kibble 0.1.0

chew through any source into clean datasets — a fast ingestion, RAG & fine-tuning toolkit
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

//! Retrieval-index corpus loading + chunking. Loads documents from the
//! configured sources (or an explicit path), then chunks them for embedding
//! and BM25 indexing.

use crate::config::IndexConfig;
use serde::{Deserialize, Serialize};
use std::path::Path;

/// A source document before chunking.
#[derive(Clone)]
#[allow(dead_code)]
pub struct IndexDoc {
    pub doc_id: String,
    pub source: String,
    pub title: String,
    pub text: String,
}

/// One indexed passage.
#[derive(Serialize, Deserialize, Clone, Debug)]
pub struct Chunk {
    pub chunk_id: String,
    pub doc_id: String,
    pub source: String,
    pub title: String,
    pub text: String,
}

#[allow(dead_code)]
const TEXT_EXTS: &[&str] = &["txt", "md", "markdown", "json", "jsonl"];

/// Path relative to `repo_root`, using forward slashes; falls back to the
/// full path string if it isn't under `repo_root`.
#[allow(dead_code)]
fn rel(path: &Path, repo_root: &Path) -> String {
    path.strip_prefix(repo_root).unwrap_or(path).to_string_lossy().replace('\\', "/")
}

/// Recursively collect text-like files under `root` as whole-file documents.
#[allow(dead_code)]
fn walk_text_files(root: &Path, repo_root: &Path, out: &mut Vec<IndexDoc>) {
    let Ok(entries) = std::fs::read_dir(root) else { return };
    for entry in entries.flatten() {
        let path = entry.path();
        if path.is_dir() {
            walk_text_files(&path, repo_root, out);
        } else if let Some(ext) = path.extension().and_then(|e| e.to_str()) {
            if !TEXT_EXTS.contains(&ext.to_lowercase().as_str()) {
                continue;
            }
            match std::fs::read_to_string(&path) {
                Ok(text) if !text.trim().is_empty() => {
                    let source = rel(&path, repo_root);
                    let title = path.file_stem().and_then(|s| s.to_str()).unwrap_or("").to_string();
                    out.push(IndexDoc { doc_id: source.clone(), source, title, text });
                }
                Ok(_) => {}
                Err(e) => eprintln!("kibble: skipping {} ({e})", path.display()),
            }
        }
    }
}

/// Load documents from an explicit path, or from the configured sources.
/// The `data/raw` source uses `corpus::load_raw_docs` (richer titles/ids);
/// everything else uses the generic text-file walker.
pub fn load_docs(repo_root: &Path, cfg: &IndexConfig, explicit: Option<&Path>) -> Vec<IndexDoc> {
    let mut docs = Vec::new();
    if let Some(p) = explicit {
        if p.is_dir() {
            walk_text_files(p, repo_root, &mut docs);
        } else if p.is_file() {
            let mut one = Vec::new();
            walk_text_files(p.parent().unwrap_or(repo_root), repo_root, &mut one);
            let target = rel(p, repo_root);
            docs.extend(one.into_iter().filter(|d| d.source == target));
        }
        return docs;
    }
    for src in &cfg.sources {
        let path = repo_root.join(src);
        if !path.exists() {
            continue;
        }
        if std::path::Path::new(src).file_name().and_then(|s| s.to_str()) == Some("raw") {
            if let Ok(raw) = crate::corpus::load_raw_docs(&path) {
                for d in raw {
                    docs.push(IndexDoc { doc_id: d.doc_id.clone(), source: d.doc_id.clone(), title: d.title, text: d.text });
                }
            }
        } else {
            walk_text_files(&path, repo_root, &mut docs);
        }
    }
    docs
}

/// Chunk every document; `chunk_id` = `"{source}#{n}"`.
pub fn chunk_docs(docs: &[IndexDoc], chunk_chars: usize) -> Vec<Chunk> {
    let mut chunks = Vec::new();
    for d in docs {
        for (n, text) in crate::text::chunk_text(&d.text, chunk_chars).into_iter().enumerate() {
            chunks.push(Chunk {
                chunk_id: format!("{}#{}", d.source, n),
                doc_id: d.doc_id.clone(),
                source: d.source.clone(),
                title: d.title.clone(),
                text,
            });
        }
    }
    chunks
}

/// Build the retrieval index from the configured sources (or an explicit path).
/// Embeds each chunk via the shared vector cache when `[understand.embed]` is
/// configured; otherwise builds a lexical-only index. Writes `chunks.jsonl`,
/// `bm25.json`, and `meta.json` under `[index].dir`. Returns the chunk count.
pub async fn build_index(repo_root: &Path, explicit: Option<&Path>) -> std::io::Result<usize> {
    let cfg = crate::config::load_config(&repo_root.join(crate::config::CONFIG_FILE));
    let docs = load_docs(repo_root, &cfg.index, explicit);
    let chunks = chunk_docs(&docs, cfg.index.chunk_chars);

    // Embed chunk texts (fail-soft → lexical-only).
    let embed = &cfg.understand.embed;
    let mut has_vectors = false;
    let mut dim = 0usize;
    if chunks.is_empty() {
        eprintln!("kibble: no documents found to index");
    } else if embed.base_url.is_empty() {
        eprintln!("kibble: no embed backend configured — building lexical-only index");
    } else {
        let texts: Vec<String> = chunks.iter().map(|c| c.text.clone()).collect();
        let store = repo_root.join(&embed.store);
        match crate::embed::EndpointEmbedder::new(embed, cfg.network.proxy.as_deref()) {
            Ok(e) => match crate::vectors::get_or_embed(&e, &store, &embed.model, &texts, embed.batch_size).await {
                Ok(vecs) => {
                    has_vectors = true;
                    dim = vecs.first().map(|v| v.len()).unwrap_or(0);
                }
                Err(err) => eprintln!("kibble: index embeddings skipped (embed failed): {err}"),
            },
            Err(err) => eprintln!("kibble: index embeddings skipped (embed init failed): {err}"),
        }
    }

    // BM25.
    let chunk_tokens: Vec<Vec<String>> = chunks.iter().map(|c| crate::bm25::tokenize(&c.text)).collect();
    let bm25 = crate::bm25::Bm25Index::build(&chunk_tokens);

    // Persist.
    let dir = repo_root.join(&cfg.index.dir);
    std::fs::create_dir_all(&dir)?;
    let mut buf = String::new();
    for c in &chunks {
        buf.push_str(&serde_json::to_string(c)?);
        buf.push('\n');
    }
    std::fs::write(dir.join("chunks.jsonl"), buf)?;
    std::fs::write(dir.join("bm25.json"), serde_json::to_string(&bm25)?)?;
    let meta = serde_json::json!({
        "embed_model": embed.model,
        "dim": dim,
        "n_chunks": chunks.len(),
        "has_vectors": has_vectors,
    });
    std::fs::write(dir.join("meta.json"), serde_json::to_string(&meta)?)?;
    Ok(chunks.len())
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::config::IndexConfig;

    #[test]
    fn load_and_chunk_a_dir() {
        let dir = std::env::temp_dir().join(format!("kibble_idx_load_{}", std::process::id()));
        let _ = std::fs::remove_dir_all(&dir);
        std::fs::create_dir_all(dir.join("data/notes")).unwrap();
        std::fs::write(dir.join("data/notes/a.md"), "hello world one two three").unwrap();
        std::fs::write(dir.join("data/notes/b.txt"), "another document here").unwrap();
        std::fs::write(dir.join("data/notes/skip.png"), "not text").unwrap();

        let cfg = IndexConfig { sources: vec!["data/notes".into()], chunk_chars: 1000, ..Default::default() };
        let docs = load_docs(&dir, &cfg, None);
        assert_eq!(docs.len(), 2); // .png skipped
        let mut sources: Vec<&str> = docs.iter().map(|d| d.source.as_str()).collect();
        sources.sort();
        assert_eq!(sources, vec!["data/notes/a.md", "data/notes/b.txt"]);

        let chunks = chunk_docs(&docs, 1000);
        assert_eq!(chunks.len(), 2);
        assert!(chunks.iter().any(|c| c.chunk_id == "data/notes/a.md#0"));
        std::fs::remove_dir_all(&dir).ok();
    }

    #[test]
    fn explicit_path_overrides_sources() {
        let dir = std::env::temp_dir().join(format!("kibble_idx_explicit_{}", std::process::id()));
        let _ = std::fs::remove_dir_all(&dir);
        std::fs::create_dir_all(dir.join("other")).unwrap();
        std::fs::write(dir.join("other/x.md"), "explicit content").unwrap();
        let cfg = IndexConfig { sources: vec!["data/raw".into()], ..Default::default() };
        let docs = load_docs(&dir, &cfg, Some(&dir.join("other")));
        assert_eq!(docs.len(), 1);
        assert_eq!(docs[0].source, "other/x.md");
        std::fs::remove_dir_all(&dir).ok();
    }

    #[test]
    fn raw_source_gives_unique_chunk_ids_and_provenance() {
        let dir = std::env::temp_dir().join(format!("kibble_idx_raw_{}", std::process::id()));
        let _ = std::fs::remove_dir_all(&dir);
        std::fs::create_dir_all(dir.join("data/raw/hacking")).unwrap();
        std::fs::write(dir.join("data/raw/hacking/alpha.txt"), "first raw document about alpha").unwrap();
        std::fs::write(dir.join("data/raw/hacking/beta.txt"), "second raw document about beta").unwrap();
        let cfg = crate::config::IndexConfig { sources: vec!["data/raw".into()], chunk_chars: 1000, ..Default::default() };
        let docs = load_docs(&dir, &cfg, None);
        assert_eq!(docs.len(), 2);
        let chunks = chunk_docs(&docs, 1000);
        // distinct chunk_ids and sources per document (no "data/raw#0" collision)
        let ids: std::collections::HashSet<&str> = chunks.iter().map(|c| c.chunk_id.as_str()).collect();
        assert_eq!(ids.len(), chunks.len());
        let sources: std::collections::HashSet<&str> = chunks.iter().map(|c| c.source.as_str()).collect();
        assert_eq!(sources.len(), 2);
        assert!(chunks.iter().all(|c| c.source != "data/raw"));
        std::fs::remove_dir_all(&dir).ok();
    }

    #[tokio::test]
    async fn build_index_lexical_only_writes_files() {
        let dir = std::env::temp_dir().join(format!("kibble_idx_build_{}", std::process::id()));
        let _ = std::fs::remove_dir_all(&dir);
        std::fs::create_dir_all(dir.join("data/notes")).unwrap();
        std::fs::write(dir.join("data/notes/a.md"), "the quick brown fox jumps").unwrap();
        // kibble.toml with sources=data/notes and NO embed backend → lexical-only
        std::fs::write(
            dir.join(crate::config::CONFIG_FILE),
            "[index]\nsources=[\"data/notes\"]\nchunk_chars=1000\n[understand.embed]\nbase_url=\"\"\n",
        ).unwrap();

        let n = build_index(&dir, None).await.unwrap();
        assert!(n >= 1);
        let idx_dir = dir.join("data/index");
        assert!(idx_dir.join("chunks.jsonl").is_file());
        assert!(idx_dir.join("bm25.json").is_file());
        let meta: serde_json::Value =
            serde_json::from_str(&std::fs::read_to_string(idx_dir.join("meta.json")).unwrap()).unwrap();
        assert_eq!(meta["has_vectors"], serde_json::json!(false));
        assert_eq!(meta["n_chunks"], serde_json::json!(n));
        std::fs::remove_dir_all(&dir).ok();
    }
}