1use anyhow::{Context as _, Result, anyhow};
2use collections::BTreeMap;
3use credentials_provider::CredentialsProvider;
4use editor::{Editor, EditorElement, EditorStyle};
5use futures::{FutureExt, Stream, StreamExt, future, future::BoxFuture};
6use google_ai::{
7 FunctionDeclaration, GenerateContentResponse, GoogleModelMode, Part, SystemInstruction,
8 ThinkingConfig, UsageMetadata,
9};
10use gpui::{
11 AnyView, App, AsyncApp, Context, Entity, FontStyle, SharedString, Task, TextStyle, WhiteSpace,
12 Window,
13};
14use http_client::HttpClient;
15use language_model::{
16 AuthenticateError, ConfigurationViewTargetAgent, LanguageModelCompletionError,
17 LanguageModelCompletionEvent, LanguageModelToolChoice, LanguageModelToolSchemaFormat,
18 LanguageModelToolUse, LanguageModelToolUseId, MessageContent, StopReason,
19};
20use language_model::{
21 LanguageModel, LanguageModelId, LanguageModelName, LanguageModelProvider,
22 LanguageModelProviderId, LanguageModelProviderName, LanguageModelProviderState,
23 LanguageModelRequest, RateLimiter, Role,
24};
25use schemars::JsonSchema;
26use serde::{Deserialize, Serialize};
27use settings::{Settings, SettingsStore};
28use std::pin::Pin;
29use std::sync::{
30 Arc, LazyLock,
31 atomic::{self, AtomicU64},
32};
33use strum::IntoEnumIterator;
34use theme::ThemeSettings;
35use ui::{Icon, IconName, List, Tooltip, prelude::*};
36use util::{ResultExt, truncate_and_trailoff};
37use zed_env_vars::EnvVar;
38
39use crate::api_key::ApiKey;
40use crate::api_key::ApiKeyState;
41use crate::ui::InstructionListItem;
42
43const PROVIDER_ID: LanguageModelProviderId = language_model::GOOGLE_PROVIDER_ID;
44const PROVIDER_NAME: LanguageModelProviderName = language_model::GOOGLE_PROVIDER_NAME;
45
46#[derive(Default, Clone, Debug, PartialEq)]
47pub struct GoogleSettings {
48 pub api_url: String,
49 pub available_models: Vec<AvailableModel>,
50}
51
52#[derive(Clone, Copy, Debug, Default, PartialEq, Serialize, Deserialize, JsonSchema)]
53#[serde(tag = "type", rename_all = "lowercase")]
54pub enum ModelMode {
55 #[default]
56 Default,
57 Thinking {
58 /// The maximum number of tokens to use for reasoning. Must be lower than the model's `max_output_tokens`.
59 budget_tokens: Option<u32>,
60 },
61}
62
63impl From<ModelMode> for GoogleModelMode {
64 fn from(value: ModelMode) -> Self {
65 match value {
66 ModelMode::Default => GoogleModelMode::Default,
67 ModelMode::Thinking { budget_tokens } => GoogleModelMode::Thinking { budget_tokens },
68 }
69 }
70}
71
72impl From<GoogleModelMode> for ModelMode {
73 fn from(value: GoogleModelMode) -> Self {
74 match value {
75 GoogleModelMode::Default => ModelMode::Default,
76 GoogleModelMode::Thinking { budget_tokens } => ModelMode::Thinking { budget_tokens },
77 }
78 }
79}
80
81#[derive(Clone, Debug, PartialEq, Serialize, Deserialize, JsonSchema)]
82pub struct AvailableModel {
83 name: String,
84 display_name: Option<String>,
85 max_tokens: u64,
86 mode: Option<ModelMode>,
87}
88
89pub struct GoogleLanguageModelProvider {
90 http_client: Arc<dyn HttpClient>,
91 state: gpui::Entity<State>,
92}
93
94pub struct State {
95 api_key_state: ApiKeyState,
96}
97
98const GEMINI_API_KEY_VAR_NAME: &str = "GEMINI_API_KEY";
99const GOOGLE_AI_API_KEY_VAR_NAME: &str = "GOOGLE_AI_API_KEY";
100
101static API_KEY_ENV_VAR: LazyLock<EnvVar> = LazyLock::new(|| {
102 // Try GEMINI_API_KEY first as primary, fallback to GOOGLE_AI_API_KEY
103 EnvVar::new(GEMINI_API_KEY_VAR_NAME.into()).or(EnvVar::new(GOOGLE_AI_API_KEY_VAR_NAME.into()))
104});
105
106impl State {
107 fn is_authenticated(&self) -> bool {
108 self.api_key_state.has_key()
109 }
110
111 fn set_api_key(&mut self, api_key: Option<String>, cx: &mut Context<Self>) -> Task<Result<()>> {
112 let api_url = GoogleLanguageModelProvider::api_url(cx);
113 self.api_key_state
114 .store(api_url, api_key, |this| &mut this.api_key_state, cx)
115 }
116
117 fn authenticate(&mut self, cx: &mut Context<Self>) -> Task<Result<(), AuthenticateError>> {
118 let api_url = GoogleLanguageModelProvider::api_url(cx);
119 self.api_key_state.load_if_needed(
120 api_url,
121 &API_KEY_ENV_VAR,
122 |this| &mut this.api_key_state,
123 cx,
124 )
125 }
126}
127
128impl GoogleLanguageModelProvider {
129 pub fn new(http_client: Arc<dyn HttpClient>, cx: &mut App) -> Self {
130 let state = cx.new(|cx| {
131 cx.observe_global::<SettingsStore>(|this: &mut State, cx| {
132 let api_url = Self::api_url(cx);
133 this.api_key_state.handle_url_change(
134 api_url,
135 &API_KEY_ENV_VAR,
136 |this| &mut this.api_key_state,
137 cx,
138 );
139 cx.notify();
140 })
141 .detach();
142 State {
143 api_key_state: ApiKeyState::new(Self::api_url(cx)),
144 }
145 });
146
147 Self { http_client, state }
148 }
149
150 fn create_language_model(&self, model: google_ai::Model) -> Arc<dyn LanguageModel> {
151 Arc::new(GoogleLanguageModel {
152 id: LanguageModelId::from(model.id().to_string()),
153 model,
154 state: self.state.clone(),
155 http_client: self.http_client.clone(),
156 request_limiter: RateLimiter::new(4),
157 })
158 }
159
160 pub fn api_key_for_gemini_cli(cx: &mut App) -> Task<Result<String>> {
161 if let Some(key) = API_KEY_ENV_VAR.value.clone() {
162 return Task::ready(Ok(key));
163 }
164 let credentials_provider = <dyn CredentialsProvider>::global(cx);
165 let api_url = Self::api_url(cx).to_string();
166 cx.spawn(async move |cx| {
167 Ok(
168 ApiKey::load_from_system_keychain(&api_url, credentials_provider.as_ref(), cx)
169 .await?
170 .key()
171 .to_string(),
172 )
173 })
174 }
175
176 fn settings(cx: &App) -> &GoogleSettings {
177 &crate::AllLanguageModelSettings::get_global(cx).google
178 }
179
180 fn api_url(cx: &App) -> SharedString {
181 let api_url = &Self::settings(cx).api_url;
182 if api_url.is_empty() {
183 google_ai::API_URL.into()
184 } else {
185 SharedString::new(api_url.as_str())
186 }
187 }
188}
189
190impl LanguageModelProviderState for GoogleLanguageModelProvider {
191 type ObservableEntity = State;
192
193 fn observable_entity(&self) -> Option<gpui::Entity<Self::ObservableEntity>> {
194 Some(self.state.clone())
195 }
196}
197
198impl LanguageModelProvider for GoogleLanguageModelProvider {
199 fn id(&self) -> LanguageModelProviderId {
200 PROVIDER_ID
201 }
202
203 fn name(&self) -> LanguageModelProviderName {
204 PROVIDER_NAME
205 }
206
207 fn icon(&self) -> IconName {
208 IconName::AiGoogle
209 }
210
211 fn default_model(&self, _cx: &App) -> Option<Arc<dyn LanguageModel>> {
212 Some(self.create_language_model(google_ai::Model::default()))
213 }
214
215 fn default_fast_model(&self, _cx: &App) -> Option<Arc<dyn LanguageModel>> {
216 Some(self.create_language_model(google_ai::Model::default_fast()))
217 }
218
219 fn provided_models(&self, cx: &App) -> Vec<Arc<dyn LanguageModel>> {
220 let mut models = BTreeMap::default();
221
222 // Add base models from google_ai::Model::iter()
223 for model in google_ai::Model::iter() {
224 if !matches!(model, google_ai::Model::Custom { .. }) {
225 models.insert(model.id().to_string(), model);
226 }
227 }
228
229 // Override with available models from settings
230 for model in &GoogleLanguageModelProvider::settings(cx).available_models {
231 models.insert(
232 model.name.clone(),
233 google_ai::Model::Custom {
234 name: model.name.clone(),
235 display_name: model.display_name.clone(),
236 max_tokens: model.max_tokens,
237 mode: model.mode.unwrap_or_default().into(),
238 },
239 );
240 }
241
242 models
243 .into_values()
244 .map(|model| {
245 Arc::new(GoogleLanguageModel {
246 id: LanguageModelId::from(model.id().to_string()),
247 model,
248 state: self.state.clone(),
249 http_client: self.http_client.clone(),
250 request_limiter: RateLimiter::new(4),
251 }) as Arc<dyn LanguageModel>
252 })
253 .collect()
254 }
255
256 fn is_authenticated(&self, cx: &App) -> bool {
257 self.state.read(cx).is_authenticated()
258 }
259
260 fn authenticate(&self, cx: &mut App) -> Task<Result<(), AuthenticateError>> {
261 self.state.update(cx, |state, cx| state.authenticate(cx))
262 }
263
264 fn configuration_view(
265 &self,
266 target_agent: language_model::ConfigurationViewTargetAgent,
267 window: &mut Window,
268 cx: &mut App,
269 ) -> AnyView {
270 cx.new(|cx| ConfigurationView::new(self.state.clone(), target_agent, window, cx))
271 .into()
272 }
273
274 fn reset_credentials(&self, cx: &mut App) -> Task<Result<()>> {
275 self.state
276 .update(cx, |state, cx| state.set_api_key(None, cx))
277 }
278}
279
280pub struct GoogleLanguageModel {
281 id: LanguageModelId,
282 model: google_ai::Model,
283 state: gpui::Entity<State>,
284 http_client: Arc<dyn HttpClient>,
285 request_limiter: RateLimiter,
286}
287
288impl GoogleLanguageModel {
289 fn stream_completion(
290 &self,
291 request: google_ai::GenerateContentRequest,
292 cx: &AsyncApp,
293 ) -> BoxFuture<
294 'static,
295 Result<futures::stream::BoxStream<'static, Result<GenerateContentResponse>>>,
296 > {
297 let http_client = self.http_client.clone();
298
299 let Ok((api_key, api_url)) = self.state.read_with(cx, |state, cx| {
300 let api_url = GoogleLanguageModelProvider::api_url(cx);
301 (state.api_key_state.key(&api_url), api_url)
302 }) else {
303 return future::ready(Err(anyhow!("App state dropped"))).boxed();
304 };
305
306 async move {
307 let api_key = api_key.context("Missing Google API key")?;
308 let request = google_ai::stream_generate_content(
309 http_client.as_ref(),
310 &api_url,
311 &api_key,
312 request,
313 );
314 request.await.context("failed to stream completion")
315 }
316 .boxed()
317 }
318}
319
320impl LanguageModel for GoogleLanguageModel {
321 fn id(&self) -> LanguageModelId {
322 self.id.clone()
323 }
324
325 fn name(&self) -> LanguageModelName {
326 LanguageModelName::from(self.model.display_name().to_string())
327 }
328
329 fn provider_id(&self) -> LanguageModelProviderId {
330 PROVIDER_ID
331 }
332
333 fn provider_name(&self) -> LanguageModelProviderName {
334 PROVIDER_NAME
335 }
336
337 fn supports_tools(&self) -> bool {
338 self.model.supports_tools()
339 }
340
341 fn supports_images(&self) -> bool {
342 self.model.supports_images()
343 }
344
345 fn supports_tool_choice(&self, choice: LanguageModelToolChoice) -> bool {
346 match choice {
347 LanguageModelToolChoice::Auto
348 | LanguageModelToolChoice::Any
349 | LanguageModelToolChoice::None => true,
350 }
351 }
352
353 fn tool_input_format(&self) -> LanguageModelToolSchemaFormat {
354 LanguageModelToolSchemaFormat::JsonSchemaSubset
355 }
356
357 fn telemetry_id(&self) -> String {
358 format!("google/{}", self.model.request_id())
359 }
360
361 fn max_token_count(&self) -> u64 {
362 self.model.max_token_count()
363 }
364
365 fn max_output_tokens(&self) -> Option<u64> {
366 self.model.max_output_tokens()
367 }
368
369 fn count_tokens(
370 &self,
371 request: LanguageModelRequest,
372 cx: &App,
373 ) -> BoxFuture<'static, Result<u64>> {
374 let model_id = self.model.request_id().to_string();
375 let request = into_google(request, model_id, self.model.mode());
376 let http_client = self.http_client.clone();
377 let api_url = GoogleLanguageModelProvider::api_url(cx);
378 let api_key = self.state.read(cx).api_key_state.key(&api_url);
379
380 async move {
381 let Some(api_key) = api_key else {
382 return Err(LanguageModelCompletionError::NoApiKey {
383 provider: PROVIDER_NAME,
384 }
385 .into());
386 };
387 let response = google_ai::count_tokens(
388 http_client.as_ref(),
389 &api_url,
390 &api_key,
391 google_ai::CountTokensRequest {
392 generate_content_request: request,
393 },
394 )
395 .await?;
396 Ok(response.total_tokens)
397 }
398 .boxed()
399 }
400
401 fn stream_completion(
402 &self,
403 request: LanguageModelRequest,
404 cx: &AsyncApp,
405 ) -> BoxFuture<
406 'static,
407 Result<
408 futures::stream::BoxStream<
409 'static,
410 Result<LanguageModelCompletionEvent, LanguageModelCompletionError>,
411 >,
412 LanguageModelCompletionError,
413 >,
414 > {
415 let request = into_google(
416 request,
417 self.model.request_id().to_string(),
418 self.model.mode(),
419 );
420 let request = self.stream_completion(request, cx);
421 let future = self.request_limiter.stream(async move {
422 let response = request.await.map_err(LanguageModelCompletionError::from)?;
423 Ok(GoogleEventMapper::new().map_stream(response))
424 });
425 async move { Ok(future.await?.boxed()) }.boxed()
426 }
427}
428
429pub fn into_google(
430 mut request: LanguageModelRequest,
431 model_id: String,
432 mode: GoogleModelMode,
433) -> google_ai::GenerateContentRequest {
434 fn map_content(content: Vec<MessageContent>) -> Vec<Part> {
435 content
436 .into_iter()
437 .flat_map(|content| match content {
438 language_model::MessageContent::Text(text) => {
439 if !text.is_empty() {
440 vec![Part::TextPart(google_ai::TextPart { text })]
441 } else {
442 vec![]
443 }
444 }
445 language_model::MessageContent::Thinking {
446 text: _,
447 signature: Some(signature),
448 } => {
449 if !signature.is_empty() {
450 vec![Part::ThoughtPart(google_ai::ThoughtPart {
451 thought: true,
452 thought_signature: signature,
453 })]
454 } else {
455 vec![]
456 }
457 }
458 language_model::MessageContent::Thinking { .. } => {
459 vec![]
460 }
461 language_model::MessageContent::RedactedThinking(_) => vec![],
462 language_model::MessageContent::Image(image) => {
463 vec![Part::InlineDataPart(google_ai::InlineDataPart {
464 inline_data: google_ai::GenerativeContentBlob {
465 mime_type: "image/png".to_string(),
466 data: image.source.to_string(),
467 },
468 })]
469 }
470 language_model::MessageContent::ToolUse(tool_use) => {
471 vec![Part::FunctionCallPart(google_ai::FunctionCallPart {
472 function_call: google_ai::FunctionCall {
473 name: tool_use.name.to_string(),
474 args: tool_use.input,
475 },
476 })]
477 }
478 language_model::MessageContent::ToolResult(tool_result) => {
479 match tool_result.content {
480 language_model::LanguageModelToolResultContent::Text(text) => {
481 vec![Part::FunctionResponsePart(
482 google_ai::FunctionResponsePart {
483 function_response: google_ai::FunctionResponse {
484 name: tool_result.tool_name.to_string(),
485 // The API expects a valid JSON object
486 response: serde_json::json!({
487 "output": text
488 }),
489 },
490 },
491 )]
492 }
493 language_model::LanguageModelToolResultContent::Image(image) => {
494 vec![
495 Part::FunctionResponsePart(google_ai::FunctionResponsePart {
496 function_response: google_ai::FunctionResponse {
497 name: tool_result.tool_name.to_string(),
498 // The API expects a valid JSON object
499 response: serde_json::json!({
500 "output": "Tool responded with an image"
501 }),
502 },
503 }),
504 Part::InlineDataPart(google_ai::InlineDataPart {
505 inline_data: google_ai::GenerativeContentBlob {
506 mime_type: "image/png".to_string(),
507 data: image.source.to_string(),
508 },
509 }),
510 ]
511 }
512 }
513 }
514 })
515 .collect()
516 }
517
518 let system_instructions = if request
519 .messages
520 .first()
521 .is_some_and(|msg| matches!(msg.role, Role::System))
522 {
523 let message = request.messages.remove(0);
524 Some(SystemInstruction {
525 parts: map_content(message.content),
526 })
527 } else {
528 None
529 };
530
531 google_ai::GenerateContentRequest {
532 model: google_ai::ModelName { model_id },
533 system_instruction: system_instructions,
534 contents: request
535 .messages
536 .into_iter()
537 .filter_map(|message| {
538 let parts = map_content(message.content);
539 if parts.is_empty() {
540 None
541 } else {
542 Some(google_ai::Content {
543 parts,
544 role: match message.role {
545 Role::User => google_ai::Role::User,
546 Role::Assistant => google_ai::Role::Model,
547 Role::System => google_ai::Role::User, // Google AI doesn't have a system role
548 },
549 })
550 }
551 })
552 .collect(),
553 generation_config: Some(google_ai::GenerationConfig {
554 candidate_count: Some(1),
555 stop_sequences: Some(request.stop),
556 max_output_tokens: None,
557 temperature: request.temperature.map(|t| t as f64).or(Some(1.0)),
558 thinking_config: match (request.thinking_allowed, mode) {
559 (true, GoogleModelMode::Thinking { budget_tokens }) => {
560 budget_tokens.map(|thinking_budget| ThinkingConfig { thinking_budget })
561 }
562 _ => None,
563 },
564 top_p: None,
565 top_k: None,
566 }),
567 safety_settings: None,
568 tools: (!request.tools.is_empty()).then(|| {
569 vec![google_ai::Tool {
570 function_declarations: request
571 .tools
572 .into_iter()
573 .map(|tool| FunctionDeclaration {
574 name: tool.name,
575 description: tool.description,
576 parameters: tool.input_schema,
577 })
578 .collect(),
579 }]
580 }),
581 tool_config: request.tool_choice.map(|choice| google_ai::ToolConfig {
582 function_calling_config: google_ai::FunctionCallingConfig {
583 mode: match choice {
584 LanguageModelToolChoice::Auto => google_ai::FunctionCallingMode::Auto,
585 LanguageModelToolChoice::Any => google_ai::FunctionCallingMode::Any,
586 LanguageModelToolChoice::None => google_ai::FunctionCallingMode::None,
587 },
588 allowed_function_names: None,
589 },
590 }),
591 }
592}
593
594pub struct GoogleEventMapper {
595 usage: UsageMetadata,
596 stop_reason: StopReason,
597}
598
599impl GoogleEventMapper {
600 pub fn new() -> Self {
601 Self {
602 usage: UsageMetadata::default(),
603 stop_reason: StopReason::EndTurn,
604 }
605 }
606
607 pub fn map_stream(
608 mut self,
609 events: Pin<Box<dyn Send + Stream<Item = Result<GenerateContentResponse>>>>,
610 ) -> impl Stream<Item = Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>
611 {
612 events
613 .map(Some)
614 .chain(futures::stream::once(async { None }))
615 .flat_map(move |event| {
616 futures::stream::iter(match event {
617 Some(Ok(event)) => self.map_event(event),
618 Some(Err(error)) => {
619 vec![Err(LanguageModelCompletionError::from(error))]
620 }
621 None => vec![Ok(LanguageModelCompletionEvent::Stop(self.stop_reason))],
622 })
623 })
624 }
625
626 pub fn map_event(
627 &mut self,
628 event: GenerateContentResponse,
629 ) -> Vec<Result<LanguageModelCompletionEvent, LanguageModelCompletionError>> {
630 static TOOL_CALL_COUNTER: AtomicU64 = AtomicU64::new(0);
631
632 let mut events: Vec<_> = Vec::new();
633 let mut wants_to_use_tool = false;
634 if let Some(usage_metadata) = event.usage_metadata {
635 update_usage(&mut self.usage, &usage_metadata);
636 events.push(Ok(LanguageModelCompletionEvent::UsageUpdate(
637 convert_usage(&self.usage),
638 )))
639 }
640 if let Some(candidates) = event.candidates {
641 for candidate in candidates {
642 if let Some(finish_reason) = candidate.finish_reason.as_deref() {
643 self.stop_reason = match finish_reason {
644 "STOP" => StopReason::EndTurn,
645 "MAX_TOKENS" => StopReason::MaxTokens,
646 _ => {
647 log::error!("Unexpected google finish_reason: {finish_reason}");
648 StopReason::EndTurn
649 }
650 };
651 }
652 candidate
653 .content
654 .parts
655 .into_iter()
656 .for_each(|part| match part {
657 Part::TextPart(text_part) => {
658 events.push(Ok(LanguageModelCompletionEvent::Text(text_part.text)))
659 }
660 Part::InlineDataPart(_) => {}
661 Part::FunctionCallPart(function_call_part) => {
662 wants_to_use_tool = true;
663 let name: Arc<str> = function_call_part.function_call.name.into();
664 let next_tool_id =
665 TOOL_CALL_COUNTER.fetch_add(1, atomic::Ordering::SeqCst);
666 let id: LanguageModelToolUseId =
667 format!("{}-{}", name, next_tool_id).into();
668
669 events.push(Ok(LanguageModelCompletionEvent::ToolUse(
670 LanguageModelToolUse {
671 id,
672 name,
673 is_input_complete: true,
674 raw_input: function_call_part.function_call.args.to_string(),
675 input: function_call_part.function_call.args,
676 },
677 )));
678 }
679 Part::FunctionResponsePart(_) => {}
680 Part::ThoughtPart(part) => {
681 events.push(Ok(LanguageModelCompletionEvent::Thinking {
682 text: "(Encrypted thought)".to_string(), // TODO: Can we populate this from thought summaries?
683 signature: Some(part.thought_signature),
684 }));
685 }
686 });
687 }
688 }
689
690 // Even when Gemini wants to use a Tool, the API
691 // responds with `finish_reason: STOP`
692 if wants_to_use_tool {
693 self.stop_reason = StopReason::ToolUse;
694 events.push(Ok(LanguageModelCompletionEvent::Stop(StopReason::ToolUse)));
695 }
696 events
697 }
698}
699
700pub fn count_google_tokens(
701 request: LanguageModelRequest,
702 cx: &App,
703) -> BoxFuture<'static, Result<u64>> {
704 // We couldn't use the GoogleLanguageModelProvider to count tokens because the github copilot doesn't have the access to google_ai directly.
705 // So we have to use tokenizer from tiktoken_rs to count tokens.
706 cx.background_spawn(async move {
707 let messages = request
708 .messages
709 .into_iter()
710 .map(|message| tiktoken_rs::ChatCompletionRequestMessage {
711 role: match message.role {
712 Role::User => "user".into(),
713 Role::Assistant => "assistant".into(),
714 Role::System => "system".into(),
715 },
716 content: Some(message.string_contents()),
717 name: None,
718 function_call: None,
719 })
720 .collect::<Vec<_>>();
721
722 // Tiktoken doesn't yet support these models, so we manually use the
723 // same tokenizer as GPT-4.
724 tiktoken_rs::num_tokens_from_messages("gpt-4", &messages).map(|tokens| tokens as u64)
725 })
726 .boxed()
727}
728
729fn update_usage(usage: &mut UsageMetadata, new: &UsageMetadata) {
730 if let Some(prompt_token_count) = new.prompt_token_count {
731 usage.prompt_token_count = Some(prompt_token_count);
732 }
733 if let Some(cached_content_token_count) = new.cached_content_token_count {
734 usage.cached_content_token_count = Some(cached_content_token_count);
735 }
736 if let Some(candidates_token_count) = new.candidates_token_count {
737 usage.candidates_token_count = Some(candidates_token_count);
738 }
739 if let Some(tool_use_prompt_token_count) = new.tool_use_prompt_token_count {
740 usage.tool_use_prompt_token_count = Some(tool_use_prompt_token_count);
741 }
742 if let Some(thoughts_token_count) = new.thoughts_token_count {
743 usage.thoughts_token_count = Some(thoughts_token_count);
744 }
745 if let Some(total_token_count) = new.total_token_count {
746 usage.total_token_count = Some(total_token_count);
747 }
748}
749
750fn convert_usage(usage: &UsageMetadata) -> language_model::TokenUsage {
751 let prompt_tokens = usage.prompt_token_count.unwrap_or(0);
752 let cached_tokens = usage.cached_content_token_count.unwrap_or(0);
753 let input_tokens = prompt_tokens - cached_tokens;
754 let output_tokens = usage.candidates_token_count.unwrap_or(0);
755
756 language_model::TokenUsage {
757 input_tokens,
758 output_tokens,
759 cache_read_input_tokens: cached_tokens,
760 cache_creation_input_tokens: 0,
761 }
762}
763
764struct ConfigurationView {
765 api_key_editor: Entity<Editor>,
766 state: gpui::Entity<State>,
767 target_agent: language_model::ConfigurationViewTargetAgent,
768 load_credentials_task: Option<Task<()>>,
769}
770
771impl ConfigurationView {
772 fn new(
773 state: gpui::Entity<State>,
774 target_agent: language_model::ConfigurationViewTargetAgent,
775 window: &mut Window,
776 cx: &mut Context<Self>,
777 ) -> Self {
778 cx.observe(&state, |_, _, cx| {
779 cx.notify();
780 })
781 .detach();
782
783 let load_credentials_task = Some(cx.spawn_in(window, {
784 let state = state.clone();
785 async move |this, cx| {
786 if let Some(task) = state
787 .update(cx, |state, cx| state.authenticate(cx))
788 .log_err()
789 {
790 // We don't log an error, because "not signed in" is also an error.
791 let _ = task.await;
792 }
793 this.update(cx, |this, cx| {
794 this.load_credentials_task = None;
795 cx.notify();
796 })
797 .log_err();
798 }
799 }));
800
801 Self {
802 api_key_editor: cx.new(|cx| {
803 let mut editor = Editor::single_line(window, cx);
804 editor.set_placeholder_text("AIzaSy...", window, cx);
805 editor
806 }),
807 target_agent,
808 state,
809 load_credentials_task,
810 }
811 }
812
813 fn save_api_key(&mut self, _: &menu::Confirm, window: &mut Window, cx: &mut Context<Self>) {
814 let api_key = self.api_key_editor.read(cx).text(cx).trim().to_string();
815 if api_key.is_empty() {
816 return;
817 }
818
819 // url changes can cause the editor to be displayed again
820 self.api_key_editor
821 .update(cx, |editor, cx| editor.set_text("", window, cx));
822
823 let state = self.state.clone();
824 cx.spawn_in(window, async move |_, cx| {
825 state
826 .update(cx, |state, cx| state.set_api_key(Some(api_key), cx))?
827 .await
828 })
829 .detach_and_log_err(cx);
830 }
831
832 fn reset_api_key(&mut self, window: &mut Window, cx: &mut Context<Self>) {
833 self.api_key_editor
834 .update(cx, |editor, cx| editor.set_text("", window, cx));
835
836 let state = self.state.clone();
837 cx.spawn_in(window, async move |_, cx| {
838 state
839 .update(cx, |state, cx| state.set_api_key(None, cx))?
840 .await
841 })
842 .detach_and_log_err(cx);
843 }
844
845 fn render_api_key_editor(&self, cx: &mut Context<Self>) -> impl IntoElement {
846 let settings = ThemeSettings::get_global(cx);
847 let text_style = TextStyle {
848 color: cx.theme().colors().text,
849 font_family: settings.ui_font.family.clone(),
850 font_features: settings.ui_font.features.clone(),
851 font_fallbacks: settings.ui_font.fallbacks.clone(),
852 font_size: rems(0.875).into(),
853 font_weight: settings.ui_font.weight,
854 font_style: FontStyle::Normal,
855 line_height: relative(1.3),
856 white_space: WhiteSpace::Normal,
857 ..Default::default()
858 };
859 EditorElement::new(
860 &self.api_key_editor,
861 EditorStyle {
862 background: cx.theme().colors().editor_background,
863 local_player: cx.theme().players().local(),
864 text: text_style,
865 ..Default::default()
866 },
867 )
868 }
869
870 fn should_render_editor(&self, cx: &mut Context<Self>) -> bool {
871 !self.state.read(cx).is_authenticated()
872 }
873}
874
875impl Render for ConfigurationView {
876 fn render(&mut self, _: &mut Window, cx: &mut Context<Self>) -> impl IntoElement {
877 let env_var_set = self.state.read(cx).api_key_state.is_from_env_var();
878
879 if self.load_credentials_task.is_some() {
880 div().child(Label::new("Loading credentials...")).into_any()
881 } else if self.should_render_editor(cx) {
882 v_flex()
883 .size_full()
884 .on_action(cx.listener(Self::save_api_key))
885 .child(Label::new(format!("To use {}, you need to add an API key. Follow these steps:", match &self.target_agent {
886 ConfigurationViewTargetAgent::ZedAgent => "Zed's agent with Google AI".into(),
887 ConfigurationViewTargetAgent::Other(agent) => agent.clone(),
888 })))
889 .child(
890 List::new()
891 .child(InstructionListItem::new(
892 "Create one by visiting",
893 Some("Google AI's console"),
894 Some("https://aistudio.google.com/app/apikey"),
895 ))
896 .child(InstructionListItem::text_only(
897 "Paste your API key below and hit enter to start using the assistant",
898 )),
899 )
900 .child(
901 h_flex()
902 .w_full()
903 .my_2()
904 .px_2()
905 .py_1()
906 .bg(cx.theme().colors().editor_background)
907 .border_1()
908 .border_color(cx.theme().colors().border)
909 .rounded_sm()
910 .child(self.render_api_key_editor(cx)),
911 )
912 .child(
913 Label::new(
914 format!("You can also assign the {GEMINI_API_KEY_VAR_NAME} environment variable and restart Zed."),
915 )
916 .size(LabelSize::Small).color(Color::Muted),
917 )
918 .into_any()
919 } else {
920 h_flex()
921 .mt_1()
922 .p_1()
923 .justify_between()
924 .rounded_md()
925 .border_1()
926 .border_color(cx.theme().colors().border)
927 .bg(cx.theme().colors().background)
928 .child(
929 h_flex()
930 .gap_1()
931 .child(Icon::new(IconName::Check).color(Color::Success))
932 .child(Label::new(if env_var_set {
933 format!("API key set in {} environment variable", API_KEY_ENV_VAR.name)
934 } else {
935 let api_url = GoogleLanguageModelProvider::api_url(cx);
936 if api_url == google_ai::API_URL {
937 "API key configured".to_string()
938 } else {
939 format!("API key configured for {}", truncate_and_trailoff(&api_url, 32))
940 }
941 })),
942 )
943 .child(
944 Button::new("reset-key", "Reset Key")
945 .label_size(LabelSize::Small)
946 .icon(Some(IconName::Trash))
947 .icon_size(IconSize::Small)
948 .icon_position(IconPosition::Start)
949 .disabled(env_var_set)
950 .when(env_var_set, |this| {
951 this.tooltip(Tooltip::text(format!("To reset your API key, make sure {GEMINI_API_KEY_VAR_NAME} and {GOOGLE_AI_API_KEY_VAR_NAME} environment variables are unset.")))
952 })
953 .on_click(cx.listener(|this, _, window, cx| this.reset_api_key(window, cx))),
954 )
955 .into_any()
956 }
957 }
958}