mermaid-cli 0.3.10

Open-source AI pair programmer with agentic capabilities. Local-first with Ollama, native tool calling, and beautiful TUI.
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

use anyhow::Result;

/// Token counting utility using character-based estimation
/// Uses ~4 characters per token as a reasonable approximation
pub struct Tokenizer {
    model_name: String,
}

impl Tokenizer {
    /// Create a new tokenizer for the given model
    pub fn new(model_name: &str) -> Self {
        Self {
            model_name: model_name.to_string(),
        }
    }

    /// Count tokens in a single text string (~4 chars per token)
    pub fn count_tokens(&self, text: &str) -> Result<usize> {
        Ok((text.len() + 3) / 4)
    }

    /// Count tokens in a chat message format
    pub fn count_chat_tokens(&self, messages: &[(String, String)]) -> Result<usize> {
        let total_chars: usize = messages
            .iter()
            .map(|(role, content)| role.len() + content.len() + 4) // +4 for message overhead
            .sum();
        Ok((total_chars + 3) / 4)
    }

    /// Get the maximum tokens for a model
    pub fn get_max_tokens(&self) -> usize {
        let model_name = self.get_base_model_name();

        if model_name.contains("gpt-4o") {
            128000
        } else if model_name.contains("gpt-4-turbo") || model_name.contains("gpt-4-1106") {
            128000
        } else if model_name.contains("gpt-4-32k") {
            32768
        } else if model_name.contains("gpt-4") {
            8192
        } else if model_name.contains("gpt-3.5-turbo-16k") {
            16384
        } else if model_name.contains("gpt-3.5-turbo") {
            4096
        } else if model_name.contains("claude-3") {
            200000
        } else if model_name.contains("claude") {
            100000
        } else if model_name.contains("llama-3") {
            8192
        } else if model_name.contains("llama-2") {
            4096
        } else if model_name.contains("codellama") {
            16384
        } else if model_name.contains("deepseek-coder") {
            65536
        } else if model_name.contains("qwen") {
            32768
        } else if model_name.contains("mistral") || model_name.contains("mixtral") {
            32768
        } else {
            8192 // Conservative default
        }
    }

    /// Calculate remaining tokens in context window
    pub fn remaining_tokens(&self, used_tokens: usize) -> usize {
        let max_tokens = self.get_max_tokens();
        max_tokens.saturating_sub(used_tokens)
    }

    /// Get the base model name (strip provider prefix)
    fn get_base_model_name(&self) -> String {
        if let Some(idx) = self.model_name.find('/') {
            self.model_name[idx + 1..].to_string()
        } else {
            self.model_name.clone()
        }
    }
}

/// Count tokens in file contents (~4 chars per token)
pub fn count_file_tokens(content: &str, _model_name: &str) -> usize {
    (content.len() + 3) / 4
}

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

    #[test]
    fn test_token_counting() {
        let tokenizer = Tokenizer::new("gpt-3.5-turbo");
        let text = "Hello, world! This is a test message.";
        let count = tokenizer.count_tokens(text).unwrap();
        assert!(count > 0);
        assert!(count < text.len());
    }

    #[test]
    fn test_model_name_extraction() {
        let tokenizer = Tokenizer::new("ollama/gpt-4");
        assert_eq!(tokenizer.get_base_model_name(), "gpt-4");

        let tokenizer = Tokenizer::new("unknown-model");
        assert_eq!(tokenizer.get_base_model_name(), "unknown-model");
    }

    #[test]
    fn test_max_tokens() {
        let tokenizer = Tokenizer::new("gpt-4");
        assert_eq!(tokenizer.get_max_tokens(), 8192);

        let tokenizer = Tokenizer::new("gpt-4o");
        assert_eq!(tokenizer.get_max_tokens(), 128000);

        let tokenizer = Tokenizer::new("gpt-3.5-turbo");
        assert_eq!(tokenizer.get_max_tokens(), 4096);
    }
}