behest 0.2.1

A Rust-native cloud agent runtime with typed tools, pluggable memory, queues, and observability.
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

//! OpenAI embedding provider adapter implementing [`EmbeddingProvider`].

use async_trait::async_trait;
use reqwest::Client;

use crate::adapt::http::{build_client, parse_retry_after, status_to_error, with_bearer_auth};
use crate::error::ProviderError;
use crate::provider::{
    Embedding, EmbeddingInput, EmbeddingProvider, EmbeddingRequest, EmbeddingResponse, ModelName,
    ProviderCapabilities, ProviderHttpConfig, ProviderId, ProviderResult, TokenUsage,
};

use super::types::{OpenAiEmbeddingRequest, OpenAiEmbeddingResponse};

/// OpenAI-compatible embedding adapter.
pub struct OpenAiEmbeddingAdapter {
    id: ProviderId,
    client: Client,
    config: ProviderHttpConfig,
}

impl OpenAiEmbeddingAdapter {
    /// Creates an OpenAI embedding adapter with a new HTTP client.
    ///
    /// # Errors
    ///
    /// Returns [`ProviderError::Transport`] when the HTTP client cannot be built.
    pub fn new(config: ProviderHttpConfig) -> Result<Self, ProviderError> {
        let client = build_client(&config)?;
        Ok(Self {
            id: config.id.clone(),
            client,
            config,
        })
    }

    /// Creates an OpenAI embedding adapter reusing an existing HTTP client.
    #[must_use]
    pub fn with_client(config: ProviderHttpConfig, client: Client) -> Self {
        Self {
            id: config.id.clone(),
            client,
            config,
        }
    }

    fn url(&self) -> String {
        format!("{}/embeddings", self.config.base_url)
    }

    fn wrap_transport(&self, source: reqwest::Error) -> ProviderError {
        if source.is_timeout() {
            ProviderError::Timeout {
                provider: self.id.clone(),
            }
        } else {
            ProviderError::Transport {
                provider: self.id.clone(),
                source,
            }
        }
    }
}

#[async_trait]
impl EmbeddingProvider for OpenAiEmbeddingAdapter {
    fn id(&self) -> ProviderId {
        self.id.clone()
    }

    fn capabilities(&self) -> ProviderCapabilities {
        ProviderCapabilities::embeddings()
    }

    async fn embed(&self, request: EmbeddingRequest) -> ProviderResult<EmbeddingResponse> {
        let input_texts = extract_texts(&request.input);
        let body = OpenAiEmbeddingRequest {
            model: request.model.as_str().to_owned(),
            input: input_texts,
            dimensions: request.dimensions,
        };

        let builder = self.client.post(self.url()).json(&body);
        let builder = with_bearer_auth(builder, &self.config);
        let response = builder.send().await.map_err(|e| self.wrap_transport(e))?;

        if !response.status().is_success() {
            let status = response.status();
            let retry_after = parse_retry_after(response.headers());
            let text = response
                .text()
                .await
                .unwrap_or_else(|e| format!("<failed to read error body: {e}>"));
            return Err(status_to_error(&self.id, status, &text, retry_after));
        }

        let parsed: OpenAiEmbeddingResponse =
            response.json().await.map_err(|e| ProviderError::Decode {
                provider: self.id.clone(),
                message: e.to_string(),
            })?;

        Ok(from_response(&self.id, &request.model, &parsed))
    }
}

fn extract_texts(inputs: &[EmbeddingInput]) -> Vec<String> {
    inputs
        .iter()
        .map(|input| match input {
            EmbeddingInput::Text { text } => text.clone(),
            EmbeddingInput::Tokens { tokens } => tokens
                .iter()
                .map(ToString::to_string)
                .collect::<Vec<_>>()
                .join(" "),
        })
        .collect()
}

fn from_response(
    provider: &ProviderId,
    model: &ModelName,
    response: &OpenAiEmbeddingResponse,
) -> EmbeddingResponse {
    let embeddings = response
        .data
        .iter()
        .map(|d| Embedding::new(d.index, d.embedding.clone()))
        .collect();

    // Embeddings only consume input tokens; output_tokens is always 0.
    let usage = response
        .usage
        .as_ref()
        .map(|u| TokenUsage::new(u.prompt_tokens, 0));

    EmbeddingResponse {
        provider: provider.clone(),
        model: model.clone(),
        embeddings,
        usage,
        raw: None,
    }
}