Pass up intent with completion requests (#31710)

This PR adds a new `intent` field to completion requests to assist in
categorizing them correctly.

Release Notes:

- N/A

---------

Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>

Cargo.lock

@@ -559,6 +559,7 @@ dependencies = [
  "workspace",
  "workspace-hack",
  "zed_actions",
+ "zed_llm_client",
 ]
 
 [[package]]
@@ -5047,6 +5048,7 @@ dependencies = [
  "util",
  "uuid",
  "workspace-hack",
+ "zed_llm_client",
 ]
 
 [[package]]
@@ -6149,6 +6151,7 @@ dependencies = [
  "workspace",
  "workspace-hack",
  "zed_actions",
+ "zed_llm_client",
  "zlog",
 ]
 
@@ -19876,9 +19879,9 @@ dependencies = [
 
 [[package]]
 name = "zed_llm_client"
-version = "0.8.3"
+version = "0.8.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "22a8b9575b215536ed8ad254ba07171e4e13bd029eda3b54cca4b184d2768050"
+checksum = "de7d9523255f4e00ee3d0918e5407bd252d798a4a8e71f6d37f23317a1588203"
 dependencies = [
  "anyhow",
  "serde",

Cargo.toml

@@ -617,7 +617,7 @@ wasmtime = { version = "29", default-features = false, features = [
 wasmtime-wasi = "29"
 which = "6.0.0"
 workspace-hack = "0.1.0"
-zed_llm_client = "0.8.3"
+zed_llm_client = "0.8.4"
 zstd = "0.11"
 
 [workspace.dependencies.async-stripe]

crates/agent/src/active_thread.rs

@@ -55,6 +55,7 @@ use util::ResultExt as _;
 use util::markdown::MarkdownCodeBlock;
 use workspace::{CollaboratorId, Workspace};
 use zed_actions::assistant::OpenRulesLibrary;
+use zed_llm_client::CompletionIntent;
 
 pub struct ActiveThread {
     context_store: Entity<ContextStore>,
@@ -1436,6 +1437,7 @@ impl ActiveThread {
                     let request = language_model::LanguageModelRequest {
                         thread_id: None,
                         prompt_id: None,
+                        intent: None,
                         mode: None,
                         messages: vec![request_message],
                         tools: vec![],
@@ -1610,7 +1612,12 @@ impl ActiveThread {
 
                         this.thread.update(cx, |thread, cx| {
                             thread.advance_prompt_id();
-                            thread.send_to_model(model.model, Some(window.window_handle()), cx);
+                            thread.send_to_model(
+                                model.model,
+                                CompletionIntent::UserPrompt,
+                                Some(window.window_handle()),
+                                cx,
+                            );
                         });
                         this._load_edited_message_context_task = None;
                         cx.notify();
@@ -3702,7 +3709,8 @@ mod tests {
 
         // Stream response to user message
         thread.update(cx, |thread, cx| {
-            let request = thread.to_completion_request(model.clone(), cx);
+            let request =
+                thread.to_completion_request(model.clone(), CompletionIntent::UserPrompt, cx);
             thread.stream_completion(request, model, cx.active_window(), cx)
         });
         // Follow the agent

crates/agent/src/agent_panel.rs

@@ -52,7 +52,7 @@ use workspace::{
 use zed_actions::agent::{OpenConfiguration, OpenOnboardingModal, ResetOnboarding};
 use zed_actions::assistant::{OpenRulesLibrary, ToggleFocus};
 use zed_actions::{DecreaseBufferFontSize, IncreaseBufferFontSize, ResetBufferFontSize};
-use zed_llm_client::UsageLimit;
+use zed_llm_client::{CompletionIntent, UsageLimit};
 
 use crate::active_thread::{self, ActiveThread, ActiveThreadEvent};
 use crate::agent_configuration::{AgentConfiguration, AssistantConfigurationEvent};
@@ -1310,7 +1310,12 @@ impl AgentPanel {
                 active_thread.thread().update(cx, |thread, cx| {
                     thread.insert_invisible_continue_message(cx);
                     thread.advance_prompt_id();
-                    thread.send_to_model(model, Some(window.window_handle()), cx);
+                    thread.send_to_model(
+                        model,
+                        CompletionIntent::UserPrompt,
+                        Some(window.window_handle()),
+                        cx,
+                    );
                 });
             });
         } else {

crates/agent/src/buffer_codegen.rs

@@ -34,6 +34,7 @@ use std::{
 };
 use streaming_diff::{CharOperation, LineDiff, LineOperation, StreamingDiff};
 use telemetry_events::{AssistantEventData, AssistantKind, AssistantPhase};
+use zed_llm_client::CompletionIntent;
 
 pub struct BufferCodegen {
     alternatives: Vec<Entity<CodegenAlternative>>,
@@ -464,6 +465,7 @@ impl CodegenAlternative {
             LanguageModelRequest {
                 thread_id: None,
                 prompt_id: None,
+                intent: Some(CompletionIntent::InlineAssist),
                 mode: None,
                 tools: Vec::new(),
                 tool_choice: None,

crates/agent/src/message_editor.rs

@@ -42,6 +42,7 @@ use theme::ThemeSettings;
 use ui::{Disclosure, KeyBinding, PopoverMenuHandle, Tooltip, prelude::*};
 use util::{ResultExt as _, maybe};
 use workspace::{CollaboratorId, Workspace};
+use zed_llm_client::CompletionIntent;
 
 use crate::context_picker::{ContextPicker, ContextPickerCompletionProvider, crease_for_mention};
 use crate::context_store::ContextStore;
@@ -375,7 +376,12 @@ impl MessageEditor {
             thread
                 .update(cx, |thread, cx| {
                     thread.advance_prompt_id();
-                    thread.send_to_model(model, Some(window_handle), cx);
+                    thread.send_to_model(
+                        model,
+                        CompletionIntent::UserPrompt,
+                        Some(window_handle),
+                        cx,
+                    );
                 })
                 .log_err();
         })
@@ -1280,6 +1286,7 @@ impl MessageEditor {
                     let request = language_model::LanguageModelRequest {
                         thread_id: None,
                         prompt_id: None,
+                        intent: None,
                         mode: None,
                         messages: vec![request_message],
                         tools: vec![],

crates/agent/src/terminal_inline_assistant.rs

@@ -25,6 +25,7 @@ use terminal_view::TerminalView;
 use ui::prelude::*;
 use util::ResultExt;
 use workspace::{Toast, Workspace, notifications::NotificationId};
+use zed_llm_client::CompletionIntent;
 
 pub fn init(
     fs: Arc<dyn Fs>,
@@ -291,6 +292,7 @@ impl TerminalInlineAssistant {
                 thread_id: None,
                 prompt_id: None,
                 mode: None,
+                intent: Some(CompletionIntent::TerminalInlineAssist),
                 messages: vec![request_message],
                 tools: Vec::new(),
                 tool_choice: None,

crates/agent/src/thread.rs

@@ -38,7 +38,7 @@ use thiserror::Error;
 use ui::Window;
 use util::{ResultExt as _, post_inc};
 use uuid::Uuid;
-use zed_llm_client::CompletionRequestStatus;
+use zed_llm_client::{CompletionIntent, CompletionRequestStatus};
 
 use crate::ThreadStore;
 use crate::context::{AgentContext, AgentContextHandle, ContextLoadResult, LoadedContext};
@@ -1184,6 +1184,7 @@ impl Thread {
     pub fn send_to_model(
         &mut self,
         model: Arc<dyn LanguageModel>,
+        intent: CompletionIntent,
         window: Option<AnyWindowHandle>,
         cx: &mut Context<Self>,
     ) {
@@ -1193,7 +1194,7 @@ impl Thread {
 
         self.remaining_turns -= 1;
 
-        let request = self.to_completion_request(model.clone(), cx);
+        let request = self.to_completion_request(model.clone(), intent, cx);
 
         self.stream_completion(request, model, window, cx);
     }
@@ -1213,11 +1214,13 @@ impl Thread {
     pub fn to_completion_request(
         &self,
         model: Arc<dyn LanguageModel>,
+        intent: CompletionIntent,
         cx: &mut Context<Self>,
     ) -> LanguageModelRequest {
         let mut request = LanguageModelRequest {
             thread_id: Some(self.id.to_string()),
             prompt_id: Some(self.last_prompt_id.to_string()),
+            intent: Some(intent),
             mode: None,
             messages: vec![],
             tools: Vec::new(),
@@ -1371,12 +1374,14 @@ impl Thread {
     fn to_summarize_request(
         &self,
         model: &Arc<dyn LanguageModel>,
+        intent: CompletionIntent,
         added_user_message: String,
         cx: &App,
     ) -> LanguageModelRequest {
         let mut request = LanguageModelRequest {
             thread_id: None,
             prompt_id: None,
+            intent: Some(intent),
             mode: None,
             messages: vec![],
             tools: Vec::new(),
@@ -1854,7 +1859,12 @@ impl Thread {
             If the conversation is about a specific subject, include it in the title. \
             Be descriptive. DO NOT speak in the first person.";
 
-        let request = self.to_summarize_request(&model.model, added_user_message.into(), cx);
+        let request = self.to_summarize_request(
+            &model.model,
+            CompletionIntent::ThreadSummarization,
+            added_user_message.into(),
+            cx,
+        );
 
         self.summary = ThreadSummary::Generating;
 
@@ -1955,7 +1965,12 @@ impl Thread {
              4. Any action items or next steps if any\n\
              Format it in Markdown with headings and bullet points.";
 
-        let request = self.to_summarize_request(&model, added_user_message.into(), cx);
+        let request = self.to_summarize_request(
+            &model,
+            CompletionIntent::ThreadContextSummarization,
+            added_user_message.into(),
+            cx,
+        );
 
         *self.detailed_summary_tx.borrow_mut() = DetailedSummaryState::Generating {
             message_id: last_message_id,
@@ -2047,7 +2062,8 @@ impl Thread {
         model: Arc<dyn LanguageModel>,
     ) -> Vec<PendingToolUse> {
         self.auto_capture_telemetry(cx);
-        let request = Arc::new(self.to_completion_request(model.clone(), cx));
+        let request =
+            Arc::new(self.to_completion_request(model.clone(), CompletionIntent::ToolResults, cx));
         let pending_tool_uses = self
             .tool_use
             .pending_tool_uses()
@@ -2243,7 +2259,7 @@ impl Thread {
         if self.all_tools_finished() {
             if let Some(ConfiguredModel { model, .. }) = self.configured_model.as_ref() {
                 if !canceled {
-                    self.send_to_model(model.clone(), window, cx);
+                    self.send_to_model(model.clone(), CompletionIntent::ToolResults, window, cx);
                 }
                 self.auto_capture_telemetry(cx);
             }
@@ -2934,7 +2950,7 @@ fn main() {{
 
         // Check message in request
         let request = thread.update(cx, |thread, cx| {
-            thread.to_completion_request(model.clone(), cx)
+            thread.to_completion_request(model.clone(), CompletionIntent::UserPrompt, cx)
         });
 
         assert_eq!(request.messages.len(), 2);
@@ -3029,7 +3045,7 @@ fn main() {{
 
         // Check entire request to make sure all contexts are properly included
         let request = thread.update(cx, |thread, cx| {
-            thread.to_completion_request(model.clone(), cx)
+            thread.to_completion_request(model.clone(), CompletionIntent::UserPrompt, cx)
         });
 
         // The request should contain all 3 messages
@@ -3136,7 +3152,7 @@ fn main() {{
 
         // Check message in request
         let request = thread.update(cx, |thread, cx| {
-            thread.to_completion_request(model.clone(), cx)
+            thread.to_completion_request(model.clone(), CompletionIntent::UserPrompt, cx)
         });
 
         assert_eq!(request.messages.len(), 2);
@@ -3162,7 +3178,7 @@ fn main() {{
 
         // Check that both messages appear in the request
         let request = thread.update(cx, |thread, cx| {
-            thread.to_completion_request(model.clone(), cx)
+            thread.to_completion_request(model.clone(), CompletionIntent::UserPrompt, cx)
         });
 
         assert_eq!(request.messages.len(), 3);
@@ -3207,7 +3223,7 @@ fn main() {{
 
         // Create a request and check that it doesn't have a stale buffer warning yet
         let initial_request = thread.update(cx, |thread, cx| {
-            thread.to_completion_request(model.clone(), cx)
+            thread.to_completion_request(model.clone(), CompletionIntent::UserPrompt, cx)
         });
 
         // Make sure we don't have a stale file warning yet
@@ -3243,7 +3259,7 @@ fn main() {{
 
         // Create a new request and check for the stale buffer warning
         let new_request = thread.update(cx, |thread, cx| {
-            thread.to_completion_request(model.clone(), cx)
+            thread.to_completion_request(model.clone(), CompletionIntent::UserPrompt, cx)
         });
 
         // We should have a stale file warning as the last message
@@ -3293,7 +3309,7 @@ fn main() {{
         });
 
         let request = thread.update(cx, |thread, cx| {
-            thread.to_completion_request(model.clone(), cx)
+            thread.to_completion_request(model.clone(), CompletionIntent::UserPrompt, cx)
         });
         assert_eq!(request.temperature, Some(0.66));
 
@@ -3313,7 +3329,7 @@ fn main() {{
         });
 
         let request = thread.update(cx, |thread, cx| {
-            thread.to_completion_request(model.clone(), cx)
+            thread.to_completion_request(model.clone(), CompletionIntent::UserPrompt, cx)
         });
         assert_eq!(request.temperature, Some(0.66));
 
@@ -3333,7 +3349,7 @@ fn main() {{
         });
 
         let request = thread.update(cx, |thread, cx| {
-            thread.to_completion_request(model.clone(), cx)
+            thread.to_completion_request(model.clone(), CompletionIntent::UserPrompt, cx)
         });
         assert_eq!(request.temperature, Some(0.66));
 
@@ -3353,7 +3369,7 @@ fn main() {{
         });
 
         let request = thread.update(cx, |thread, cx| {
-            thread.to_completion_request(model.clone(), cx)
+            thread.to_completion_request(model.clone(), CompletionIntent::UserPrompt, cx)
         });
         assert_eq!(request.temperature, None);
     }
@@ -3385,7 +3401,12 @@ fn main() {{
         // Send a message
         thread.update(cx, |thread, cx| {
             thread.insert_user_message("Hi!", ContextLoadResult::default(), None, vec![], cx);
-            thread.send_to_model(model.clone(), None, cx);
+            thread.send_to_model(
+                model.clone(),
+                CompletionIntent::ThreadSummarization,
+                None,
+                cx,
+            );
         });
 
         let fake_model = model.as_fake();
@@ -3480,7 +3501,7 @@ fn main() {{
                 vec![],
                 cx,
             );
-            thread.send_to_model(model.clone(), None, cx);
+            thread.send_to_model(model.clone(), CompletionIntent::UserPrompt, None, cx);
         });
 
         let fake_model = model.as_fake();
@@ -3518,7 +3539,12 @@ fn main() {{
     ) {
         thread.update(cx, |thread, cx| {
             thread.insert_user_message("Hi!", ContextLoadResult::default(), None, vec![], cx);
-            thread.send_to_model(model.clone(), None, cx);
+            thread.send_to_model(
+                model.clone(),
+                CompletionIntent::ThreadSummarization,
+                None,
+                cx,
+            );
         });
 
         let fake_model = model.as_fake();

crates/assistant_context_editor/Cargo.toml

@@ -57,6 +57,7 @@ uuid.workspace = true
 workspace-hack.workspace = true
 workspace.workspace = true
 zed_actions.workspace = true
+zed_llm_client.workspace = true
 
 [dev-dependencies]
 language_model = { workspace = true, features = ["test-support"] }

crates/assistant_context_editor/src/context.rs

@@ -45,6 +45,7 @@ use text::{BufferSnapshot, ToPoint};
 use ui::IconName;
 use util::{ResultExt, TryFutureExt, post_inc};
 use uuid::Uuid;
+use zed_llm_client::CompletionIntent;
 
 #[derive(Clone, Debug, Eq, PartialEq, Hash, PartialOrd, Ord, Serialize, Deserialize)]
 pub struct ContextId(String);
@@ -2272,6 +2273,7 @@ impl AssistantContext {
         let mut completion_request = LanguageModelRequest {
             thread_id: None,
             prompt_id: None,
+            intent: Some(CompletionIntent::UserPrompt),
             mode: None,
             messages: Vec::new(),
             tools: Vec::new(),

crates/assistant_tools/src/edit_agent.rs

@@ -28,6 +28,7 @@ use std::{cmp, iter, mem, ops::Range, path::PathBuf, pin::Pin, sync::Arc, task::
 use streaming_diff::{CharOperation, StreamingDiff};
 use streaming_fuzzy_matcher::StreamingFuzzyMatcher;
 use util::debug_panic;
+use zed_llm_client::CompletionIntent;
 
 #[derive(Serialize)]
 struct CreateFilePromptTemplate {
@@ -106,7 +107,9 @@ impl EditAgent {
                 edit_description,
             }
             .render(&this.templates)?;
-            let new_chunks = this.request(conversation, prompt, cx).await?;
+            let new_chunks = this
+                .request(conversation, CompletionIntent::CreateFile, prompt, cx)
+                .await?;
 
             let (output, mut inner_events) = this.overwrite_with_chunks(buffer, new_chunks, cx);
             while let Some(event) = inner_events.next().await {
@@ -213,7 +216,9 @@ impl EditAgent {
                 edit_description,
             }
             .render(&this.templates)?;
-            let edit_chunks = this.request(conversation, prompt, cx).await?;
+            let edit_chunks = this
+                .request(conversation, CompletionIntent::EditFile, prompt, cx)
+                .await?;
             this.apply_edit_chunks(buffer, edit_chunks, events_tx, cx)
                 .await
         });
@@ -589,6 +594,7 @@ impl EditAgent {
     async fn request(
         &self,
         mut conversation: LanguageModelRequest,
+        intent: CompletionIntent,
         prompt: String,
         cx: &mut AsyncApp,
     ) -> Result<BoxStream<'static, Result<String, LanguageModelCompletionError>>> {
@@ -646,6 +652,7 @@ impl EditAgent {
         let request = LanguageModelRequest {
             thread_id: conversation.thread_id,
             prompt_id: conversation.prompt_id,
+            intent: Some(intent),
             mode: conversation.mode,
             messages: conversation.messages,
             tool_choice,

crates/eval/Cargo.toml

@@ -67,3 +67,4 @@ unindent.workspace = true
 util.workspace = true
 uuid.workspace = true
 workspace-hack.workspace = true
+zed_llm_client.workspace = true

crates/eval/src/example.rs

@@ -19,6 +19,7 @@ use collections::HashMap;
 use futures::{FutureExt as _, StreamExt, channel::mpsc, select_biased};
 use gpui::{App, AppContext, AsyncApp, Entity};
 use language_model::{LanguageModel, Role, StopReason};
+use zed_llm_client::CompletionIntent;
 
 pub const THREAD_EVENT_TIMEOUT: Duration = Duration::from_secs(60 * 2);
 
@@ -308,7 +309,7 @@ impl ExampleContext {
 
         let message_count_before = self.app.update_entity(&self.agent_thread, |thread, cx| {
             thread.set_remaining_turns(iterations);
-            thread.send_to_model(model, None, cx);
+            thread.send_to_model(model, CompletionIntent::UserPrompt, None, cx);
             thread.messages().len()
         })?;
 

crates/eval/src/instance.rs

@@ -576,6 +576,7 @@ impl ExampleInstance {
                 thread_id: None,
                 prompt_id: None,
                 mode: None,
+                intent: None,
                 messages: vec![LanguageModelRequestMessage {
                     role: Role::User,
                     content: vec![MessageContent::Text(to_prompt(assertion.description))],

crates/git_ui/Cargo.toml

@@ -59,6 +59,7 @@ util.workspace = true
 workspace-hack.workspace = true
 workspace.workspace = true
 zed_actions.workspace = true
+zed_llm_client.workspace = true
 
 [target.'cfg(windows)'.dependencies]
 windows.workspace = true

crates/git_ui/src/git_panel.rs

@@ -13,7 +13,6 @@ use agent_settings::AgentSettings;
 use anyhow::Context as _;
 use askpass::AskPassDelegate;
 use db::kvp::KEY_VALUE_STORE;
-
 use editor::{
     Editor, EditorElement, EditorMode, EditorSettings, MultiBuffer, ShowScrollbar,
     scroll::ScrollbarAutoHide,
@@ -42,6 +41,7 @@ use language_model::{
 };
 use menu::{Confirm, SecondaryConfirm, SelectFirst, SelectLast, SelectNext, SelectPrevious};
 use multi_buffer::ExcerptInfo;
+use notifications::status_toast::{StatusToast, ToastIcon};
 use panel::{
     PanelHeader, panel_button, panel_editor_container, panel_editor_style, panel_filled_button,
     panel_icon_button,
@@ -64,13 +64,12 @@ use ui::{
 };
 use util::{ResultExt, TryFutureExt, maybe};
 use workspace::AppState;
-
-use notifications::status_toast::{StatusToast, ToastIcon};
 use workspace::{
     Workspace,
     dock::{DockPosition, Panel, PanelEvent},
     notifications::DetachAndPromptErr,
 };
+use zed_llm_client::CompletionIntent;
 
 actions!(
     git_panel,
@@ -1767,6 +1766,7 @@ impl GitPanel {
                 let request = LanguageModelRequest {
                     thread_id: None,
                     prompt_id: None,
+                    intent: Some(CompletionIntent::GenerateGitCommitMessage),
                     mode: None,
                     messages: vec![LanguageModelRequestMessage {
                         role: Role::User,

crates/language_model/src/request.rs

@@ -12,7 +12,7 @@ use gpui::{
 use image::codecs::png::PngEncoder;
 use serde::{Deserialize, Serialize};
 use util::ResultExt;
-use zed_llm_client::CompletionMode;
+use zed_llm_client::{CompletionIntent, CompletionMode};
 
 #[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Hash)]
 pub struct LanguageModelImage {
@@ -384,6 +384,7 @@ pub enum LanguageModelToolChoice {
 pub struct LanguageModelRequest {
     pub thread_id: Option<String>,
     pub prompt_id: Option<String>,
+    pub intent: Option<CompletionIntent>,
     pub mode: Option<CompletionMode>,
     pub messages: Vec<LanguageModelRequestMessage>,
     pub tools: Vec<LanguageModelRequestTool>,

crates/language_models/src/provider/cloud.rs

@@ -809,6 +809,7 @@ impl LanguageModel for CloudLanguageModel {
     > {
         let thread_id = request.thread_id.clone();
         let prompt_id = request.prompt_id.clone();
+        let intent = request.intent;
         let mode = request.mode;
         let app_version = cx.update(|cx| AppVersion::global(cx)).ok();
         match self.model.provider {
@@ -841,6 +842,7 @@ impl LanguageModel for CloudLanguageModel {
                         CompletionBody {
                             thread_id,
                             prompt_id,
+                            intent,
                             mode,
                             provider: zed_llm_client::LanguageModelProvider::Anthropic,
                             model: request.model.clone(),
@@ -897,6 +899,7 @@ impl LanguageModel for CloudLanguageModel {
                         CompletionBody {
                             thread_id,
                             prompt_id,
+                            intent,
                             mode,
                             provider: zed_llm_client::LanguageModelProvider::OpenAi,
                             model: request.model.clone(),
@@ -934,6 +937,7 @@ impl LanguageModel for CloudLanguageModel {
                         CompletionBody {
                             thread_id,
                             prompt_id,
+                            intent,
                             mode,
                             provider: zed_llm_client::LanguageModelProvider::Google,
                             model: request.model.model_id.clone(),

crates/language_models/src/provider/mistral.rs

@@ -816,6 +816,7 @@ mod tests {
             tool_choice: None,
             thread_id: None,
             prompt_id: None,
+            intent: None,
             mode: None,
             stop: Vec::new(),
         };

crates/language_models/src/provider/open_ai.rs

@@ -860,6 +860,7 @@ mod tests {
         let request = LanguageModelRequest {
             thread_id: None,
             prompt_id: None,
+            intent: None,
             mode: None,
             messages: vec![LanguageModelRequestMessage {
                 role: Role::User,

crates/rules_library/src/rules_library.rs

@@ -923,6 +923,7 @@ impl RulesLibrary {
                                 LanguageModelRequest {
                                     thread_id: None,
                                     prompt_id: None,
+                                    intent: None,
                                     mode: None,
                                     messages: vec![LanguageModelRequestMessage {
                                         role: Role::System,

crates/semantic_index/src/summary_index.rs

@@ -560,6 +560,7 @@ impl SummaryIndex {
             thread_id: None,
             prompt_id: None,
             mode: None,
+            intent: None,
             messages: vec![LanguageModelRequestMessage {
                 role: Role::User,
                 content: vec![prompt.into()],