diff --git a/crates/assistant/src/assistant_panel.rs b/crates/assistant/src/assistant_panel.rs
index 5d9e2e84a3..ae39254980 100644
--- a/crates/assistant/src/assistant_panel.rs
+++ b/crates/assistant/src/assistant_panel.rs
@@ -58,6 +58,7 @@ use std::{
     time::Duration,
 };
 use terminal_view::{terminal_panel::TerminalPanel, TerminalView};
+use ui::TintColor;
 use ui::{
     prelude::*,
     utils::{format_distance_from_now, DateTimeType},
@@ -2320,8 +2321,34 @@ impl ContextEditor {
             },
             None => "Send",
         };
+
+        let (style, tooltip) = match token_state(&self.context, cx) {
+            Some(TokenState::NoTokensLeft { .. }) => (
+                ButtonStyle::Tinted(TintColor::Negative),
+                Some(Tooltip::text("Token limit reached", cx)),
+            ),
+            Some(TokenState::HasMoreTokens {
+                over_warn_threshold,
+                ..
+            }) => {
+                let (style, tooltip) = if over_warn_threshold {
+                    (
+                        ButtonStyle::Tinted(TintColor::Warning),
+                        Some(Tooltip::text("Token limit is close to exhaustion", cx)),
+                    )
+                } else {
+                    (ButtonStyle::Filled, None)
+                };
+                (style, tooltip)
+            }
+            None => (ButtonStyle::Filled, None),
+        };
+
         ButtonLike::new("send_button")
-            .style(ButtonStyle::Filled)
+            .style(style)
+            .when_some(tooltip, |button, tooltip| {
+                button.tooltip(move |_| tooltip.clone())
+            })
             .layer(ElevationIndex::ModalSurface)
             .children(
                 KeyBinding::for_action_in(&Assist, &focus_handle, cx)
@@ -2726,25 +2753,30 @@ impl ContextEditorToolbarItem {
     }
 
     fn render_remaining_tokens(&self, cx: &mut ViewContext<Self>) -> Option<impl IntoElement> {
-        let model = LanguageModelRegistry::read_global(cx).active_model()?;
         let context = &self
             .active_context_editor
             .as_ref()?
             .upgrade()?
             .read(cx)
             .context;
-        let token_count = context.read(cx).token_count()?;
-        let max_token_count = model.max_token_count();
-
-        let remaining_tokens = max_token_count as isize - token_count as isize;
-        let token_count_color = if remaining_tokens <= 0 {
-            Color::Error
-        } else if token_count as f32 / max_token_count as f32 >= 0.8 {
-            Color::Warning
-        } else {
-            Color::Muted
+        let (token_count_color, token_count, max_token_count) = match token_state(context, cx)? {
+            TokenState::NoTokensLeft {
+                max_token_count,
+                token_count,
+            } => (Color::Error, token_count, max_token_count),
+            TokenState::HasMoreTokens {
+                max_token_count,
+                token_count,
+                over_warn_threshold,
+            } => {
+                let color = if over_warn_threshold {
+                    Color::Warning
+                } else {
+                    Color::Muted
+                };
+                (color, token_count, max_token_count)
+            }
         };
-
         Some(
             h_flex()
                 .gap_0p5()
@@ -3077,3 +3109,40 @@ fn slash_command_error_block_renderer(message: String) -> RenderBlock {
             .into_any()
     })
 }
+
+enum TokenState {
+    NoTokensLeft {
+        max_token_count: usize,
+        token_count: usize,
+    },
+    HasMoreTokens {
+        max_token_count: usize,
+        token_count: usize,
+        over_warn_threshold: bool,
+    },
+}
+
+fn token_state(context: &Model<Context>, cx: &AppContext) -> Option<TokenState> {
+    const WARNING_TOKEN_THRESHOLD: f32 = 0.8;
+
+    let model = LanguageModelRegistry::read_global(cx).active_model()?;
+    let token_count = context.read(cx).token_count()?;
+    let max_token_count = model.max_token_count();
+
+    let remaining_tokens = max_token_count as isize - token_count as isize;
+    let token_state = if remaining_tokens <= 0 {
+        TokenState::NoTokensLeft {
+            max_token_count,
+            token_count,
+        }
+    } else {
+        let over_warn_threshold =
+            token_count as f32 / max_token_count as f32 >= WARNING_TOKEN_THRESHOLD;
+        TokenState::HasMoreTokens {
+            max_token_count,
+            token_count,
+            over_warn_threshold,
+        }
+    };
+    Some(token_state)
+}