results[1].similarity.unwrap());
    assert!(results[1].similarity.unwrap() >= results[2].similarity.unwrap());
}

#[test]
fn test_negative_similarity() {
    use tempfile::TempDir;
    let dir = TempDir::new().unwrap();
    let path = dir.path().join("test.db");
    std::mem::forget(dir);

    let db = Database::open(&path).unwrap();
    let embedding_pos = vec![1.0f32; 384];
    let embedding_neg = vec![-1.0f32; 384];

    db.insert("test-project", "positive", &embedding_pos, None)
        .unwrap();

    let results = db.search("test-project", &embedding_neg, 10).unwrap();
    assert!(!results.is_empty());
    assert!(results[0].similarity.unwrap() < 0.0);
}

#[test]
fn test_hybrid_search_basic() {
    use tempfile::TempDir;
    let dir = TempDir::new().unwrap();
    let path = dir.path().join("test.db");
    std::mem::forget(dir);

    let db = Database::open(&path).unwrap();
    db.initialize_fts().unwrap();

    // Add memories with different embeddings for semantic search
    let embedding_rust: Vec<f32> = vec![1.0, 0.0, 0.0f32]
        .into_iter()
        .cycle()
        .take(384)
        .collect();
    let embedding_python: Vec<f32> = vec![0.0, 1.0, 0.0f32]
        .into_iter()
        .cycle()
        .take(384)
        .collect();
    let embedding_general: Vec<f32> = vec![0.5, 0.5, 0.5f32]
        .into_iter()
        .cycle()
        .take(384)
        .collect();

    let id_rust = db
        .insert(
            "test-project",
            "rust programming language",
            &embedding_rust,
            None,
        )
        .unwrap();
    let _id_python = db
        .insert(
            "test-project",
            "python code examples",
            &embedding_python,
            None,
        )
        .unwrap();
    let _id_general = db
        .insert(
            "test-project",
            "general software development",
            &embedding_general,
            None,
        )
        .unwrap();

    // Test hybrid search combines semantic and BM25
    let semantic_results = db.search("test-project", &embedding_rust, 50).unwrap();
    let bm25_results = db.search_bm25("rust", "test-project", 50).unwrap();

    // Verify semantic search finds rust-related content
    assert!(!semantic_results.is_empty());
    assert!(semantic_results.iter().any(|m| m.id == id_rust));
    assert!(semantic_results.iter().any(|m| m.content.contains("rust")));

    // BM25 should find matches for "rust" query
    assert!(!bm25_results.is_empty());

    // Rust memory should appear in both semantic and BM25 results
    assert!(semantic_results.iter().any(|m| m.id == id_rust));
    assert!(
        bm25_results
            .iter()
            .any(|m| m.content.to_lowercase().contains("rust"))
    );
}

#[test]
fn test_hybrid_search_empty_results() {
    use tempfile::TempDir;
    let dir = TempDir::new().unwrap();
    let path = dir.path().join("test.db");
    std::mem::forget(dir);

    let db = Database::open(&path).unwrap();
    db.initialize_fts().unwrap();

    // Add one memory
    let embedding = vec![0.5f32; 384];
    db.insert("test-project", "some content", &embedding, None)
        .unwrap();

    // Query with non-matching text
    let semantic_results = db.search("test-project", &vec![0.1f32; 384], 50).unwrap();
    let bm25_results = db
        .search_bm25("nonexistent term xyz", "test-project", 50)
        .unwrap();

    // One may be empty, but fusion should handle it
    assert!(!semantic_results.is_empty() || !bm25_results.is_empty());
}

#[test]
fn test_hybrid_search_with_recency() {
    use chrono::{Duration, Utc};
    use tempfile::TempDir;
    let dir = TempDir::new().unwrap();
    let path = dir.path().join("test.db");
    std::mem::forget(dir);

    let db = Database::open(&path).unwrap();
    db.initialize_fts().unwrap();

    // Add memories with known embedding
    let embedding_good = vec![1.0f32; 384];
    let embedding_bad = vec![0.0f32; 384];

    let now = Utc::now();
    let old_time = (now - Duration::days(100)).to_rfc3339();
    let new_time = now.to_rfc3339();

    // Old high-similarity memory
    let id_old = db
        .insert_with_time(
            "test-project",
            "old but relevant",
            &embedding_good,
            None,
            &old_time,
            &old_time,
        )
        .unwrap();

    // New low-similarity memory (same content for BM25 relevance)
    let _id_new = db
        .insert_with_time(
            "test-project",
            "new but less relevant",
            &embedding_bad,
            None,
            &new_time,
            &new_time,
        )
        .unwrap();

    // Search without recency weighting - should find high-similarity first
    let semantic_results = db.search("test-project", &embedding_good, 10).unwrap();
    if !semantic_results.is_empty() {
        let top_id = &semantic_results[0].id;
        // With zero recency weight, similarity dominates
        // Either the high-similarity old memory is first, or we have a single result
        assert!(top_id == &id_old || semantic_results.len() == 1);
    }
}

#[ignore]
#[test]
fn test_integration_add_search_roundtrip() {
    // Full integration test with real model
    // Requires: cargo test -- --ignored
    use crate::config::Config;
    use tempfile::TempDir;
    let dir = TempDir::new().unwrap();
    let path = dir.path().join("test.db");
    let config = Config::default();

    let mut store = MemoryStore::new(&path, "BAAI/bge-small-en-v1.5", config).unwrap();

    let id = match store
        .add_with_conflict("test-project", "semantic search is useful", None, false)
        .unwrap()
    {
        crate::memory_types::AddResult::Added { id } => id,
        _ => panic!("Expected AddResult::Added"),
    };

    let results = store
        .search("test-project", "finding information", 5, 0.0)
        .unwrap();
    assert!(!results.is_empty());

    let memory = store.get(&id).unwrap().unwrap();
    assert_eq!(memory.content, "semantic search is useful");
}

#[ignore]
#[test]
fn test_integration_update_changes_embedding() {
    // Full integration test with real model
    // Requires: cargo test -- --ignored
    use crate::config::Config;
    use tempfile::TempDir;
    let dir = TempDir::new().unwrap();
    let path = dir.path().join("test.db");
    let config = Config::default();

    let mut store = MemoryStore::new(&path, "BAAI/bge-small-en-v1.5", config).unwrap();

    let id = match store
        .add_with_conflict("test-project", "original content", None, false)
        .unwrap()
    {
        crate::memory_types::AddResult::Added { id } => id,
        _ => panic!("Expected AddResult::Added"),
    };

    store.update(&id, "completely different content").unwrap();

    let memory = store.get(&id).unwrap().unwrap();
    assert_eq!(memory.content, "completely different content");
}