agent: Show an error when the model requests limit has been reached (#28868)

This PR adds an error message when the model requests limit has been
hit.

Release Notes:

- N/A

Co-authored-by: Oleksiy Syvokon <oleksiy.syvokon@gmail.com>

crates/agent/src/assistant_panel.rs

@@ -25,6 +25,7 @@ use language_model_selector::ToggleModelSelector;
 use project::Project;
 use prompt_library::{PromptLibrary, open_prompt_library};
 use prompt_store::PromptBuilder;
+use proto::Plan;
 use settings::{Settings, update_settings_file};
 use time::UtcOffset;
 use ui::{
@@ -1449,6 +1450,9 @@ impl AssistantPanel {
                     ThreadError::MaxMonthlySpendReached => {
                         self.render_max_monthly_spend_reached_error(cx)
                     }
+                    ThreadError::ModelRequestLimitReached { plan } => {
+                        self.render_model_request_limit_reached_error(plan, cx)
+                    }
                     ThreadError::Message { header, message } => {
                         self.render_error_message(header, message, cx)
                     }
@@ -1551,6 +1555,67 @@ impl AssistantPanel {
             .into_any()
     }
 
+    fn render_model_request_limit_reached_error(
+        &self,
+        plan: Plan,
+        cx: &mut Context<Self>,
+    ) -> AnyElement {
+        let error_message = match plan {
+            Plan::Free => "Model request limit reached. Upgrade to Zed Pro for more requests.",
+            Plan::ZedPro => {
+                "Model request limit reached. Upgrade to usage-based billing for more requests."
+            }
+        };
+        let call_to_action = match plan {
+            Plan::Free => "Upgrade to Zed Pro",
+            Plan::ZedPro => "Upgrade to usage-based billing",
+        };
+
+        v_flex()
+            .gap_0p5()
+            .child(
+                h_flex()
+                    .gap_1p5()
+                    .items_center()
+                    .child(Icon::new(IconName::XCircle).color(Color::Error))
+                    .child(Label::new("Model Request Limit Reached").weight(FontWeight::MEDIUM)),
+            )
+            .child(
+                div()
+                    .id("error-message")
+                    .max_h_24()
+                    .overflow_y_scroll()
+                    .child(Label::new(error_message)),
+            )
+            .child(
+                h_flex()
+                    .justify_end()
+                    .mt_1()
+                    .child(
+                        Button::new("subscribe", call_to_action).on_click(cx.listener(
+                            |this, _, _, cx| {
+                                this.thread.update(cx, |this, _cx| {
+                                    this.clear_last_error();
+                                });
+
+                                cx.open_url(&zed_urls::account_url(cx));
+                                cx.notify();
+                            },
+                        )),
+                    )
+                    .child(Button::new("dismiss", "Dismiss").on_click(cx.listener(
+                        |this, _, _, cx| {
+                            this.thread.update(cx, |this, _cx| {
+                                this.clear_last_error();
+                            });
+
+                            cx.notify();
+                        },
+                    ))),
+            )
+            .into_any()
+    }
+
     fn render_error_message(
         &self,
         header: SharedString,

crates/agent/src/thread.rs

@@ -18,12 +18,13 @@ use language_model::{
     ConfiguredModel, LanguageModel, LanguageModelCompletionEvent, LanguageModelId,
     LanguageModelKnownError, LanguageModelRegistry, LanguageModelRequest,
     LanguageModelRequestMessage, LanguageModelRequestTool, LanguageModelToolResult,
-    LanguageModelToolUseId, MaxMonthlySpendReachedError, MessageContent, PaymentRequiredError,
-    Role, StopReason, TokenUsage,
+    LanguageModelToolUseId, MaxMonthlySpendReachedError, MessageContent,
+    ModelRequestLimitReachedError, PaymentRequiredError, Role, StopReason, TokenUsage,
 };
 use project::Project;
 use project::git_store::{GitStore, GitStoreCheckpoint, RepositoryState};
 use prompt_store::PromptBuilder;
+use proto::Plan;
 use schemars::JsonSchema;
 use serde::{Deserialize, Serialize};
 use settings::Settings;
@@ -1150,6 +1151,12 @@ impl Thread {
                                 cx.emit(ThreadEvent::ShowError(
                                     ThreadError::MaxMonthlySpendReached,
                                 ));
+                            } else if let Some(error) =
+                                error.downcast_ref::<ModelRequestLimitReachedError>()
+                            {
+                                cx.emit(ThreadEvent::ShowError(
+                                    ThreadError::ModelRequestLimitReached { plan: error.plan },
+                                ));
                             } else if let Some(known_error) =
                                 error.downcast_ref::<LanguageModelKnownError>()
                             {
@@ -1929,6 +1936,8 @@ pub enum ThreadError {
     PaymentRequired,
     #[error("Max monthly spend reached")]
     MaxMonthlySpendReached,
+    #[error("Model request limit reached")]
+    ModelRequestLimitReached { plan: Plan },
     #[error("Message {header}: {message}")]
     Message {
         header: SharedString,

crates/language_model/src/model/cloud_model.rs

@@ -142,6 +142,24 @@ impl fmt::Display for MaxMonthlySpendReachedError {
     }
 }
 
+#[derive(Error, Debug)]
+pub struct ModelRequestLimitReachedError {
+    pub plan: Plan,
+}
+
+impl fmt::Display for ModelRequestLimitReachedError {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        let message = match self.plan {
+            Plan::Free => "Model request limit reached. Upgrade to Zed Pro for more requests.",
+            Plan::ZedPro => {
+                "Model request limit reached. Upgrade to usage-based billing for more requests."
+            }
+        };
+
+        write!(f, "{message}")
+    }
+}
+
 #[derive(Clone, Default)]
 pub struct LlmApiToken(Arc<RwLock<Option<String>>>);
 

crates/language_models/Cargo.toml

@@ -53,6 +53,7 @@ tokio = { workspace = true, features = ["rt", "rt-multi-thread"] }
 ui.workspace = true
 util.workspace = true
 workspace-hack.workspace = true
+zed_llm_client.workspace = true
 
 [dev-dependencies]
 editor = { workspace = true, features = ["test-support"] }

crates/language_models/src/provider/cloud.rs

@@ -16,18 +16,21 @@ use language_model::{
     AuthenticateError, CloudModel, LanguageModel, LanguageModelCacheConfiguration, LanguageModelId,
     LanguageModelKnownError, LanguageModelName, LanguageModelProviderId, LanguageModelProviderName,
     LanguageModelProviderState, LanguageModelProviderTosView, LanguageModelRequest,
-    LanguageModelToolSchemaFormat, RateLimiter, ZED_CLOUD_PROVIDER_ID,
+    LanguageModelToolSchemaFormat, ModelRequestLimitReachedError, RateLimiter,
+    ZED_CLOUD_PROVIDER_ID,
 };
 use language_model::{
     LanguageModelAvailability, LanguageModelCompletionEvent, LanguageModelProvider, LlmApiToken,
     MaxMonthlySpendReachedError, PaymentRequiredError, RefreshLlmTokenListener,
 };
+use proto::Plan;
 use schemars::JsonSchema;
 use serde::{Deserialize, Serialize, de::DeserializeOwned};
 use serde_json::value::RawValue;
 use settings::{Settings, SettingsStore};
 use smol::Timer;
 use smol::io::{AsyncReadExt, BufReader};
+use std::str::FromStr as _;
 use std::{
     sync::{Arc, LazyLock},
     time::Duration,
@@ -35,6 +38,7 @@ use std::{
 use strum::IntoEnumIterator;
 use thiserror::Error;
 use ui::{TintColor, prelude::*};
+use zed_llm_client::{CURRENT_PLAN_HEADER_NAME, SUBSCRIPTION_LIMIT_RESOURCE_HEADER_NAME};
 
 use crate::AllLanguageModelSettings;
 use crate::provider::anthropic::{count_anthropic_tokens, into_anthropic};
@@ -551,6 +555,32 @@ impl CloudLanguageModel {
                     .is_some()
             {
                 return Err(anyhow!(MaxMonthlySpendReachedError));
+            } else if status == StatusCode::FORBIDDEN
+                && response
+                    .headers()
+                    .get(SUBSCRIPTION_LIMIT_RESOURCE_HEADER_NAME)
+                    .is_some()
+            {
+                if let Some("model_requests") = response
+                    .headers()
+                    .get(SUBSCRIPTION_LIMIT_RESOURCE_HEADER_NAME)
+                    .and_then(|resource| resource.to_str().ok())
+                {
+                    if let Some(plan) = response
+                        .headers()
+                        .get(CURRENT_PLAN_HEADER_NAME)
+                        .and_then(|plan| plan.to_str().ok())
+                        .and_then(|plan| zed_llm_client::Plan::from_str(plan).ok())
+                    {
+                        let plan = match plan {
+                            zed_llm_client::Plan::Free => Plan::Free,
+                            zed_llm_client::Plan::ZedPro => Plan::ZedPro,
+                        };
+                        return Err(anyhow!(ModelRequestLimitReachedError { plan }));
+                    }
+                }
+
+                return Err(anyhow!("Forbidden"));
             } else if status.as_u16() >= 500 && status.as_u16() < 600 {
                 // If we encounter an error in the 500 range, retry after a delay.
                 // We've seen at least these in the wild from API providers: