Marshall Bowers created
This PR adds a "max mode" toggle to the Agent panel, for models that
support it.
Only visible to folks in the `new-billing` feature flag.
Icon is just a placeholder.
Release Notes:
- N/Acrates/agent/src/active_thread.rs | 1
crates/agent/src/buffer_codegen.rs | 1
crates/agent/src/message_editor.rs | 36 ++++++++++++++++
crates/agent/src/terminal_inline_assistant.rs | 1
crates/agent/src/thread.rs | 20 +++++++++
crates/assistant/src/inline_assistant.rs | 1
crates/assistant/src/terminal_inline_assistant.rs | 1
crates/assistant_context_editor/src/context.rs | 1
crates/eval/src/instance.rs | 1
crates/git_ui/src/git_panel.rs | 1
crates/language_model/src/language_model.rs | 20 +++++++++
crates/language_model/src/request.rs | 2
crates/rules_library/src/rules_library.rs | 1
crates/semantic_index/src/summary_index.rs | 1
14 files changed, 87 insertions(+), 1 deletion(-)
@@ -1293,6 +1293,7 @@ impl ActiveThread {
let request = language_model::LanguageModelRequest {
thread_id: None,
prompt_id: None,
+ mode: None,
messages: vec![request_message],
tools: vec![],
stop: vec![],
@@ -460,6 +460,7 @@ impl CodegenAlternative {
LanguageModelRequest {
thread_id: None,
prompt_id: None,
+ mode: None,
tools: Vec::new(),
stop: Vec::new(),
temperature: None,
@@ -11,6 +11,7 @@ use editor::{
ContextMenuOptions, ContextMenuPlacement, Editor, EditorElement, EditorEvent, EditorMode,
EditorStyle, MultiBuffer,
};
+use feature_flags::{FeatureFlagAppExt, NewBillingFeatureFlag};
use file_icons::FileIcons;
use fs::Fs;
use futures::future::Shared;
@@ -33,6 +34,7 @@ use theme::ThemeSettings;
use ui::{Disclosure, KeyBinding, PopoverMenuHandle, Tooltip, prelude::*};
use util::ResultExt as _;
use workspace::Workspace;
+use zed_llm_client::CompletionMode;
use crate::assistant_model_selector::AssistantModelSelector;
use crate::context_picker::{ContextPicker, ContextPickerCompletionProvider};
@@ -50,7 +52,6 @@ pub struct MessageEditor {
thread: Entity<Thread>,
incompatible_tools_state: Entity<IncompatibleToolsState>,
editor: Entity<Editor>,
- #[allow(dead_code)]
workspace: WeakEntity<Workspace>,
project: Entity<Project>,
context_store: Entity<ContextStore>,
@@ -419,6 +420,37 @@ impl MessageEditor {
}
}
+ fn render_max_mode_toggle(&self, cx: &mut Context<Self>) -> Option<AnyElement> {
+ if !cx.has_flag::<NewBillingFeatureFlag>() {
+ return None;
+ }
+
+ let model = LanguageModelRegistry::read_global(cx)
+ .default_model()
+ .map(|default| default.model.clone())?;
+ if !model.supports_max_mode() {
+ return None;
+ }
+
+ let active_completion_mode = self.thread.read(cx).completion_mode();
+
+ Some(
+ IconButton::new("max-mode", IconName::SquarePlus)
+ .icon_size(IconSize::Small)
+ .toggle_state(active_completion_mode == Some(CompletionMode::Max))
+ .on_click(cx.listener(move |this, _event, _window, cx| {
+ this.thread.update(cx, |thread, _cx| {
+ thread.set_completion_mode(match active_completion_mode {
+ Some(CompletionMode::Max) => Some(CompletionMode::Normal),
+ Some(CompletionMode::Normal) | None => Some(CompletionMode::Max),
+ });
+ });
+ }))
+ .tooltip(Tooltip::text("Max Mode"))
+ .into_any_element(),
+ )
+ }
+
fn render_editor(
&self,
font_size: Rems,
@@ -579,6 +611,7 @@ impl MessageEditor {
}),
)
})
+ .children(self.render_max_mode_toggle(cx))
.child(self.model_selector.clone())
.map({
let focus_handle = focus_handle.clone();
@@ -1100,6 +1133,7 @@ impl MessageEditor {
let request = language_model::LanguageModelRequest {
thread_id: None,
prompt_id: None,
+ mode: None,
messages: vec![request_message],
tools: vec![],
stop: vec![],
@@ -276,6 +276,7 @@ impl TerminalInlineAssistant {
LanguageModelRequest {
thread_id: None,
prompt_id: None,
+ mode: None,
messages: vec![request_message],
tools: Vec::new(),
stop: Vec::new(),
@@ -34,6 +34,7 @@ use settings::Settings;
use thiserror::Error;
use util::{ResultExt as _, TryFutureExt as _, post_inc};
use uuid::Uuid;
+use zed_llm_client::CompletionMode;
use crate::context::{AgentContext, ContextLoadResult, LoadedContext};
use crate::thread_store::{
@@ -290,6 +291,7 @@ pub struct Thread {
summary: Option<SharedString>,
pending_summary: Task<Option<()>>,
detailed_summary_state: DetailedSummaryState,
+ completion_mode: Option<CompletionMode>,
messages: Vec<Message>,
next_message_id: MessageId,
last_prompt_id: PromptId,
@@ -339,6 +341,7 @@ impl Thread {
summary: None,
pending_summary: Task::ready(None),
detailed_summary_state: DetailedSummaryState::NotGenerated,
+ completion_mode: None,
messages: Vec::new(),
next_message_id: MessageId(0),
last_prompt_id: PromptId::new(),
@@ -394,6 +397,7 @@ impl Thread {
summary: Some(serialized.summary),
pending_summary: Task::ready(None),
detailed_summary_state: serialized.detailed_summary_state,
+ completion_mode: None,
messages: serialized
.messages
.into_iter()
@@ -518,6 +522,14 @@ impl Thread {
}
}
+ pub fn completion_mode(&self) -> Option<CompletionMode> {
+ self.completion_mode
+ }
+
+ pub fn set_completion_mode(&mut self, mode: Option<CompletionMode>) {
+ self.completion_mode = mode;
+ }
+
pub fn message(&self, id: MessageId) -> Option<&Message> {
let index = self
.messages
@@ -930,6 +942,12 @@ impl Thread {
self.remaining_turns -= 1;
let mut request = self.to_completion_request(cx);
+ request.mode = if model.supports_max_mode() {
+ self.completion_mode
+ } else {
+ None
+ };
+
if model.supports_tools() {
request.tools = self
.tools()
@@ -967,6 +985,7 @@ impl Thread {
let mut request = LanguageModelRequest {
thread_id: Some(self.id.to_string()),
prompt_id: Some(self.last_prompt_id.to_string()),
+ mode: None,
messages: vec![],
tools: Vec::new(),
stop: Vec::new(),
@@ -1063,6 +1082,7 @@ impl Thread {
let mut request = LanguageModelRequest {
thread_id: None,
prompt_id: None,
+ mode: None,
messages: vec![],
tools: Vec::new(),
stop: Vec::new(),
@@ -2981,6 +2981,7 @@ impl CodegenAlternative {
Ok(LanguageModelRequest {
thread_id: None,
prompt_id: None,
+ mode: None,
messages,
tools: Vec::new(),
stop: Vec::new(),
@@ -294,6 +294,7 @@ impl TerminalInlineAssistant {
Ok(LanguageModelRequest {
thread_id: None,
prompt_id: None,
+ mode: None,
messages,
tools: Vec::new(),
stop: Vec::new(),
@@ -2559,6 +2559,7 @@ impl AssistantContext {
let mut completion_request = LanguageModelRequest {
thread_id: None,
prompt_id: None,
+ mode: None,
messages: Vec::new(),
tools: Vec::new(),
stop: Vec::new(),
@@ -573,6 +573,7 @@ impl ExampleInstance {
let request = LanguageModelRequest {
thread_id: None,
prompt_id: None,
+ mode: None,
messages: vec![LanguageModelRequestMessage {
role: Role::User,
content: vec![MessageContent::Text(to_prompt(assertion.description))],
@@ -1746,6 +1746,7 @@ impl GitPanel {
let request = LanguageModelRequest {
thread_id: None,
prompt_id: None,
+ mode: None,
messages: vec![LanguageModelRequestMessage {
role: Role::User,
content: vec![content.into()],
@@ -240,6 +240,26 @@ pub trait LanguageModel: Send + Sync {
/// Whether this model supports tools.
fn supports_tools(&self) -> bool;
+ /// Returns whether this model supports "max mode";
+ fn supports_max_mode(&self) -> bool {
+ if self.provider_id().0 != ZED_CLOUD_PROVIDER_ID {
+ return false;
+ }
+
+ const MAX_MODE_CAPABLE_MODELS: &[CloudModel] = &[
+ CloudModel::Anthropic(anthropic::Model::Claude3_7Sonnet),
+ CloudModel::Anthropic(anthropic::Model::Claude3_7SonnetThinking),
+ ];
+
+ for model in MAX_MODE_CAPABLE_MODELS {
+ if self.id().0 == model.id() {
+ return true;
+ }
+ }
+
+ false
+ }
+
fn tool_input_format(&self) -> LanguageModelToolSchemaFormat {
LanguageModelToolSchemaFormat::JsonSchema
}
@@ -11,6 +11,7 @@ use gpui::{
use image::{DynamicImage, ImageDecoder, codecs::png::PngEncoder, imageops::resize};
use serde::{Deserialize, Serialize};
use util::ResultExt;
+use zed_llm_client::CompletionMode;
#[derive(Clone, PartialEq, Eq, Serialize, Deserialize, Hash)]
pub struct LanguageModelImage {
@@ -252,6 +253,7 @@ pub struct LanguageModelRequestTool {
pub struct LanguageModelRequest {
pub thread_id: Option<String>,
pub prompt_id: Option<String>,
+ pub mode: Option<CompletionMode>,
pub messages: Vec<LanguageModelRequestMessage>,
pub tools: Vec<LanguageModelRequestTool>,
pub stop: Vec<String>,
@@ -922,6 +922,7 @@ impl RulesLibrary {
LanguageModelRequest {
thread_id: None,
prompt_id: None,
+ mode: None,
messages: vec![LanguageModelRequestMessage {
role: Role::System,
content: vec![body.to_string().into()],
@@ -559,6 +559,7 @@ impl SummaryIndex {
let request = LanguageModelRequest {
thread_id: None,
prompt_id: None,
+ mode: None,
messages: vec![LanguageModelRequestMessage {
role: Role::User,
content: vec![prompt.into()],