synth_ai_core/api/
inference.rs

1//! Inference API client.
2
3use serde_json::Value;
4
5use crate::errors::CoreError;
6use crate::http::HttpClient;
7use crate::models::normalize_model_identifier;
8
9/// Inference client for chat completions.
10pub struct InferenceClient<'a> {
11    http: &'a HttpClient,
12}
13
14impl<'a> InferenceClient<'a> {
15    pub(crate) fn new(http: &'a HttpClient) -> Self {
16        Self { http }
17    }
18
19    /// Create a chat completion request through Synth inference proxy.
20    pub async fn chat_completion(&self, mut body: Value) -> Result<Value, CoreError> {
21        let obj = body
22            .as_object_mut()
23            .ok_or_else(|| CoreError::Validation("request must be an object".to_string()))?;
24
25        let model = obj
26            .get("model")
27            .and_then(|v| v.as_str())
28            .ok_or_else(|| CoreError::Validation("model is required".to_string()))?;
29        let normalized = normalize_model_identifier(model, false)?;
30        obj.insert("model".to_string(), Value::String(normalized));
31
32        if !obj.contains_key("thinking_budget") {
33            obj.insert("thinking_budget".to_string(), Value::Number(256.into()));
34        }
35
36        self.http
37            .post_json("/api/inference/v1/chat/completions", &body)
38            .await
39            .map_err(CoreError::from)
40    }
41}
synth_ai_core/api/inference.rs

synth_ai_core/api/
inference.rs
