sgr-agent 0.5.1

SGR LLM client + agent framework — structured output, function calling, agent loop, 3 agent variants
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

//! Model metadata — context window sizes and compaction budgets.
//!
//! Maps model name → context window. Used by any agent doing compaction
//! to determine when summarization is needed.

/// Estimate context window size for a model by name.
///
/// Recognizes major model families: Gemini, GPT-5, Claude, DeepSeek.
/// Returns token count. Falls back to 128K for unknown models.
pub fn context_window(model_id: &str) -> usize {
    let id = model_id.to_lowercase();
    if id.contains("gemini") {
        1_000_000
    } else if id.contains("gpt-5") && id.contains("mini") {
        400_000
    } else if id.contains("gpt-5") {
        1_050_000
    } else if id.contains("claude") {
        200_000
    } else if id.contains("deepseek") {
        128_000
    } else {
        // GPT-4o, Mistral, Llama, etc.
        128_000
    }
}

/// Suggested compaction threshold (60% of context window).
///
/// Delays compaction as long as possible since it costs an LLM call.
/// For a 400K model this means ~240K tokens.
pub fn compaction_budget(model_id: &str) -> usize {
    context_window(model_id) * 6 / 10
}

/// Strip provider namespace: `"openai/gpt-5.4"` → `"gpt-5.4"`.
///
/// Handles common prefixes: `openai/`, `google/`, `anthropic/`.
pub fn strip_provider(model_id: &str) -> &str {
    for prefix in ["openai/", "google/", "anthropic/"] {
        if let Some(stripped) = model_id.strip_prefix(prefix) {
            return stripped;
        }
    }
    model_id
}

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

    #[test]
    fn gemini_context() {
        assert_eq!(context_window("gemini-2.5-flash"), 1_000_000);
        assert_eq!(context_window("gemini-2.5-pro"), 1_000_000);
    }

    #[test]
    fn gpt5_context() {
        assert_eq!(context_window("gpt-5.4-mini"), 400_000);
        assert_eq!(context_window("gpt-5.4"), 1_050_000);
        assert_eq!(context_window("gpt-5.4-pro"), 1_050_000);
    }

    #[test]
    fn claude_context() {
        assert_eq!(context_window("claude-sonnet-4-5"), 200_000);
        assert_eq!(context_window("claude-opus-4-6"), 200_000);
    }

    #[test]
    fn deepseek_context() {
        assert_eq!(context_window("deepseek-chat"), 128_000);
    }

    #[test]
    fn fallback_context() {
        assert_eq!(context_window("gpt-4o"), 128_000);
        assert_eq!(context_window("unknown-model"), 128_000);
    }

    #[test]
    fn budget_is_60_percent() {
        assert_eq!(compaction_budget("gemini-2.5-flash"), 600_000);
        assert_eq!(compaction_budget("gpt-5.4-mini"), 240_000);
        assert_eq!(compaction_budget("claude-opus-4-6"), 120_000);
        assert_eq!(compaction_budget("gpt-4o"), 76_800);
    }

    #[test]
    fn strip_provider_openai() {
        assert_eq!(strip_provider("openai/gpt-5.4-mini"), "gpt-5.4-mini");
        assert_eq!(strip_provider("openai/gpt-5.4"), "gpt-5.4");
    }

    #[test]
    fn strip_provider_google() {
        assert_eq!(
            strip_provider("google/gemini-2.5-flash"),
            "gemini-2.5-flash"
        );
    }

    #[test]
    fn strip_provider_anthropic() {
        assert_eq!(
            strip_provider("anthropic/claude-opus-4-6"),
            "claude-opus-4-6"
        );
    }

    #[test]
    fn strip_provider_noop() {
        assert_eq!(strip_provider("gpt-5.4-mini"), "gpt-5.4-mini");
        assert_eq!(strip_provider("claude-opus-4-6"), "claude-opus-4-6");
    }
}