llmstat 0.1.0

Token usage distribution and cost across local LLM CLIs
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

//! LiteLLM's community pricebook as the base pricing source.
//!
//! `model_prices_and_context_window.json` is fetched once and cached under
//! `~/.cache/llmstat/` for `TTL`; a stale cache is used when the fetch fails
//! (offline), and an absent cache+failed fetch just leaves models unpriced.

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

use super::{Price, normalize};

const URL: &str =
    "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json";
const TTL: std::time::Duration = std::time::Duration::from_secs(24 * 3600);

#[derive(Deserialize)]
struct LiteEntry {
    input_cost_per_token: Option<f64>,
    output_cost_per_token: Option<f64>,
    cache_read_input_token_cost: Option<f64>,
}

/// LiteLLM prices keyed by normalized model key.
pub struct LiteBook {
    map: HashMap<String, LiteEntry>,
    /// e.g. "litellm (cache)" or "unavailable" — shown in notes.
    pub note: String,
}

fn cache_path() -> PathBuf {
    std::env::home_dir()
        .unwrap_or_else(|| PathBuf::from("~"))
        .join(".cache/llmstat/litellm-prices.json")
}

fn fresh(path: &Path) -> bool {
    std::fs::metadata(path)
        .and_then(|m| m.modified())
        .ok()
        .and_then(|t| t.elapsed().ok())
        .is_some_and(|age| age < TTL)
}

fn fetch() -> Option<String> {
    let resp = ureq::get(URL).call().ok()?;
    resp.into_body().read_to_string().ok()
}

fn parse(text: &str) -> HashMap<String, LiteEntry> {
    let raw: HashMap<String, serde_json::Value> = match serde_json::from_str(text) {
        Ok(r) => r,
        Err(_) => return HashMap::new(),
    };
    raw.into_iter()
        .filter(|(k, _)| k != "sample_spec" && !k.starts_with("spec."))
        .filter_map(|(k, v)| serde_json::from_value::<LiteEntry>(v).ok().map(|e| (k, e)))
        .filter(|(_, e)| e.input_cost_per_token.is_some() || e.output_cost_per_token.is_some())
        .map(|(k, e)| (normalize(&k), e))
        .collect()
}

impl LiteBook {
    /// Load the pricebook: fresh cache → fetch+cache → stale cache → empty.
    pub fn load(refresh: bool) -> Self {
        let path = cache_path();
        let cached = std::fs::read_to_string(&path).ok();
        let cache_fresh = cached.is_some() && fresh(&path);

        if (!cache_fresh || refresh)
            && let Some(text) = fetch()
        {
            if let Some(dir) = path.parent() {
                let _ = std::fs::create_dir_all(dir);
            }
            let _ = std::fs::write(&path, &text);
            return Self {
                map: parse(&text),
                note: "litellm (fetched)".into(),
            };
        }
        match cached {
            Some(text) => Self {
                map: parse(&text),
                note: if cache_fresh {
                    "litellm (cache)".into()
                } else {
                    "litellm (stale cache — offline?)".into()
                },
            },
            None => Self {
                map: HashMap::new(),
                note: "litellm prices unavailable (offline, no cache)".into(),
            },
        }
    }

    /// Look up a normalized model name, returning the litellm key matched and
    /// the per-1M-token price. Exact match first, then the longest key that
    /// prefixes the model name or vice versa (dated aliases etc.).
    pub fn lookup(&self, normalized_name: &str) -> Option<(String, Price)> {
        let to_price = |e: &LiteEntry| Price {
            input: e.input_cost_per_token.unwrap_or(0.0) * 1e6,
            // no cache-read price listed -> cached bills at input rate
            cached: e
                .cache_read_input_token_cost
                .or(e.input_cost_per_token)
                .unwrap_or(0.0)
                * 1e6,
            output: e.output_cost_per_token.unwrap_or(0.0) * 1e6,
        };
        if let Some(e) = self.map.get(normalized_name) {
            return Some((normalized_name.to_string(), to_price(e)));
        }
        // prefix matching in either direction; keys <5 chars are too generic
        let mut best: Option<(&String, &LiteEntry)> = None;
        for (k, e) in &self.map {
            if k.len() < 5 {
                continue;
            }
            // prefix only counts at a '-' boundary: "gpt-5-6-sol" may match
            // key "gpt-5-6" but "gpt-4o" must not match key "gpt-4".
            let pref = |a: &str, b: &str| {
                a.len() > b.len() && a.starts_with(b) && a.as_bytes()[b.len()] == b'-'
            };
            let hit = normalized_name == k || pref(normalized_name, k) || pref(k, normalized_name);
            if hit && best.is_none_or(|(bk, _)| k.len() > bk.len()) {
                best = Some((k, e));
            }
        }
        best.map(|(k, e)| (k.clone(), to_price(e)))
    }
}