cxpak 1.0.1

Spends CPU cycles so you don't spend tokens. The LLM gets a briefing packet instead of a flashlight in a dark room.
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

use serde::{Deserialize, Serialize};
use std::collections::HashMap;
use std::path::Path;

/// Flat-matrix vector index. Vectors are stored contiguously:
/// `vectors[i * dims .. (i + 1) * dims]` is the embedding for `paths[i]`.
#[derive(Debug, Serialize, Deserialize)]
pub struct EmbeddingIndex {
    paths: Vec<String>,
    path_index: HashMap<String, usize>,
    vectors: Vec<f32>,
    dims: usize,
}

impl EmbeddingIndex {
    /// Create a new empty index with the given dimensionality.
    pub fn new(dims: usize) -> Self {
        Self {
            paths: Vec::new(),
            path_index: HashMap::new(),
            vectors: Vec::new(),
            dims,
        }
    }

    /// Add or update the vector for a file path.
    ///
    /// If the path already exists its embedding is replaced in-place.
    /// Panics if `vector.len() != self.dims`.
    pub fn add(&mut self, path: impl Into<String>, vector: Vec<f32>) {
        assert_eq!(
            vector.len(),
            self.dims,
            "vector length {} does not match index dims {}",
            vector.len(),
            self.dims
        );

        let path = path.into();

        if let Some(&idx) = self.path_index.get(&path) {
            // Update existing entry in-place.
            let start = idx * self.dims;
            self.vectors[start..start + self.dims].copy_from_slice(&vector);
        } else {
            let idx = self.paths.len();
            self.path_index.insert(path.clone(), idx);
            self.paths.push(path);
            self.vectors.extend_from_slice(&vector);
        }
    }

    /// Search for the `top_k` most similar paths to `query_vector`.
    ///
    /// Returns `(path, cosine_similarity)` pairs sorted descending by similarity.
    pub fn search(&self, query: &[f32], top_k: usize) -> Vec<(String, f64)> {
        if self.paths.is_empty() || top_k == 0 {
            return vec![];
        }

        let query_norm = l2_norm(query);
        if query_norm == 0.0 {
            return vec![];
        }

        let mut scores: Vec<(usize, f64)> = self
            .paths
            .iter()
            .enumerate()
            .map(|(i, _)| {
                let v = &self.vectors[i * self.dims..(i + 1) * self.dims];
                let sim = cosine_sim_unnormed(query, v, query_norm);
                (i, sim)
            })
            .collect();

        scores.sort_unstable_by(|a, b| b.1.partial_cmp(&a.1).unwrap_or(std::cmp::Ordering::Equal));
        scores.truncate(top_k);

        scores
            .into_iter()
            .map(|(i, sim)| (self.paths[i].clone(), sim))
            .collect()
    }

    /// Number of entries in the index.
    pub fn len(&self) -> usize {
        self.paths.len()
    }

    /// Returns true if the index contains no entries.
    pub fn is_empty(&self) -> bool {
        self.paths.is_empty()
    }

    /// Serialize to `path` using bincode 1.x.
    pub fn save(&self, path: &Path) -> Result<(), String> {
        let bytes =
            bincode::serialize(self).map_err(|e| format!("bincode serialize error: {e}"))?;
        std::fs::write(path, bytes).map_err(|e| format!("write error: {e}"))
    }

    /// Deserialize from `path` using bincode 1.x.
    pub fn load(path: &Path) -> Result<Self, String> {
        let bytes = std::fs::read(path).map_err(|e| format!("read error: {e}"))?;
        bincode::deserialize(&bytes).map_err(|e| format!("bincode deserialize error: {e}"))
    }

    /// Cosine similarity of the stored vector for `path` against `query`.
    ///
    /// Returns `None` if the path is not in the index or dimensions mismatch.
    pub fn cosine_similarity(&self, path: &str, query: &[f32]) -> Option<f64> {
        let &idx = self.path_index.get(path)?;
        let v = &self.vectors[idx * self.dims..(idx + 1) * self.dims];
        let qn = l2_norm(query);
        if qn == 0.0 {
            return Some(0.0);
        }
        Some(cosine_sim_unnormed(query, v, qn))
    }
}

// ---------------------------------------------------------------------------
// Internal helpers
// ---------------------------------------------------------------------------

fn l2_norm(v: &[f32]) -> f64 {
    let sum: f32 = v.iter().map(|x| x * x).sum();
    (sum as f64).sqrt()
}

/// Cosine similarity where caller supplies the pre-computed query norm.
fn cosine_sim_unnormed(a: &[f32], b: &[f32], a_norm: f64) -> f64 {
    let dot: f32 = a.iter().zip(b.iter()).map(|(x, y)| x * y).sum();
    let b_norm = l2_norm(b);
    if b_norm == 0.0 || a_norm == 0.0 {
        return 0.0;
    }
    (dot as f64) / (a_norm * b_norm)
}

#[cfg(test)]
mod tests {
    use super::*;
    use tempfile::TempDir;

    fn unit_vec(dims: usize, val: f32) -> Vec<f32> {
        let mut v = vec![0.0f32; dims];
        v[0] = val;
        v
    }

    #[test]
    fn add_and_search() {
        let mut idx = EmbeddingIndex::new(4);
        idx.add("a.rs", vec![1.0, 0.0, 0.0, 0.0]);
        idx.add("b.rs", vec![0.0, 1.0, 0.0, 0.0]);
        idx.add("c.rs", vec![0.0, 0.0, 1.0, 0.0]);

        // Query exactly matches "a.rs"
        let results = idx.search(&[1.0, 0.0, 0.0, 0.0], 2);
        assert_eq!(results.len(), 2);
        assert_eq!(results[0].0, "a.rs");
        assert!((results[0].1 - 1.0).abs() < 1e-6);
        // second result is one of the orthogonal ones with sim 0.0
        assert!((results[1].1).abs() < 1e-6);
    }

    #[test]
    fn cosine_similarity() {
        let mut idx = EmbeddingIndex::new(3);
        idx.add("file.rs", vec![1.0, 0.0, 0.0]);

        let sim = idx.cosine_similarity("file.rs", &[1.0, 0.0, 0.0]).unwrap();
        assert!((sim - 1.0).abs() < 1e-6, "identical vectors => sim=1.0");

        let sim2 = idx.cosine_similarity("file.rs", &[0.0, 1.0, 0.0]).unwrap();
        assert!((sim2).abs() < 1e-6, "orthogonal vectors => sim=0.0");

        assert!(idx
            .cosine_similarity("missing.rs", &[1.0, 0.0, 0.0])
            .is_none());
    }

    #[test]
    fn incremental_update() {
        let mut idx = EmbeddingIndex::new(2);
        idx.add("x.rs", vec![1.0, 0.0]);
        assert_eq!(idx.len(), 1);

        // Update same path — should replace, not add.
        idx.add("x.rs", vec![0.0, 1.0]);
        assert_eq!(idx.len(), 1, "update should not increase count");

        // The stored vector should now be [0.0, 1.0].
        let sim = idx.cosine_similarity("x.rs", &[0.0, 1.0]).unwrap();
        assert!((sim - 1.0).abs() < 1e-6, "updated vector should be used");
    }

    #[test]
    fn save_load() {
        let dir = TempDir::new().unwrap();
        let path = dir.path().join("index.bin");

        let mut idx = EmbeddingIndex::new(3);
        idx.add("alpha.rs", vec![0.5, 0.5, 0.0]);
        idx.add("beta.rs", vec![0.0, 0.5, 0.5]);

        idx.save(&path).expect("save should succeed");
        let loaded = EmbeddingIndex::load(&path).expect("load should succeed");

        assert_eq!(loaded.len(), 2);
        let sim = loaded
            .cosine_similarity("alpha.rs", &[0.5, 0.5, 0.0])
            .unwrap();
        assert!(sim > 0.99, "loaded similarity: {sim}");
    }

    #[test]
    fn empty_search() {
        let idx = EmbeddingIndex::new(4);
        let results = idx.search(&[1.0, 0.0, 0.0, 0.0], 5);
        assert!(results.is_empty(), "empty index should return no results");
    }

    #[test]
    fn search_top_k_respected() {
        let mut idx = EmbeddingIndex::new(2);
        for i in 0..10 {
            idx.add(format!("file{i}.rs"), unit_vec(2, i as f32 * 0.1 + 0.1));
        }
        let results = idx.search(&[1.0, 0.0], 3);
        assert_eq!(results.len(), 3);
        // Results should be sorted descending
        assert!(results[0].1 >= results[1].1);
        assert!(results[1].1 >= results[2].1);
    }
}