diff --git a/crates/editor/src/editor.rs b/crates/editor/src/editor.rs
index 29e009fdf8..06894bf1f9 100644
--- a/crates/editor/src/editor.rs
+++ b/crates/editor/src/editor.rs
@@ -126,7 +126,7 @@ use language::{
     Language, OffsetRangeExt, Point, Runnable, RunnableRange, Selection, SelectionGoal, TextObject,
     TransactionId, TreeSitterOptions, WordsQuery,
     language_settings::{
-        self, InlayHintSettings, LspInsertMode, RewrapBehavior, WordsCompletionMode,
+        self, InlayHintKind, InlayHintSettings, LspInsertMode, RewrapBehavior, WordsCompletionMode,
         all_language_settings, language_settings,
     },
     point_from_lsp, point_to_lsp, text_diff_with_options,
@@ -150,18 +150,18 @@ use project::{
     BreakpointWithPosition, CodeAction, Completion, CompletionIntent, CompletionResponse,
     CompletionSource, DisableAiSettings, DocumentHighlight, InlayHint, Location, LocationLink,
     PrepareRenameResponse, Project, ProjectItem, ProjectPath, ProjectTransaction, TaskSourceKind,
-    debugger::breakpoint_store::Breakpoint,
     debugger::{
         breakpoint_store::{
-            BreakpointEditAction, BreakpointSessionState, BreakpointState, BreakpointStore,
-            BreakpointStoreEvent,
+            Breakpoint, BreakpointEditAction, BreakpointSessionState, BreakpointState,
+            BreakpointStore, BreakpointStoreEvent,
         },
         session::{Session, SessionEvent},
     },
     git_store::{GitStoreEvent, RepositoryEvent},
     lsp_store::{CompletionDocumentation, FormatTrigger, LspFormatTarget, OpenLspBufferHandle},
-    project_settings::{DiagnosticSeverity, GoToDiagnosticSeverityFilter},
-    project_settings::{GitGutterSetting, ProjectSettings},
+    project_settings::{
+        DiagnosticSeverity, GitGutterSetting, GoToDiagnosticSeverityFilter, ProjectSettings,
+    },
 };
 use rand::{seq::SliceRandom, thread_rng};
 use rpc::{ErrorCode, ErrorExt, proto::PeerId};
@@ -1177,9 +1177,137 @@ pub struct Editor {
     selection_drag_state: SelectionDragState,
     next_color_inlay_id: usize,
     colors: Option<LspColorData>,
+    inlay_hints: Option<LspInlayHintData>,
     folding_newlines: Task<()>,
 }
 
+#[derive(Debug)]
+struct LspInlayHintData {
+    enabled: bool,
+    modifiers_override: bool,
+    enabled_in_settings: bool,
+    allowed_hint_kinds: HashSet<Option<InlayHintKind>>,
+    invalidate_debounce: Option<Duration>,
+    append_debounce: Option<Duration>,
+    inlays_for_version: Option<clock::Global>,
+    inlays: HashMap<BufferId, Vec<InlayId>>,
+    inlay_tasks: HashMap<BufferId, HashMap<Range<BufferRow>, Task<()>>>,
+}
+
+impl LspInlayHintData {
+    fn new(settings: InlayHintSettings) -> Self {
+        Self {
+            modifiers_override: false,
+            enabled: settings.enabled,
+            enabled_in_settings: settings.enabled,
+            inlays: HashMap::default(),
+            inlays_for_version: None,
+            inlay_tasks: HashMap::default(),
+            invalidate_debounce: debounce_value(settings.edit_debounce_ms),
+            append_debounce: debounce_value(settings.scroll_debounce_ms),
+            allowed_hint_kinds: settings.enabled_inlay_hint_kinds(),
+        }
+    }
+
+    fn modifiers_override(&mut self, new_override: bool) -> Option<bool> {
+        if self.modifiers_override == new_override {
+            return None;
+        }
+        self.modifiers_override = new_override;
+        if (self.enabled && self.modifiers_override) || (!self.enabled && !self.modifiers_override)
+        {
+            self.clear();
+            Some(false)
+        } else {
+            Some(true)
+        }
+    }
+
+    fn toggle(&mut self, enabled: bool) -> bool {
+        if self.enabled == enabled {
+            return false;
+        }
+        self.enabled = enabled;
+        self.modifiers_override = false;
+        if !enabled {
+            self.clear();
+        }
+        true
+    }
+
+    fn clear(&mut self) {
+        self.inlay_tasks.clear();
+        // TODO kb splice!?
+        self.inlays.clear();
+    }
+
+    /// Checks inlay hint settings for enabled hint kinds and general enabled state.
+    /// Generates corresponding inlay_map splice updates on settings changes.
+    /// Does not update inlay hint cache state on disabling or inlay hint kinds change: only reenabling forces new LSP queries.
+    fn update_settings(
+        &mut self,
+        multi_buffer: &Entity<MultiBuffer>,
+        new_hint_settings: InlayHintSettings,
+        visible_hints: Vec<Inlay>,
+        cx: &mut App,
+    ) -> ControlFlow<Option<InlaySplice>> {
+        let old_enabled = self.enabled;
+        // If the setting for inlay hints has changed, update `enabled`. This condition avoids inlay
+        // hint visibility changes when other settings change (such as theme).
+        //
+        // Another option might be to store whether the user has manually toggled inlay hint
+        // visibility, and prefer this. This could lead to confusion as it means inlay hint
+        // visibility would not change when updating the setting if they were ever toggled.
+        if new_hint_settings.enabled != self.enabled_in_settings {
+            self.enabled = new_hint_settings.enabled;
+            self.enabled_in_settings = new_hint_settings.enabled;
+            self.modifiers_override = false;
+        };
+        self.invalidate_debounce = debounce_value(new_hint_settings.edit_debounce_ms);
+        self.append_debounce = debounce_value(new_hint_settings.scroll_debounce_ms);
+        let new_allowed_hint_kinds = new_hint_settings.enabled_inlay_hint_kinds();
+        match (old_enabled, self.enabled) {
+            (false, false) => {
+                self.allowed_hint_kinds = new_allowed_hint_kinds;
+                ControlFlow::Break(None)
+            }
+            (true, true) => {
+                if new_allowed_hint_kinds == self.allowed_hint_kinds {
+                    ControlFlow::Break(None)
+                } else {
+                    todo!("TODO kb")
+                }
+            }
+            (true, false) => {
+                self.modifiers_override = false;
+                self.allowed_hint_kinds = new_allowed_hint_kinds;
+                if self.inlays.is_empty() {
+                    ControlFlow::Break(None)
+                } else {
+                    self.clear();
+                    ControlFlow::Break(Some(InlaySplice {
+                        to_remove: visible_hints.iter().map(|inlay| inlay.id).collect(),
+                        to_insert: Vec::new(),
+                    }))
+                }
+            }
+            (false, true) => {
+                self.modifiers_override = false;
+                self.allowed_hint_kinds = new_allowed_hint_kinds;
+                ControlFlow::Continue(())
+            }
+        }
+    }
+}
+
+fn debounce_value(debounce_ms: u64) -> Option<Duration> {
+    if debounce_ms > 0 {
+        Some(Duration::from_millis(debounce_ms))
+    } else {
+        None
+    }
+}
+
 #[derive(Copy, Clone, Debug, PartialEq, Eq, Default)]
 enum NextScrollCursorCenterTopBottom {
     #[default]
@@ -1611,7 +1739,7 @@ enum InlayHintRefreshReason {
     SettingsChange(InlayHintSettings),
     NewLinesShown,
     BufferEdited(HashSet<Arc<Language>>),
-    RefreshRequested,
+    RefreshRequested(LanguageServerId),
     ExcerptsRemoved(Vec<ExcerptId>),
 }
 
@@ -1623,7 +1751,7 @@ impl InlayHintRefreshReason {
             Self::SettingsChange(_) => "settings change",
             Self::NewLinesShown => "new lines shown",
             Self::BufferEdited(_) => "buffer edited",
-            Self::RefreshRequested => "refresh requested",
+            Self::RefreshRequested(_) => "refresh requested",
             Self::ExcerptsRemoved(_) => "excerpts removed",
         }
     }
@@ -1864,8 +1992,11 @@ impl Editor {
                     project::Event::RefreshCodeLens => {
                         // we always query lens with actions, without storing them, always refreshing them
                     }
-                    project::Event::RefreshInlayHints => {
-                        editor.refresh_inlay_hints(InlayHintRefreshReason::RefreshRequested, cx);
+                    project::Event::RefreshInlayHints(server_id) => {
+                        editor.refresh_inlay_hints(
+                            InlayHintRefreshReason::RefreshRequested(*server_id),
+                            cx,
+                        );
                     }
                     project::Event::LanguageServerAdded(..)
                     | project::Event::LanguageServerRemoved(..) => {
@@ -2230,6 +2361,7 @@ impl Editor {
             tasks_update_task: None,
             pull_diagnostics_task: Task::ready(()),
             colors: None,
+            inlay_hints: None,
             next_color_inlay_id: 0,
             linked_edit_ranges: Default::default(),
             in_project_search: false,
@@ -2378,6 +2510,7 @@ impl Editor {
             editor.minimap =
                 editor.create_minimap(EditorSettings::get_global(cx).minimap, window, cx);
             editor.colors = Some(LspColorData::new(cx));
+            editor.inlay_hints = Some(LspInlayHintData::new(inlay_hint_settings));
             editor.update_lsp_data(false, None, window, cx);
         }
 
@@ -5229,7 +5362,7 @@ impl Editor {
             InlayHintRefreshReason::BufferEdited(buffer_languages) => {
                 (InvalidationStrategy::BufferEdited, Some(buffer_languages))
             }
-            InlayHintRefreshReason::RefreshRequested => {
+            InlayHintRefreshReason::RefreshRequested(_) => {
                 (InvalidationStrategy::RefreshRequested, None)
             }
         };
@@ -5248,6 +5381,209 @@ impl Editor {
         }
     }
 
+    fn refresh_inlay_hints_2(&mut self, reason: InlayHintRefreshReason, cx: &mut Context<Self>) {
+        if !self.mode.is_full() || self.semantics_provider.is_none() {
+            return;
+        }
+
+        {
+            let Some(inlay_hints) = self.inlay_hints.as_mut() else {
+                return;
+            };
+
+            let reason_description = reason.description();
+            let ignore_debounce = matches!(
+                reason,
+                InlayHintRefreshReason::SettingsChange(_)
+                    | InlayHintRefreshReason::Toggle(_)
+                    | InlayHintRefreshReason::ExcerptsRemoved(_)
+                    | InlayHintRefreshReason::ModifiersChanged(_)
+            );
+            let (invalidate_cache, required_languages) = match reason {
+                InlayHintRefreshReason::ModifiersChanged(enabled) => {
+                    match inlay_hints.modifiers_override(enabled) {
+                        Some(enabled) => {
+                            if enabled {
+                                (InvalidationStrategy::RefreshRequested, None)
+                            } else {
+                                self.splice_inlays(
+                                    &self
+                                        .visible_inlay_hints(cx)
+                                        .iter()
+                                        .map(|inlay| inlay.id)
+                                        .collect::<Vec<InlayId>>(),
+                                    Vec::new(),
+                                    cx,
+                                );
+                                return;
+                            }
+                        }
+                        None => return,
+                    }
+                }
+                InlayHintRefreshReason::Toggle(enabled) => {
+                    if inlay_hints.toggle(enabled) {
+                        if enabled {
+                            (InvalidationStrategy::RefreshRequested, None)
+                        } else {
+                            self.splice_inlays(
+                                &self
+                                    .visible_inlay_hints(cx)
+                                    .iter()
+                                    .map(|inlay| inlay.id)
+                                    .collect::<Vec<InlayId>>(),
+                                Vec::new(),
+                                cx,
+                            );
+                            return;
+                        }
+                    } else {
+                        return;
+                    }
+                }
+                InlayHintRefreshReason::SettingsChange(new_settings) => {
+                    return;
+                }
+                InlayHintRefreshReason::ExcerptsRemoved(excerpts_removed) => {
+                    if let Some(InlaySplice {
+                        to_remove,
+                        to_insert,
+                    }) = self.inlay_hint_cache.remove_excerpts(&excerpts_removed)
+                    {
+                        self.splice_inlays(&to_remove, to_insert, cx);
+                    }
+                    self.display_map.update(cx, |display_map, _| {
+                        display_map.remove_inlays_for_excerpts(&excerpts_removed)
+                    });
+                    return;
+                }
+                InlayHintRefreshReason::NewLinesShown => (InvalidationStrategy::None, None),
+                InlayHintRefreshReason::BufferEdited(buffer_languages) => {
+                    (InvalidationStrategy::BufferEdited, Some(buffer_languages))
+                }
+                InlayHintRefreshReason::RefreshRequested(_) => {
+                    (InvalidationStrategy::RefreshRequested, None)
+                }
+            };
+        }
+
+        let multi_buffer_snapshot = self.buffer().read(cx).snapshot(cx);
+        let Some(semantics_provider) = self.semantics_provider.clone() else {
+            return;
+        };
+        for (excerpt_id, (buffer, buffer_version, range)) in self.visible_excerpts(None, cx) {
+            let Some(excerpt_text_anchor_range) =
+                multi_buffer_snapshot.context_range_for_excerpt(excerpt_id)
+            else {
+                continue;
+            };
+            let buffer_id = buffer.read(cx).remote_id();
+            let buffer_snapshot = buffer.read(cx).snapshot();
+            let buffer_anchor_range =
+                buffer_snapshot.anchor_before(range.start)..buffer_snapshot.anchor_after(range.end);
+
+            let new_hints =
+                semantics_provider.inlay_hints_2(buffer, buffer_anchor_range.clone(), cx);
+            let Some(inlay_hints) = self.inlay_hints.as_mut() else {
+                return;
+            };
+            if let Some((hints_range, new_hints)) = new_hints {
+                let buffer_hints = inlay_hints.inlay_tasks.entry(buffer_id).or_default();
+                buffer_hints.insert(
+                    hints_range.clone(),
+                    cx.spawn(async move |editor, cx| {
+                        let new_hints = new_hints.await;
+                        editor
+                            .update(cx, |editor, cx| {
+                                let multi_buffer_snapshot = editor.buffer.read(cx).snapshot(cx);
+                                let Some(buffer_snapshot) =
+                                    multi_buffer_snapshot.buffer_for_excerpt(excerpt_id)
+                                else {
+                                    return;
+                                };
+
+                                let mut update_data = None;
+                                if let Some(inlay_hints) = editor.inlay_hints.as_mut() {
+                                    let inlay_tasks =
+                                        inlay_hints.inlay_tasks.entry(buffer_id).or_default();
+                                    match new_hints {
+                                        Ok(new_hints) => {
+                                            if inlay_hints.inlays_for_version.as_ref().is_some_and(
+                                                |inlays_for_version| {
+                                                    !inlays_for_version
+                                                        .changed_since(&buffer_version)
+                                                },
+                                            ) {
+                                                let hints_to_remove = if inlay_hints
+                                                    .inlays_for_version
+                                                    .as_ref()
+                                                    .is_some_and(|inlays_for_version| {
+                                                        buffer_version
+                                                            .changed_since(&inlays_for_version)
+                                                    }) {
+                                                    inlay_hints
+                                                        .inlays
+                                                        .remove(&buffer_id)
+                                                        .unwrap_or_default()
+                                                } else {
+                                                    Vec::new()
+                                                };
+                                                let hints_to_insert = new_hints
+                                                    .into_iter()
+                                                    .filter_map(|lsp_hint| {
+                                                        if buffer_anchor_range
+                                                            .start
+                                                            .cmp(
+                                                                &lsp_hint.position,
+                                                                buffer_snapshot,
+                                                            )
+                                                            .is_ge()
+                                                            && buffer_anchor_range
+                                                                .end
+                                                                .cmp(
+                                                                    &lsp_hint.position,
+                                                                    buffer_snapshot,
+                                                                )
+                                                                .is_le()
+                                                        {
+                                                            let position = multi_buffer_snapshot
+                                                                .anchor_in_excerpt(
+                                                                    excerpt_id,
+                                                                    lsp_hint.position,
+                                                                )?;
+                                                            return Some(Inlay::hint(
+                                                                post_inc(&mut editor.next_inlay_id),
+                                                                position,
+                                                                &lsp_hint,
+                                                            ));
+                                                        }
+                                                        None
+                                                    })
+                                                    .collect();
+                                                update_data =
+                                                    Some((hints_to_remove, hints_to_insert));
+                                                inlay_hints.inlays_for_version =
+                                                    Some(buffer_version);
+                                            }
+                                        }
+                                        // TODO kb who should log and clean up the errored state? Could we do that with `lsp_store_cx.spawn`?
+                                        Err(_) => {}
+                                    }
+
+                                    inlay_tasks.remove(&hints_range);
+                                }
+
+                                if let Some((hints_to_remove, hints_to_insert)) = update_data {
+                                    editor.splice_inlays(&hints_to_remove, hints_to_insert, cx);
+                                }
+                            })
+                            .ok();
+                    }),
+                );
+            }
+        }
+    }
+
     fn visible_inlay_hints(&self, cx: &Context<Editor>) -> Vec<Inlay> {
         self.display_map
             .read(cx)
@@ -16751,9 +17087,9 @@ impl Editor {
                         HashSet::default(),
                         cx,
                     );
-                    cx.emit(project::Event::RefreshInlayHints);
                 });
             });
+            self.refresh_inlay_hints(InlayHintRefreshReason::NewLinesShown, cx);
         }
     }
 
@@ -21874,6 +22210,18 @@ pub trait SemanticsProvider {
         cx: &mut App,
     ) -> Option<Task<anyhow::Result<Vec<InlayHint>>>>;
 
+    fn inlay_hints_2(
+        &self,
+        buffer: Entity<Buffer>,
+        range: Range<text::Anchor>,
+        cx: &mut App,
+    ) -> Option<(
+        Range<BufferRow>,
+        Shared<Task<Result<Vec<InlayHint>, Arc<anyhow::Error>>>>,
+    )> {
+        todo!("TODO kb")
+    }
+
     fn resolve_inlay_hint(
         &self,
         hint: InlayHint,
diff --git a/crates/project/src/lsp_store.rs b/crates/project/src/lsp_store.rs
index deebaedd74..c5e0b726d3 100644
--- a/crates/project/src/lsp_store.rs
+++ b/crates/project/src/lsp_store.rs
@@ -14,6 +14,9 @@ pub mod json_language_server_ext;
 pub mod lsp_ext_command;
 pub mod rust_analyzer_ext;
 
+mod inlay_hint_cache;
+
+use self::inlay_hint_cache::BufferInlayHints;
 use crate::{
     CodeAction, ColorPresentation, Completion, CompletionResponse, CompletionSource,
     CoreCompletion, DocumentColor, Hover, InlayHint, LocationLink, LspAction, LspPullDiagnostics,
@@ -564,7 +567,7 @@ impl LocalLspStore {
     }
 
     fn setup_lsp_messages(
-        this: WeakEntity<LspStore>,
+        lsp_store: WeakEntity<LspStore>,
         fs: Arc<dyn Fs>,
         language_server: &LanguageServer,
         delegate: Arc<dyn LspAdapterDelegate>,
@@ -575,7 +578,7 @@ impl LocalLspStore {
         language_server
             .on_notification::<lsp::notification::PublishDiagnostics, _>({
                 let adapter = adapter.clone();
-                let this = this.clone();
+                let this = lsp_store.clone();
                 move |mut params, cx| {
                     let adapter = adapter.clone();
                     if let Some(this) = this.upgrade() {
@@ -619,7 +622,7 @@ impl LocalLspStore {
             .on_request::<lsp::request::WorkspaceConfiguration, _, _>({
                 let adapter = adapter.adapter.clone();
                 let delegate = delegate.clone();
-                let this = this.clone();
+                let this = lsp_store.clone();
                 let fs = fs.clone();
                 move |params, cx| {
                     let adapter = adapter.clone();
@@ -667,7 +670,7 @@ impl LocalLspStore {
 
         language_server
             .on_request::<lsp::request::WorkspaceFoldersRequest, _, _>({
-                let this = this.clone();
+                let this = lsp_store.clone();
                 move |_, cx| {
                     let this = this.clone();
                     let cx = cx.clone();
@@ -695,7 +698,7 @@ impl LocalLspStore {
         // to these requests when initializing.
         language_server
             .on_request::<lsp::request::WorkDoneProgressCreate, _, _>({
-                let this = this.clone();
+                let this = lsp_store.clone();
                 move |params, cx| {
                     let this = this.clone();
                     let mut cx = cx.clone();
@@ -716,7 +719,7 @@ impl LocalLspStore {
 
         language_server
             .on_request::<lsp::request::RegisterCapability, _, _>({
-                let lsp_store = this.clone();
+                let lsp_store = lsp_store.clone();
                 move |params, cx| {
                     let lsp_store = lsp_store.clone();
                     let mut cx = cx.clone();
@@ -745,7 +748,7 @@ impl LocalLspStore {
 
         language_server
             .on_request::<lsp::request::UnregisterCapability, _, _>({
-                let lsp_store = this.clone();
+                let lsp_store = lsp_store.clone();
                 move |params, cx| {
                     let lsp_store = lsp_store.clone();
                     let mut cx = cx.clone();
@@ -774,7 +777,7 @@ impl LocalLspStore {
 
         language_server
             .on_request::<lsp::request::ApplyWorkspaceEdit, _, _>({
-                let this = this.clone();
+                let this = lsp_store.clone();
                 move |params, cx| {
                     let mut cx = cx.clone();
                     let this = this.clone();
@@ -793,18 +796,22 @@ impl LocalLspStore {
 
         language_server
             .on_request::<lsp::request::InlayHintRefreshRequest, _, _>({
-                let this = this.clone();
+                let lsp_store = lsp_store.clone();
                 move |(), cx| {
-                    let this = this.clone();
+                    let this = lsp_store.clone();
                     let mut cx = cx.clone();
                     async move {
-                        this.update(&mut cx, |this, cx| {
-                            cx.emit(LspStoreEvent::RefreshInlayHints);
-                            this.downstream_client.as_ref().map(|(client, project_id)| {
-                                client.send(proto::RefreshInlayHints {
-                                    project_id: *project_id,
+                        this.update(&mut cx, |lsp_store, cx| {
+                            cx.emit(LspStoreEvent::RefreshInlayHints(server_id));
+                            lsp_store
+                                .downstream_client
+                                .as_ref()
+                                .map(|(client, project_id)| {
+                                    client.send(proto::RefreshInlayHints {
+                                        project_id: *project_id,
+                                        server_id: server_id.to_proto(),
+                                    })
                                 })
-                            })
                         })?
                         .transpose()?;
                         Ok(())
@@ -815,7 +822,7 @@ impl LocalLspStore {
 
         language_server
             .on_request::<lsp::request::CodeLensRefresh, _, _>({
-                let this = this.clone();
+                let this = lsp_store.clone();
                 move |(), cx| {
                     let this = this.clone();
                     let mut cx = cx.clone();
@@ -837,7 +844,7 @@ impl LocalLspStore {
 
         language_server
             .on_request::<lsp::request::WorkspaceDiagnosticRefresh, _, _>({
-                let this = this.clone();
+                let this = lsp_store.clone();
                 move |(), cx| {
                     let this = this.clone();
                     let mut cx = cx.clone();
@@ -863,7 +870,7 @@ impl LocalLspStore {
 
         language_server
             .on_request::<lsp::request::ShowMessageRequest, _, _>({
-                let this = this.clone();
+                let this = lsp_store.clone();
                 let name = name.to_string();
                 move |params, cx| {
                     let this = this.clone();
@@ -901,7 +908,7 @@ impl LocalLspStore {
             .detach();
         language_server
             .on_notification::<lsp::notification::ShowMessage, _>({
-                let this = this.clone();
+                let this = lsp_store.clone();
                 let name = name.to_string();
                 move |params, cx| {
                     let this = this.clone();
@@ -933,7 +940,7 @@ impl LocalLspStore {
 
         language_server
             .on_notification::<lsp::notification::Progress, _>({
-                let this = this.clone();
+                let this = lsp_store.clone();
                 move |params, cx| {
                     if let Some(this) = this.upgrade() {
                         this.update(cx, |this, cx| {
@@ -952,7 +959,7 @@ impl LocalLspStore {
 
         language_server
             .on_notification::<lsp::notification::LogMessage, _>({
-                let this = this.clone();
+                let this = lsp_store.clone();
                 move |params, cx| {
                     if let Some(this) = this.upgrade() {
                         this.update(cx, |_, cx| {
@@ -970,7 +977,7 @@ impl LocalLspStore {
 
         language_server
             .on_notification::<lsp::notification::LogTrace, _>({
-                let this = this.clone();
+                let this = lsp_store.clone();
                 move |params, cx| {
                     let mut cx = cx.clone();
                     if let Some(this) = this.upgrade() {
@@ -987,9 +994,9 @@ impl LocalLspStore {
             })
             .detach();
 
-        json_language_server_ext::register_requests(this.clone(), language_server);
-        rust_analyzer_ext::register_notifications(this.clone(), language_server);
-        clangd_ext::register_notifications(this, language_server, adapter);
+        json_language_server_ext::register_requests(lsp_store.clone(), language_server);
+        rust_analyzer_ext::register_notifications(lsp_store.clone(), language_server);
+        clangd_ext::register_notifications(lsp_store, language_server, adapter);
     }
 
     fn shutdown_language_servers_on_quit(
@@ -3491,6 +3498,7 @@ pub struct LspStore {
     pub(super) lsp_server_capabilities: HashMap<LanguageServerId, lsp::ServerCapabilities>,
     lsp_document_colors: HashMap<BufferId, DocumentColorData>,
     lsp_code_lens: HashMap<BufferId, CodeLensData>,
+    inlay_hint_data: HashMap<BufferId, BufferInlayHints>,
     running_lsp_requests: HashMap<TypeId, (Global, HashMap<LspRequestId, Task<()>>)>,
 }
 
@@ -3540,7 +3548,7 @@ pub enum LspStoreEvent {
         new_language: Option<Arc<Language>>,
     },
     Notification(String),
-    RefreshInlayHints,
+    RefreshInlayHints(LanguageServerId),
     RefreshCodeLens,
     DiagnosticsUpdated {
         server_id: LanguageServerId,
@@ -3762,6 +3770,7 @@ impl LspStore {
             lsp_server_capabilities: HashMap::default(),
             lsp_document_colors: HashMap::default(),
             lsp_code_lens: HashMap::default(),
+            inlay_hint_data: HashMap::default(),
             running_lsp_requests: HashMap::default(),
             active_entry: None,
             _maintain_workspace_config,
@@ -3824,6 +3833,7 @@ impl LspStore {
             lsp_server_capabilities: HashMap::default(),
             lsp_document_colors: HashMap::default(),
             lsp_code_lens: HashMap::default(),
+            inlay_hint_data: HashMap::default(),
             running_lsp_requests: HashMap::default(),
             active_entry: None,
 
@@ -4125,6 +4135,7 @@ impl LspStore {
                     if refcount == 0 {
                         lsp_store.lsp_document_colors.remove(&buffer_id);
                         lsp_store.lsp_code_lens.remove(&buffer_id);
+                        lsp_store.inlay_hint_data.remove(&buffer_id);
                         let local = lsp_store.as_local_mut().unwrap();
                         local.registered_buffers.remove(&buffer_id);
                         local.buffers_opened_in_servers.remove(&buffer_id);
@@ -9524,7 +9535,7 @@ impl LspStore {
             if let Some(work) = status.pending_work.remove(&token)
                 && !work.is_disk_based_diagnostics_progress
             {
-                cx.emit(LspStoreEvent::RefreshInlayHints);
+                cx.emit(LspStoreEvent::RefreshInlayHints(language_server_id));
             }
             cx.notify();
         }
@@ -9656,12 +9667,14 @@ impl LspStore {
     }
 
     async fn handle_refresh_inlay_hints(
-        this: Entity<Self>,
-        _: TypedEnvelope<proto::RefreshInlayHints>,
+        lsp_store: Entity<Self>,
+        envelope: TypedEnvelope<proto::RefreshInlayHints>,
         mut cx: AsyncApp,
     ) -> Result<proto::Ack> {
-        this.update(&mut cx, |_, cx| {
-            cx.emit(LspStoreEvent::RefreshInlayHints);
+        lsp_store.update(&mut cx, |_, cx| {
+            cx.emit(LspStoreEvent::RefreshInlayHints(
+                LanguageServerId::from_proto(envelope.payload.server_id),
+            ));
         })?;
         Ok(proto::Ack {})
     }
@@ -10913,7 +10926,7 @@ impl LspStore {
             language_server.name(),
             Some(key.worktree_id),
         ));
-        cx.emit(LspStoreEvent::RefreshInlayHints);
+        cx.emit(LspStoreEvent::RefreshInlayHints(server_id));
 
         let server_capabilities = language_server.capabilities();
         if let Some((downstream_client, project_id)) = self.downstream_client.as_ref() {
@@ -11517,6 +11530,10 @@ impl LspStore {
                 buffer_servers.remove(&for_server);
             }
         }
+
+        for inlay_hint_cache in self.inlay_hint_data.values_mut() {
+            inlay_hint_cache.remove_server_data(for_server);
+        }
     }
 
     pub fn result_id(
diff --git a/crates/project/src/lsp_store/inlay_hint_cache.rs b/crates/project/src/lsp_store/inlay_hint_cache.rs
new file mode 100644
index 0000000000..d69c4c3038
--- /dev/null
+++ b/crates/project/src/lsp_store/inlay_hint_cache.rs
@@ -0,0 +1,57 @@
+use std::{collections::BTreeMap, ops::Range};
+
+use clock::Global;
+use collections::HashMap;
+use futures::future::Shared;
+use gpui::{Context, Entity, Task};
+use language::BufferRow;
+use lsp::LanguageServerId;
+use text::BufferId;
+
+use crate::{InlayHint, buffer_store::BufferStore};
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
+pub struct InlayHintId(usize);
+
+#[derive(Debug, Default)]
+pub struct BufferInlayHints {
+    all_hints: HashMap<InlayHintId, InlayHint>,
+    hints: HashMap<LanguageServerId, HintChunks>,
+    chunks_for_version: Global,
+    cache_version: usize,
+}
+
+#[derive(Debug, Default)]
+struct HintChunks {
+    hints_by_chunks: BTreeMap<Range<BufferRow>, Option<Vec<InlayHintId>>>,
+    chunk_updates: HashMap<Range<BufferRow>, Shared<Task<Vec<InlayHint>>>>,
+}
+
+#[derive(Debug, PartialEq, Eq, Clone, Copy)]
+pub enum HintFetchStrategy {
+    IgnoreCache,
+    UseCache { known_cache_version: Option<usize> },
+}
+
+impl BufferInlayHints {
+    pub fn remove_server_data(&mut self, for_server: LanguageServerId) -> bool {
+        let removed = self.hints.remove(&for_server).is_some();
+        if removed {
+            self.cache_version += 1;
+        }
+        removed
+    }
+
+    pub fn hints(
+        &self,
+        buffer_store: Entity<BufferStore>,
+        buffer: BufferId,
+        strategy: HintFetchStrategy,
+        range: impl text::ToOffset,
+        cx: &mut Context<Self>,
+    ) -> Option<(Range<BufferRow>, Shared<Task<Vec<InlayHint>>>)> {
+        todo!("TODO kb")
+    }
+    // we want to store the cache version outbound, so they can query with it: we can return nothing (`Option`) if the version matches
+    // we can get a new server up/down, so we want to re-query for them things, ignoring the cache version
+}
diff --git a/crates/project/src/project.rs b/crates/project/src/project.rs
index 9fd4eed641..8f4417774c 100644
--- a/crates/project/src/project.rs
+++ b/crates/project/src/project.rs
@@ -322,7 +322,7 @@ pub enum Event {
     HostReshared,
     Reshared,
     Rejoined,
-    RefreshInlayHints,
+    RefreshInlayHints(LanguageServerId),
     RefreshCodeLens,
     RevealInProjectPanel(ProjectEntryId),
     SnippetEdit(BufferId, Vec<(lsp::Range, Snippet)>),
@@ -2922,7 +2922,9 @@ impl Project {
                     return;
                 };
             }
-            LspStoreEvent::RefreshInlayHints => cx.emit(Event::RefreshInlayHints),
+            LspStoreEvent::RefreshInlayHints(server_id) => {
+                cx.emit(Event::RefreshInlayHints(*server_id))
+            }
             LspStoreEvent::RefreshCodeLens => cx.emit(Event::RefreshCodeLens),
             LspStoreEvent::LanguageServerPrompt(prompt) => {
                 cx.emit(Event::LanguageServerPrompt(prompt.clone()))
diff --git a/crates/project/src/project_tests.rs b/crates/project/src/project_tests.rs
index 6dcd07482e..b4f87c4afa 100644
--- a/crates/project/src/project_tests.rs
+++ b/crates/project/src/project_tests.rs
@@ -1804,7 +1804,10 @@ async fn test_disk_based_diagnostics_progress(cx: &mut gpui::TestAppContext) {
     fake_server
         .start_progress(format!("{}/0", progress_token))
         .await;
-    assert_eq!(events.next().await.unwrap(), Event::RefreshInlayHints);
+    assert_eq!(
+        events.next().await.unwrap(),
+        Event::RefreshInlayHints(fake_server.server.server_id())
+    );
     assert_eq!(
         events.next().await.unwrap(),
         Event::DiskBasedDiagnosticsStarted {
@@ -1943,7 +1946,10 @@ async fn test_restarting_server_with_diagnostics_running(cx: &mut gpui::TestAppC
             Some(worktree_id)
         )
     );
-    assert_eq!(events.next().await.unwrap(), Event::RefreshInlayHints);
+    assert_eq!(
+        events.next().await.unwrap(),
+        Event::RefreshInlayHints(fake_server.server.server_id())
+    );
     fake_server.start_progress(progress_token).await;
     assert_eq!(
         events.next().await.unwrap(),
diff --git a/crates/proto/proto/lsp.proto b/crates/proto/proto/lsp.proto
index 473ef5c38c..97b07ecf76 100644
--- a/crates/proto/proto/lsp.proto
+++ b/crates/proto/proto/lsp.proto
@@ -465,6 +465,7 @@ message ResolveInlayHintResponse {
 
 message RefreshInlayHints {
     uint64 project_id = 1;
+    uint64 server_id = 2;
 }
 
 message CodeLens {