chore: extraction model api (no proxying implemented yet)

src/server/chat/completions.rs

@@ -21,8 +21,6 @@ use secrecy::ExposeSecret;
 use serde::Deserialize;
 use serde_with::DeserializeFromStr;
 use serde_with::SerializeDisplay;
-use tokio::select;
-use tokio::time::sleep;
 use std::collections::VecDeque;
 use std::convert::Infallible;
 use std::fmt::Display;
@@ -31,6 +29,8 @@ use std::time::Duration;
 use strum::EnumString;
 use strum::VariantArray;
 use tokio::runtime::Handle;
+use tokio::select;
+use tokio::time::sleep;
 
 #[derive(Deserialize, TypedPath)]
 #[typed_path("/v1/chat/completions")]
@@ -201,7 +201,6 @@ impl Shortcut {
                 (P::OpenRouter, "poolside/laguna-s-2.1:free".into()),
                 (P::VercelAiGateway, "stepfun/step-3.7-flash".into()),
                 (P::VercelAiGateway, "inclusionai/ling-3.0-flash-fin".into()),
-                
                 (
                     P::VercelAiGateway,
                     "inclusionai/ling-3.0-flash-sante".into(),
@@ -259,7 +258,10 @@ impl Shortcut {
                 (P::Ollama, "nemotron-3-ultra".into()),
                 (P::NvidiaNim, "nvidia/nemotron-3-ultra-550b-a55b".into()),
                 (P::Kilo, "nvidia-nemotron-3-ultra-550b-a55b-free".into()),
-                (P::OpenRouter, "nvidia/nemotron-3-ultra-550b-a55b:free".into()),
+                (
+                    P::OpenRouter,
+                    "nvidia/nemotron-3-ultra-550b-a55b:free".into(),
+                ),
                 (P::OpenCode, "nemotron-3-ultra-free".into()),
             ],
 
@@ -268,8 +270,14 @@ impl Shortcut {
                 (P::HuggingFace, "zai-org/GLM-5.3".into()),
                 (P::CloudflareWorkersAi, "@cf/zai-org/glm-5.3".into()),
                 (P::Mistral, "zai-glm-5-3".into()),
-                (P::VercelAiGateway, "deepseek/deepseek-v4.1-flash-1017".into()),
-                (P::CloudflareWorkersAi, "@cf/deepseek-ai/deepseek-v4.1-flash-1017".into()),
+                (
+                    P::VercelAiGateway,
+                    "deepseek/deepseek-v4.1-flash-1017".into(),
+                ),
+                (
+                    P::CloudflareWorkersAi,
+                    "@cf/deepseek-ai/deepseek-v4.1-flash-1017".into(),
+                ),
                 (P::OpenCode, "muse-spark-1.3-contributor-free".into()),
                 (P::NvidiaNim, "moonshotai/kimi-k3".into()),
                 (P::HuggingFace, "moonshotai/Kimi-K3".into()),

src/server/embeddings.rs

@@ -73,8 +73,10 @@ impl Shortcut {
 
         match self {
             S::CodestralEmbed2505 => vec![(P::Mistral, "codestral-embed-2505".into())],
-            S::CohereEmbedV4_0 => vec![(P::Cohere, "embed-v4.0".into()),
-                (P::VercelAiGateway, "cohere/embed-v4.0".into())],
+            S::CohereEmbedV4_0 => vec![
+                (P::Cohere, "embed-v4.0".into()),
+                (P::VercelAiGateway, "cohere/embed-v4.0".into()),
+            ],
             S::CohereEmbedV5_0Fast => vec![
                 (P::Cohere, "embed-v5.0-fast".into()),
                 (P::VercelAiGateway, "cohere/embed-v5.0-fast".into()),
@@ -209,18 +211,22 @@ pub async fn create_embeddings(
                         .build()
                 })
                 .collect(),
-                Provider::VercelAiGateway => state
-                    .app_args
-                    .vercel_ai_gateway_api_key
-                    .iter()
-                    .cloned()
-                    .map(|api_key| {
-                        Upstream::builder()
-                            .url("https://ai-gateway.vercel.sh/v1/embeddings".parse().expect("TODO"))
-                            .api_key(api_key)
-                            .build()
-                    })
-                    .collect(),
+            Provider::VercelAiGateway => state
+                .app_args
+                .vercel_ai_gateway_api_key
+                .iter()
+                .cloned()
+                .map(|api_key| {
+                    Upstream::builder()
+                        .url(
+                            "https://ai-gateway.vercel.sh/v1/embeddings"
+                                .parse()
+                                .expect("TODO"),
+                        )
+                        .api_key(api_key)
+                        .build()
+                })
+                .collect(),
         };
 
         if let Ok(upstreams) = Vec1::try_from(upstreams) {

src/server/mod.rs

@@ -12,8 +12,8 @@ use url::Url;
 
 mod chat;
 mod decisions;
-mod frontend;
 mod embeddings;
+mod frontend;
 mod models;
 mod system_one;
 

src/server/system_one.rs

@@ -53,7 +53,10 @@ impl Shortcut {
         match self {
             S::S => {
                 vec![
-                    (P::VercelAiGateway, "perplexity/pplx-decider-flash-v1.2".into()),
+                    (
+                        P::VercelAiGateway,
+                        "perplexity/pplx-decider-flash-v1.2".into(),
+                    ),
                     (P::LiquidAi, "d1:free".into()),
                     (P::CloudflareWorkersAi, "clef-omni".into()),
                     (P::CloudflareWorkersAi, "clef-flash".into()),
@@ -74,9 +77,7 @@ impl Shortcut {
             }
 
             S::L => {
-                vec![
-                    (P::VercelAiGateway, "liquid/d1-ultra".into())
-                ]
+                vec![(P::VercelAiGateway, "liquid/d1-ultra".into())]
             }
         }
     }