skippy-server 0.76.1

Embedded Skippy staged runtime server
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

use crate::cli::OpenAiGuardrailsCliMode;
use openai_frontend::CompactingOpenAiBackend;
use openai_frontend::CompactionConfig;
use openai_frontend::GuardedOpenAiBackend;
use openai_frontend::GuardrailMode;
use openai_frontend::GuardrailPolicy;
use openai_frontend::GuardrailPolicyHandle;
use openai_frontend::OpenAiBackend;
use openai_frontend::RetryExhaustionMode;
use openai_frontend::StreamingGuardrailMode;
use serde::Serialize;
use std::sync::Arc;

#[derive(Clone, Copy, Debug, Eq, PartialEq)]
pub enum OpenAiGuardrailsTarget {
    Skippy,
}

impl OpenAiGuardrailsTarget {
    const fn as_status_label(self) -> &'static str {
        match self {
            Self::Skippy => "skippy",
        }
    }
}

#[derive(Clone, Debug, PartialEq)]
pub struct OpenAiGuardrailsConfig {
    pub target: OpenAiGuardrailsTarget,
    pub policy: GuardrailPolicyHandle,
    pub compaction: Option<CompactionConfig>,
}

impl OpenAiGuardrailsConfig {
    pub fn disabled_for_skippy() -> Self {
        Self {
            target: OpenAiGuardrailsTarget::Skippy,
            policy: GuardrailPolicyHandle::default(),
            compaction: None,
        }
    }

    pub fn compatibility_for_skippy() -> Self {
        Self {
            target: OpenAiGuardrailsTarget::Skippy,
            policy: GuardrailPolicy {
                mode: GuardrailMode::MetricsOnly,
                apply_to_all_models: true,
                retry_exhaustion_mode: RetryExhaustionMode::PassLastText,
                ..GuardrailPolicy::default()
            }
            .into(),
            compaction: None,
        }
    }

    pub fn for_standalone_mode(mode: OpenAiGuardrailsCliMode) -> Self {
        match mode {
            OpenAiGuardrailsCliMode::Disabled => Self::disabled_for_skippy(),
            OpenAiGuardrailsCliMode::Metrics => Self::compatibility_for_skippy(),
            OpenAiGuardrailsCliMode::Enforce => Self {
                target: OpenAiGuardrailsTarget::Skippy,
                policy: GuardrailPolicy {
                    mode: GuardrailMode::Enforce,
                    apply_to_all_models: true,
                    ..GuardrailPolicy::default()
                }
                .into(),
                compaction: None,
            },
        }
    }

    pub fn status(&self) -> OpenAiGuardrailsStatus {
        let policy = self.policy.snapshot();
        OpenAiGuardrailsStatus {
            mode: guardrail_mode_label(policy.mode),
            target: self.target.as_status_label(),
            streaming: streaming_mode_label(policy.streaming_mode),
            retry_exhaustion: retry_exhaustion_label(&policy),
            small_model_policy: small_model_policy_label(&policy),
            small_param_threshold_b: policy.small_param_threshold_b,
            max_tool_retries: policy.max_tool_retries,
            max_structured_retries: policy.max_structured_retries,
        }
    }

    fn should_wrap_guardrail_backend(&self) -> bool {
        matches!(self.target, OpenAiGuardrailsTarget::Skippy)
    }

    #[cfg(test)]
    pub(super) fn wrap_backend(&self, backend: Arc<dyn OpenAiBackend>) -> Arc<dyn OpenAiBackend> {
        self.wrap_backend_with_context_limit(backend, None)
    }

    pub(super) fn wrap_backend_with_context_limit(
        &self,
        backend: Arc<dyn OpenAiBackend>,
        context_limit_tokens: Option<usize>,
    ) -> Arc<dyn OpenAiBackend> {
        let backend = self.wrap_compacting_backend(backend, context_limit_tokens);
        if self.should_wrap_guardrail_backend() {
            Arc::new(GuardedOpenAiBackend::with_policy_handle(
                backend,
                self.policy.clone(),
            ))
        } else {
            backend
        }
    }

    fn wrap_compacting_backend(
        &self,
        backend: Arc<dyn OpenAiBackend>,
        context_limit_tokens: Option<usize>,
    ) -> Arc<dyn OpenAiBackend> {
        let Some(mut compaction) = self.compaction else {
            return backend;
        };
        if compaction.context_limit_tokens.is_none() {
            compaction.context_limit_tokens = context_limit_tokens;
        }
        Arc::new(CompactingOpenAiBackend::new(backend, compaction))
    }
}

#[derive(Clone, Debug, PartialEq, Serialize)]
pub struct OpenAiGuardrailsStatus {
    pub mode: &'static str,
    pub target: &'static str,
    pub streaming: &'static str,
    pub retry_exhaustion: &'static str,
    pub small_model_policy: &'static str,
    pub small_param_threshold_b: f32,
    pub max_tool_retries: u8,
    pub max_structured_retries: u8,
}

fn guardrail_mode_label(mode: GuardrailMode) -> &'static str {
    match mode {
        GuardrailMode::Disabled => "disabled",
        GuardrailMode::MetricsOnly => "metrics",
        GuardrailMode::Enforce => "enforce",
    }
}

fn streaming_mode_label(mode: StreamingGuardrailMode) -> &'static str {
    match mode {
        StreamingGuardrailMode::PassThrough => "pass_through",
    }
}

fn retry_exhaustion_label(policy: &GuardrailPolicy) -> &'static str {
    match policy.retry_exhaustion_mode {
        RetryExhaustionMode::Error => "error",
        RetryExhaustionMode::PassLastText => "pass_last_text",
    }
}

fn small_model_policy_label(policy: &GuardrailPolicy) -> &'static str {
    if policy.apply_to_all_models {
        "all"
    } else {
        "small_models_only"
    }
}