feat(server): refactor for byok (#14911)

packages/backend/native/src/llm/assets/prompts/built-in.json

@@ -686,17 +686,6 @@
       }
     ]
   },
-  {
-    "name": "Apply Updates",
-    "action": "Apply Updates",
-    "model": "claude-sonnet-4-5@20250929",
-    "messages": [
-      {
-        "role": "user",
-        "template": "\nYou are a Markdown document update engine.\n\nYou will be given:\n\n1. content: The original Markdown document\n   - The content is structured into blocks.\n   - Each block starts with a comment like <!-- block_id=... flavour=... --> and contains the block's content.\n   - The content is {{content}}\n\n2. op: A description of the edit intention\n   - This describes the semantic meaning of the edit, such as \"Bold the first paragraph\".\n   - The op is {{op}}\n\n3. updates: A Markdown snippet\n   - The updates is {{updates}}\n   - This represents the block-level changes to apply to the original Markdown.\n   - The update may:\n     - **Replace** an existing block (same block_id, new content)\n     - **Delete** block(s) using <!-- delete block BLOCK_ID -->\n     - **Insert** new block(s) with a new unique block_id\n   - When performing deletions, the update will include **surrounding context blocks** (or use <!-- existing blocks -->) to help you determine where and what to delete.\n\nYour task:\n- Apply the update in <updates> to the document in <code>, following the intent described in <op>.\n- Preserve all block_id and flavour comments.\n- Maintain the original block order unless the update clearly appends new blocks.\n- Do not remove or alter unrelated blocks.\n- Output only the fully updated Markdown content. Do not wrap the content in ```markdown.\n\n---\n\n✍️ Examples\n\n✅ Replacement (modifying an existing block)\n\n<code>\n<!-- block_id=101 flavour=paragraph -->\n## Introduction\n\n<!-- block_id=102 flavour=paragraph -->\nThis document provides an overview of the system architecture and its components.\n</code>\n\n<op>\nMake the introduction more formal.\n</op>\n\n<updates>\n<!-- block_id=102 flavour=paragraph -->\nThis document outlines the architectural design and individual components of the system in detail.\n</updates>\n\nExpected Output:\n<!-- block_id=101 flavour=paragraph -->\n## Introduction\n\n<!-- block_id=102 flavour=paragraph -->\nThis document outlines the architectural design and individual components of the system in detail.\n\n---\n\n➕ Insertion (adding new content)\n\n<code>\n<!-- block_id=201 flavour=paragraph -->\n# Project Summary\n\n<!-- block_id=202 flavour=paragraph -->\nThis project aims to build a collaborative text editing tool.\n</code>\n\n<op>\nAdd a disclaimer section at the end.\n</op>\n\n<updates>\n<!-- block_id=new-301 flavour=paragraph -->\n## Disclaimer\n\n<!-- block_id=new-302 flavour=paragraph -->\nThis document is subject to change. Do not distribute externally.\n</updates>\n\nExpected Output:\n<!-- block_id=201 flavour=paragraph -->\n# Project Summary\n\n<!-- block_id=202 flavour=paragraph -->\nThis project aims to build a collaborative text editing tool.\n\n<!-- block_id=new-301 flavour=paragraph -->\n## Disclaimer\n\n<!-- block_id=new-302 flavour=paragraph -->\nThis document is subject to change. Do not distribute externally.\n\n---\n\n❌ Deletion (removing blocks)\n\n<code>\n<!-- block_id=401 flavour=paragraph -->\n## Author\n\n<!-- block_id=402 flavour=paragraph -->\nWritten by the AI team at OpenResearch.\n\n<!-- block_id=403 flavour=paragraph -->\n## Experimental Section\n\n<!-- block_id=404 flavour=paragraph -->\nThe following section is still under development and may change without notice.\n\n<!-- block_id=405 flavour=paragraph -->\n## License\n\n<!-- block_id=406 flavour=paragraph -->\nThis document is licensed under CC BY-NC 4.0.\n</code>\n\n<op>\nRemove the experimental section.\n</op>\n\n<updates>\n<!-- delete block_id=403 -->\n<!-- delete block_id=404 -->\n</updates>\n\nExpected Output:\n<!-- block_id=401 flavour=paragraph -->\n## Author\n\n<!-- block_id=402 flavour=paragraph -->\nWritten by the AI team at OpenResearch.\n\n<!-- block_id=405 flavour=paragraph -->\n## License\n\n<!-- block_id=406 flavour=paragraph -->\nThis document is licensed under CC BY-NC 4.0.\n\n---\n\nNow apply the `updates` to the `content`, following the intent in `op`, and return the updated Markdown.\n"
-      }
-    ]
-  },
   {
     "name": "Code Artifact",
     "model": "claude-sonnet-4-5@20250929",

packages/backend/native/src/llm/core/contracts/mod.rs

@@ -319,7 +319,7 @@ pub struct RequestedModelMatchResponse {
 pub struct ModelRegistryResolveRequest {
   #[napi(
     ts_type = "'openai_chat' | 'openai_responses' | 'anthropic' | 'cloudflare_workers_ai' | 'gemini_api' | \
-               'gemini_vertex' | 'fal' | 'perplexity' | 'anthropic_vertex' | 'morph'"
+               'gemini_vertex' | 'fal' | 'anthropic_vertex'"
   )]
   #[serde(skip_serializing_if = "Option::is_none")]
   pub backend_kind: Option<String>,
@@ -333,7 +333,7 @@ pub struct ModelRegistryResolveRequest {
 pub struct ModelRegistryMatchRequest {
   #[napi(
     ts_type = "'openai_chat' | 'openai_responses' | 'anthropic' | 'cloudflare_workers_ai' | 'gemini_api' | \
-               'gemini_vertex' | 'fal' | 'perplexity' | 'anthropic_vertex' | 'morph'"
+               'gemini_vertex' | 'fal' | 'anthropic_vertex'"
   )]
   pub backend_kind: String,
   pub cond: ModelConditionsContract,
@@ -346,7 +346,7 @@ pub struct ModelRegistryMatchRequest {
 pub struct ModelRegistryVariantContract {
   #[napi(
     ts_type = "'openai_chat' | 'openai_responses' | 'anthropic' | 'cloudflare_workers_ai' | 'gemini_api' | \
-               'gemini_vertex' | 'fal' | 'perplexity' | 'anthropic_vertex' | 'morph'"
+               'gemini_vertex' | 'fal' | 'anthropic_vertex'"
   )]
   pub backend_kind: String,
   pub canonical_key: String,

packages/backend/native/src/llm/host/mod.rs

@@ -7,7 +7,7 @@ pub(crate) use error::{
   STREAM_ABORTED_REASON, STREAM_CALLBACK_DISPATCH_FAILED_REASON, STREAM_END_MARKER, callback_dispatch_failed_reason,
   invalid_arg,
 };
-pub(crate) use stream::emit_error_event;
+pub(crate) use stream::{emit_error_event, emit_provider_selected_event};
 pub use stream::{
   llm_dispatch_prepared_stream, llm_dispatch_tool_loop_stream, llm_dispatch_tool_loop_stream_prepared,
   llm_dispatch_tool_loop_stream_routed,

packages/backend/native/src/llm/host/stream.rs

@@ -106,14 +106,18 @@ fn spawn_prepared_stream(
         if reason.starts_with(STREAM_CALLBACK_DISPATCH_FAILED_REASON)
     );
 
-    if let Err(error) = result
+    if let Err(error) = &result
       && !aborted_in_worker.load(Ordering::Relaxed)
       && !callback_dispatch_failed
-      && !is_abort_error(&error)
+      && !is_abort_error(error)
     {
       emit_error_event(&callback, error.to_string(), "dispatch_error");
     }
 
+    if let Ok(provider_id) = result {
+      emit_provider_selected_event(&callback, provider_id);
+    }
+
     if !callback_dispatch_failed {
       let _ = callback.call(
         Ok(STREAM_END_MARKER.to_string()),
@@ -129,7 +133,7 @@ fn dispatch_prepared_stream_with_fallback(
   routes: &[PreparedDispatchRoute],
   callback: &ThreadsafeFunction<String, ()>,
   aborted: &AtomicBool,
-) -> std::result::Result<(), BackendError> {
+) -> std::result::Result<String, BackendError> {
   dispatch_prepared_stream_with_fallback_using_client(&DefaultHttpClient::default(), routes, aborted, |event| {
     emit_stream_event(callback, event)
   })
@@ -140,7 +144,7 @@ fn dispatch_prepared_stream_with_fallback_using_client<F>(
   routes: &[PreparedDispatchRoute],
   aborted: &AtomicBool,
   mut emit_event: F,
-) -> std::result::Result<(), BackendError>
+) -> std::result::Result<String, BackendError>
 where
   F: FnMut(&StreamEvent) -> Status,
 {
@@ -154,7 +158,7 @@ where
     .collect::<std::result::Result<Vec<_>, BackendError>>()?;
   let mut callback_dispatch_failed = false;
 
-  dispatch_prepared_stream_with_pipeline(
+  let provider_id = dispatch_prepared_stream_with_pipeline(
     client,
     &mut adapter_routes,
     || aborted.load(Ordering::Relaxed),
@@ -174,7 +178,7 @@ where
       "{STREAM_CALLBACK_DISPATCH_FAILED_REASON}:unknown"
     )))
   } else {
-    Ok(())
+    Ok(provider_id)
   }
 }
 
@@ -195,6 +199,16 @@ pub(crate) fn emit_error_event(callback: &ThreadsafeFunction<String, ()>, messag
   let _ = callback.call(Ok(error_event), ThreadsafeFunctionCallMode::NonBlocking);
 }
 
+pub(crate) fn emit_provider_selected_event(callback: &ThreadsafeFunction<String, ()>, provider_id: String) {
+  let event = serde_json::json!({
+    "type": "provider_selected",
+    "provider_id": provider_id,
+  })
+  .to_string();
+
+  let _ = callback.call(Ok(event), ThreadsafeFunctionCallMode::NonBlocking);
+}
+
 fn emit_stream_event(callback: &ThreadsafeFunction<String, ()>, event: &StreamEvent) -> Status {
   let value = serde_json::to_string(event).unwrap_or_else(|error| {
     serde_json::json!({

packages/backend/native/src/llm/host/tool_loop/engine.rs

@@ -14,7 +14,10 @@ use napi::{
   threadsafe_function::{ThreadsafeFunction, ThreadsafeFunctionCallMode},
 };
 
-use super::callback::{NapiEventSink, NapiToolExecutor, emit_tool_loop_event};
+use super::{
+  super::emit_provider_selected_event,
+  callback::{NapiEventSink, NapiToolExecutor, emit_tool_loop_event},
+};
 use crate::llm::{
   LlmDispatchPayload, LlmMiddlewarePayload, LlmStreamHandle, STREAM_ABORTED_REASON,
   STREAM_CALLBACK_DISPATCH_FAILED_REASON, STREAM_END_MARKER, StreamPipeline, apply_request_middlewares,
@@ -39,11 +42,13 @@ fn dispatch_prepared_round_with_fallback(
     })
     .collect::<std::result::Result<Vec<_>, BackendError>>()?;
 
-  run_prepared_stream_round_with_fallback(
+  let mut selected_provider_id: Option<String> = None;
+  let outcome = run_prepared_stream_round_with_fallback(
     &mut pipelines,
     |on_event| {
-      let (selected_index, _) =
+      let (selected_index, provider_id) =
         dispatch_prepared_stream_with_fallback_index(&DefaultHttpClient::default(), &adapter_routes, on_event)?;
+      selected_provider_id = Some(provider_id);
       Ok(selected_index)
     },
     || aborted.load(Ordering::Relaxed),
@@ -53,7 +58,11 @@ fn dispatch_prepared_round_with_fallback(
       emitted.store(true, Ordering::Relaxed);
       emit_tool_loop_event(callback, loop_event)
     },
-  )
+  )?;
+  if let Some(provider_id) = selected_provider_id {
+    emit_provider_selected_event(callback, provider_id);
+  }
+  Ok(outcome)
 }
 
 fn prepare_tool_loop_route(