1use anyhow::{Context as _, Result};
2use collections::BTreeMap;
3use futures::{FutureExt, Stream, StreamExt, future::BoxFuture};
4use google_ai::{
5 FunctionDeclaration, GenerateContentResponse, GoogleModelMode, Part, SystemInstruction,
6 ThinkingConfig, UsageMetadata,
7};
8use gpui::{AnyView, App, AsyncApp, Context, Entity, SharedString, Task, Window};
9use http_client::HttpClient;
10use language_model::{
11 AuthenticateError, ConfigurationViewTargetAgent, EnvVar, LanguageModelCompletionError,
12 LanguageModelCompletionEvent, LanguageModelToolChoice, LanguageModelToolSchemaFormat,
13 LanguageModelToolUse, LanguageModelToolUseId, MessageContent, StopReason,
14};
15use language_model::{
16 IconOrSvg, LanguageModel, LanguageModelId, LanguageModelName, LanguageModelProvider,
17 LanguageModelProviderId, LanguageModelProviderName, LanguageModelProviderState,
18 LanguageModelRequest, RateLimiter, Role,
19};
20use schemars::JsonSchema;
21use serde::{Deserialize, Serialize};
22pub use settings::GoogleAvailableModel as AvailableModel;
23use settings::{Settings, SettingsStore};
24use std::pin::Pin;
25use std::sync::{
26 Arc, LazyLock,
27 atomic::{self, AtomicU64},
28};
29use strum::IntoEnumIterator;
30use ui::{ButtonLink, ConfiguredApiCard, List, ListBulletItem, prelude::*};
31use ui_input::InputField;
32use util::ResultExt;
33
34use language_model::ApiKeyState;
35
36const PROVIDER_ID: LanguageModelProviderId = language_model::GOOGLE_PROVIDER_ID;
37const PROVIDER_NAME: LanguageModelProviderName = language_model::GOOGLE_PROVIDER_NAME;
38
39#[derive(Default, Clone, Debug, PartialEq)]
40pub struct GoogleSettings {
41 pub api_url: String,
42 pub available_models: Vec<AvailableModel>,
43}
44
45#[derive(Clone, Copy, Debug, Default, PartialEq, Serialize, Deserialize, JsonSchema)]
46#[serde(tag = "type", rename_all = "lowercase")]
47pub enum ModelMode {
48 #[default]
49 Default,
50 Thinking {
51 /// The maximum number of tokens to use for reasoning. Must be lower than the model's `max_output_tokens`.
52 budget_tokens: Option<u32>,
53 },
54}
55
56pub struct GoogleLanguageModelProvider {
57 http_client: Arc<dyn HttpClient>,
58 state: Entity<State>,
59}
60
61pub struct State {
62 api_key_state: ApiKeyState,
63}
64
65const GEMINI_API_KEY_VAR_NAME: &str = "GEMINI_API_KEY";
66const GOOGLE_AI_API_KEY_VAR_NAME: &str = "GOOGLE_AI_API_KEY";
67
68static API_KEY_ENV_VAR: LazyLock<EnvVar> = LazyLock::new(|| {
69 // Try GEMINI_API_KEY first as primary, fallback to GOOGLE_AI_API_KEY
70 EnvVar::new(GEMINI_API_KEY_VAR_NAME.into()).or(EnvVar::new(GOOGLE_AI_API_KEY_VAR_NAME.into()))
71});
72
73impl State {
74 fn is_authenticated(&self) -> bool {
75 self.api_key_state.has_key()
76 }
77
78 fn set_api_key(&mut self, api_key: Option<String>, cx: &mut Context<Self>) -> Task<Result<()>> {
79 let api_url = GoogleLanguageModelProvider::api_url(cx);
80 self.api_key_state
81 .store(api_url, api_key, |this| &mut this.api_key_state, cx)
82 }
83
84 fn authenticate(&mut self, cx: &mut Context<Self>) -> Task<Result<(), AuthenticateError>> {
85 let api_url = GoogleLanguageModelProvider::api_url(cx);
86 self.api_key_state
87 .load_if_needed(api_url, |this| &mut this.api_key_state, cx)
88 }
89}
90
91impl GoogleLanguageModelProvider {
92 pub fn new(http_client: Arc<dyn HttpClient>, cx: &mut App) -> Self {
93 let state = cx.new(|cx| {
94 cx.observe_global::<SettingsStore>(|this: &mut State, cx| {
95 let api_url = Self::api_url(cx);
96 this.api_key_state
97 .handle_url_change(api_url, |this| &mut this.api_key_state, cx);
98 cx.notify();
99 })
100 .detach();
101 State {
102 api_key_state: ApiKeyState::new(Self::api_url(cx), (*API_KEY_ENV_VAR).clone()),
103 }
104 });
105
106 Self { http_client, state }
107 }
108
109 fn create_language_model(&self, model: google_ai::Model) -> Arc<dyn LanguageModel> {
110 Arc::new(GoogleLanguageModel {
111 id: LanguageModelId::from(model.id().to_string()),
112 model,
113 state: self.state.clone(),
114 http_client: self.http_client.clone(),
115 request_limiter: RateLimiter::new(4),
116 })
117 }
118
119 fn settings(cx: &App) -> &GoogleSettings {
120 &crate::AllLanguageModelSettings::get_global(cx).google
121 }
122
123 fn api_url(cx: &App) -> SharedString {
124 let api_url = &Self::settings(cx).api_url;
125 if api_url.is_empty() {
126 google_ai::API_URL.into()
127 } else {
128 SharedString::new(api_url.as_str())
129 }
130 }
131}
132
133impl LanguageModelProviderState for GoogleLanguageModelProvider {
134 type ObservableEntity = State;
135
136 fn observable_entity(&self) -> Option<Entity<Self::ObservableEntity>> {
137 Some(self.state.clone())
138 }
139}
140
141impl LanguageModelProvider for GoogleLanguageModelProvider {
142 fn id(&self) -> LanguageModelProviderId {
143 PROVIDER_ID
144 }
145
146 fn name(&self) -> LanguageModelProviderName {
147 PROVIDER_NAME
148 }
149
150 fn icon(&self) -> IconOrSvg {
151 IconOrSvg::Icon(IconName::AiGoogle)
152 }
153
154 fn default_model(&self, _cx: &App) -> Option<Arc<dyn LanguageModel>> {
155 Some(self.create_language_model(google_ai::Model::default()))
156 }
157
158 fn default_fast_model(&self, _cx: &App) -> Option<Arc<dyn LanguageModel>> {
159 Some(self.create_language_model(google_ai::Model::default_fast()))
160 }
161
162 fn provided_models(&self, cx: &App) -> Vec<Arc<dyn LanguageModel>> {
163 let mut models = BTreeMap::default();
164
165 // Add base models from google_ai::Model::iter()
166 for model in google_ai::Model::iter() {
167 if !matches!(model, google_ai::Model::Custom { .. }) {
168 models.insert(model.id().to_string(), model);
169 }
170 }
171
172 // Override with available models from settings
173 for model in &GoogleLanguageModelProvider::settings(cx).available_models {
174 models.insert(
175 model.name.clone(),
176 google_ai::Model::Custom {
177 name: model.name.clone(),
178 display_name: model.display_name.clone(),
179 max_tokens: model.max_tokens,
180 mode: model.mode.unwrap_or_default(),
181 },
182 );
183 }
184
185 models
186 .into_values()
187 .map(|model| {
188 Arc::new(GoogleLanguageModel {
189 id: LanguageModelId::from(model.id().to_string()),
190 model,
191 state: self.state.clone(),
192 http_client: self.http_client.clone(),
193 request_limiter: RateLimiter::new(4),
194 }) as Arc<dyn LanguageModel>
195 })
196 .collect()
197 }
198
199 fn is_authenticated(&self, cx: &App) -> bool {
200 self.state.read(cx).is_authenticated()
201 }
202
203 fn authenticate(&self, cx: &mut App) -> Task<Result<(), AuthenticateError>> {
204 self.state.update(cx, |state, cx| state.authenticate(cx))
205 }
206
207 fn configuration_view(
208 &self,
209 target_agent: language_model::ConfigurationViewTargetAgent,
210 window: &mut Window,
211 cx: &mut App,
212 ) -> AnyView {
213 cx.new(|cx| ConfigurationView::new(self.state.clone(), target_agent, window, cx))
214 .into()
215 }
216
217 fn reset_credentials(&self, cx: &mut App) -> Task<Result<()>> {
218 self.state
219 .update(cx, |state, cx| state.set_api_key(None, cx))
220 }
221}
222
223pub struct GoogleLanguageModel {
224 id: LanguageModelId,
225 model: google_ai::Model,
226 state: Entity<State>,
227 http_client: Arc<dyn HttpClient>,
228 request_limiter: RateLimiter,
229}
230
231impl GoogleLanguageModel {
232 fn stream_completion(
233 &self,
234 request: google_ai::GenerateContentRequest,
235 cx: &AsyncApp,
236 ) -> BoxFuture<
237 'static,
238 Result<futures::stream::BoxStream<'static, Result<GenerateContentResponse>>>,
239 > {
240 let http_client = self.http_client.clone();
241
242 let (api_key, api_url) = self.state.read_with(cx, |state, cx| {
243 let api_url = GoogleLanguageModelProvider::api_url(cx);
244 (state.api_key_state.key(&api_url), api_url)
245 });
246
247 async move {
248 let api_key = api_key.context("Missing Google API key")?;
249 let request = google_ai::stream_generate_content(
250 http_client.as_ref(),
251 &api_url,
252 &api_key,
253 request,
254 );
255 request.await.context("failed to stream completion")
256 }
257 .boxed()
258 }
259}
260
261impl LanguageModel for GoogleLanguageModel {
262 fn id(&self) -> LanguageModelId {
263 self.id.clone()
264 }
265
266 fn name(&self) -> LanguageModelName {
267 LanguageModelName::from(self.model.display_name().to_string())
268 }
269
270 fn provider_id(&self) -> LanguageModelProviderId {
271 PROVIDER_ID
272 }
273
274 fn provider_name(&self) -> LanguageModelProviderName {
275 PROVIDER_NAME
276 }
277
278 fn supports_tools(&self) -> bool {
279 self.model.supports_tools()
280 }
281
282 fn supports_images(&self) -> bool {
283 self.model.supports_images()
284 }
285
286 fn supports_tool_choice(&self, choice: LanguageModelToolChoice) -> bool {
287 match choice {
288 LanguageModelToolChoice::Auto
289 | LanguageModelToolChoice::Any
290 | LanguageModelToolChoice::None => true,
291 }
292 }
293
294 fn tool_input_format(&self) -> LanguageModelToolSchemaFormat {
295 LanguageModelToolSchemaFormat::JsonSchemaSubset
296 }
297
298 fn telemetry_id(&self) -> String {
299 format!("google/{}", self.model.request_id())
300 }
301
302 fn max_token_count(&self) -> u64 {
303 self.model.max_token_count()
304 }
305
306 fn max_output_tokens(&self) -> Option<u64> {
307 self.model.max_output_tokens()
308 }
309
310 fn count_tokens(
311 &self,
312 request: LanguageModelRequest,
313 cx: &App,
314 ) -> BoxFuture<'static, Result<u64>> {
315 let model_id = self.model.request_id().to_string();
316 let request = into_google(request, model_id, self.model.mode());
317 let http_client = self.http_client.clone();
318 let api_url = GoogleLanguageModelProvider::api_url(cx);
319 let api_key = self.state.read(cx).api_key_state.key(&api_url);
320
321 async move {
322 let Some(api_key) = api_key else {
323 return Err(LanguageModelCompletionError::NoApiKey {
324 provider: PROVIDER_NAME,
325 }
326 .into());
327 };
328 let response = google_ai::count_tokens(
329 http_client.as_ref(),
330 &api_url,
331 &api_key,
332 google_ai::CountTokensRequest {
333 generate_content_request: request,
334 },
335 )
336 .await?;
337 Ok(response.total_tokens)
338 }
339 .boxed()
340 }
341
342 fn stream_completion(
343 &self,
344 request: LanguageModelRequest,
345 cx: &AsyncApp,
346 ) -> BoxFuture<
347 'static,
348 Result<
349 futures::stream::BoxStream<
350 'static,
351 Result<LanguageModelCompletionEvent, LanguageModelCompletionError>,
352 >,
353 LanguageModelCompletionError,
354 >,
355 > {
356 let request = into_google(
357 request,
358 self.model.request_id().to_string(),
359 self.model.mode(),
360 );
361 let request = self.stream_completion(request, cx);
362 let future = self.request_limiter.stream(async move {
363 let response = request.await.map_err(LanguageModelCompletionError::from)?;
364 Ok(GoogleEventMapper::new().map_stream(response))
365 });
366 async move { Ok(future.await?.boxed()) }.boxed()
367 }
368}
369
370pub fn into_google(
371 mut request: LanguageModelRequest,
372 model_id: String,
373 mode: GoogleModelMode,
374) -> google_ai::GenerateContentRequest {
375 fn map_content(content: Vec<MessageContent>) -> Vec<Part> {
376 content
377 .into_iter()
378 .flat_map(|content| match content {
379 language_model::MessageContent::Text(text) => {
380 if !text.is_empty() {
381 vec![Part::TextPart(google_ai::TextPart { text })]
382 } else {
383 vec![]
384 }
385 }
386 language_model::MessageContent::Thinking {
387 text: _,
388 signature: Some(signature),
389 } => {
390 if !signature.is_empty() {
391 vec![Part::ThoughtPart(google_ai::ThoughtPart {
392 thought: true,
393 thought_signature: signature,
394 })]
395 } else {
396 vec![]
397 }
398 }
399 language_model::MessageContent::Thinking { .. } => {
400 vec![]
401 }
402 language_model::MessageContent::RedactedThinking(_) => vec![],
403 language_model::MessageContent::Image(image) => {
404 vec![Part::InlineDataPart(google_ai::InlineDataPart {
405 inline_data: google_ai::GenerativeContentBlob {
406 mime_type: "image/png".to_string(),
407 data: image.source.to_string(),
408 },
409 })]
410 }
411 language_model::MessageContent::ToolUse(tool_use) => {
412 // Normalize empty string signatures to None
413 let thought_signature = tool_use.thought_signature.filter(|s| !s.is_empty());
414
415 vec![Part::FunctionCallPart(google_ai::FunctionCallPart {
416 function_call: google_ai::FunctionCall {
417 name: tool_use.name.to_string(),
418 args: tool_use.input,
419 },
420 thought_signature,
421 })]
422 }
423 language_model::MessageContent::ToolResult(tool_result) => {
424 match tool_result.content {
425 language_model::LanguageModelToolResultContent::Text(text) => {
426 vec![Part::FunctionResponsePart(
427 google_ai::FunctionResponsePart {
428 function_response: google_ai::FunctionResponse {
429 name: tool_result.tool_name.to_string(),
430 // The API expects a valid JSON object
431 response: serde_json::json!({
432 "output": text
433 }),
434 },
435 },
436 )]
437 }
438 language_model::LanguageModelToolResultContent::Image(image) => {
439 vec![
440 Part::FunctionResponsePart(google_ai::FunctionResponsePart {
441 function_response: google_ai::FunctionResponse {
442 name: tool_result.tool_name.to_string(),
443 // The API expects a valid JSON object
444 response: serde_json::json!({
445 "output": "Tool responded with an image"
446 }),
447 },
448 }),
449 Part::InlineDataPart(google_ai::InlineDataPart {
450 inline_data: google_ai::GenerativeContentBlob {
451 mime_type: "image/png".to_string(),
452 data: image.source.to_string(),
453 },
454 }),
455 ]
456 }
457 }
458 }
459 })
460 .collect()
461 }
462
463 let system_instructions = if request
464 .messages
465 .first()
466 .is_some_and(|msg| matches!(msg.role, Role::System))
467 {
468 let message = request.messages.remove(0);
469 Some(SystemInstruction {
470 parts: map_content(message.content),
471 })
472 } else {
473 None
474 };
475
476 google_ai::GenerateContentRequest {
477 model: google_ai::ModelName { model_id },
478 system_instruction: system_instructions,
479 contents: request
480 .messages
481 .into_iter()
482 .filter_map(|message| {
483 let parts = map_content(message.content);
484 if parts.is_empty() {
485 None
486 } else {
487 Some(google_ai::Content {
488 parts,
489 role: match message.role {
490 Role::User => google_ai::Role::User,
491 Role::Assistant => google_ai::Role::Model,
492 Role::System => google_ai::Role::User, // Google AI doesn't have a system role
493 },
494 })
495 }
496 })
497 .collect(),
498 generation_config: Some(google_ai::GenerationConfig {
499 candidate_count: Some(1),
500 stop_sequences: Some(request.stop),
501 max_output_tokens: None,
502 temperature: request.temperature.map(|t| t as f64).or(Some(1.0)),
503 thinking_config: match (request.thinking_allowed, mode) {
504 (true, GoogleModelMode::Thinking { budget_tokens }) => {
505 budget_tokens.map(|thinking_budget| ThinkingConfig { thinking_budget })
506 }
507 _ => None,
508 },
509 top_p: None,
510 top_k: None,
511 }),
512 safety_settings: None,
513 tools: (!request.tools.is_empty()).then(|| {
514 vec![google_ai::Tool {
515 function_declarations: request
516 .tools
517 .into_iter()
518 .map(|tool| FunctionDeclaration {
519 name: tool.name,
520 description: tool.description,
521 parameters: tool.input_schema,
522 })
523 .collect(),
524 }]
525 }),
526 tool_config: request.tool_choice.map(|choice| google_ai::ToolConfig {
527 function_calling_config: google_ai::FunctionCallingConfig {
528 mode: match choice {
529 LanguageModelToolChoice::Auto => google_ai::FunctionCallingMode::Auto,
530 LanguageModelToolChoice::Any => google_ai::FunctionCallingMode::Any,
531 LanguageModelToolChoice::None => google_ai::FunctionCallingMode::None,
532 },
533 allowed_function_names: None,
534 },
535 }),
536 }
537}
538
539pub struct GoogleEventMapper {
540 usage: UsageMetadata,
541 stop_reason: StopReason,
542}
543
544impl GoogleEventMapper {
545 pub fn new() -> Self {
546 Self {
547 usage: UsageMetadata::default(),
548 stop_reason: StopReason::EndTurn,
549 }
550 }
551
552 pub fn map_stream(
553 mut self,
554 events: Pin<Box<dyn Send + Stream<Item = Result<GenerateContentResponse>>>>,
555 ) -> impl Stream<Item = Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>
556 {
557 events
558 .map(Some)
559 .chain(futures::stream::once(async { None }))
560 .flat_map(move |event| {
561 futures::stream::iter(match event {
562 Some(Ok(event)) => self.map_event(event),
563 Some(Err(error)) => {
564 vec![Err(LanguageModelCompletionError::from(error))]
565 }
566 None => vec![Ok(LanguageModelCompletionEvent::Stop(self.stop_reason))],
567 })
568 })
569 }
570
571 pub fn map_event(
572 &mut self,
573 event: GenerateContentResponse,
574 ) -> Vec<Result<LanguageModelCompletionEvent, LanguageModelCompletionError>> {
575 static TOOL_CALL_COUNTER: AtomicU64 = AtomicU64::new(0);
576
577 let mut events: Vec<_> = Vec::new();
578 let mut wants_to_use_tool = false;
579 if let Some(usage_metadata) = event.usage_metadata {
580 update_usage(&mut self.usage, &usage_metadata);
581 events.push(Ok(LanguageModelCompletionEvent::UsageUpdate(
582 convert_usage(&self.usage),
583 )))
584 }
585
586 if let Some(prompt_feedback) = event.prompt_feedback
587 && let Some(block_reason) = prompt_feedback.block_reason.as_deref()
588 {
589 self.stop_reason = match block_reason {
590 "SAFETY" | "OTHER" | "BLOCKLIST" | "PROHIBITED_CONTENT" | "IMAGE_SAFETY" => {
591 StopReason::Refusal
592 }
593 _ => {
594 log::error!("Unexpected Google block_reason: {block_reason}");
595 StopReason::Refusal
596 }
597 };
598 events.push(Ok(LanguageModelCompletionEvent::Stop(self.stop_reason)));
599
600 return events;
601 }
602
603 if let Some(candidates) = event.candidates {
604 for candidate in candidates {
605 if let Some(finish_reason) = candidate.finish_reason.as_deref() {
606 self.stop_reason = match finish_reason {
607 "STOP" => StopReason::EndTurn,
608 "MAX_TOKENS" => StopReason::MaxTokens,
609 _ => {
610 log::error!("Unexpected google finish_reason: {finish_reason}");
611 StopReason::EndTurn
612 }
613 };
614 }
615 candidate
616 .content
617 .parts
618 .into_iter()
619 .for_each(|part| match part {
620 Part::TextPart(text_part) => {
621 events.push(Ok(LanguageModelCompletionEvent::Text(text_part.text)))
622 }
623 Part::InlineDataPart(_) => {}
624 Part::FunctionCallPart(function_call_part) => {
625 wants_to_use_tool = true;
626 let name: Arc<str> = function_call_part.function_call.name.into();
627 let next_tool_id =
628 TOOL_CALL_COUNTER.fetch_add(1, atomic::Ordering::SeqCst);
629 let id: LanguageModelToolUseId =
630 format!("{}-{}", name, next_tool_id).into();
631
632 // Normalize empty string signatures to None
633 let thought_signature = function_call_part
634 .thought_signature
635 .filter(|s| !s.is_empty());
636
637 events.push(Ok(LanguageModelCompletionEvent::ToolUse(
638 LanguageModelToolUse {
639 id,
640 name,
641 is_input_complete: true,
642 raw_input: function_call_part.function_call.args.to_string(),
643 input: function_call_part.function_call.args,
644 thought_signature,
645 },
646 )));
647 }
648 Part::FunctionResponsePart(_) => {}
649 Part::ThoughtPart(part) => {
650 events.push(Ok(LanguageModelCompletionEvent::Thinking {
651 text: "(Encrypted thought)".to_string(), // TODO: Can we populate this from thought summaries?
652 signature: Some(part.thought_signature),
653 }));
654 }
655 });
656 }
657 }
658
659 // Even when Gemini wants to use a Tool, the API
660 // responds with `finish_reason: STOP`
661 if wants_to_use_tool {
662 self.stop_reason = StopReason::ToolUse;
663 events.push(Ok(LanguageModelCompletionEvent::Stop(StopReason::ToolUse)));
664 }
665 events
666 }
667}
668
669pub fn count_google_tokens(
670 request: LanguageModelRequest,
671 cx: &App,
672) -> BoxFuture<'static, Result<u64>> {
673 // We couldn't use the GoogleLanguageModelProvider to count tokens because the github copilot doesn't have the access to google_ai directly.
674 // So we have to use tokenizer from tiktoken_rs to count tokens.
675 cx.background_spawn(async move {
676 let messages = request
677 .messages
678 .into_iter()
679 .map(|message| tiktoken_rs::ChatCompletionRequestMessage {
680 role: match message.role {
681 Role::User => "user".into(),
682 Role::Assistant => "assistant".into(),
683 Role::System => "system".into(),
684 },
685 content: Some(message.string_contents()),
686 name: None,
687 function_call: None,
688 })
689 .collect::<Vec<_>>();
690
691 // Tiktoken doesn't yet support these models, so we manually use the
692 // same tokenizer as GPT-4.
693 tiktoken_rs::num_tokens_from_messages("gpt-4", &messages).map(|tokens| tokens as u64)
694 })
695 .boxed()
696}
697
698fn update_usage(usage: &mut UsageMetadata, new: &UsageMetadata) {
699 if let Some(prompt_token_count) = new.prompt_token_count {
700 usage.prompt_token_count = Some(prompt_token_count);
701 }
702 if let Some(cached_content_token_count) = new.cached_content_token_count {
703 usage.cached_content_token_count = Some(cached_content_token_count);
704 }
705 if let Some(candidates_token_count) = new.candidates_token_count {
706 usage.candidates_token_count = Some(candidates_token_count);
707 }
708 if let Some(tool_use_prompt_token_count) = new.tool_use_prompt_token_count {
709 usage.tool_use_prompt_token_count = Some(tool_use_prompt_token_count);
710 }
711 if let Some(thoughts_token_count) = new.thoughts_token_count {
712 usage.thoughts_token_count = Some(thoughts_token_count);
713 }
714 if let Some(total_token_count) = new.total_token_count {
715 usage.total_token_count = Some(total_token_count);
716 }
717}
718
719fn convert_usage(usage: &UsageMetadata) -> language_model::TokenUsage {
720 let prompt_tokens = usage.prompt_token_count.unwrap_or(0);
721 let cached_tokens = usage.cached_content_token_count.unwrap_or(0);
722 let input_tokens = prompt_tokens - cached_tokens;
723 let output_tokens = usage.candidates_token_count.unwrap_or(0);
724
725 language_model::TokenUsage {
726 input_tokens,
727 output_tokens,
728 cache_read_input_tokens: cached_tokens,
729 cache_creation_input_tokens: 0,
730 }
731}
732
733struct ConfigurationView {
734 api_key_editor: Entity<InputField>,
735 state: Entity<State>,
736 target_agent: language_model::ConfigurationViewTargetAgent,
737 load_credentials_task: Option<Task<()>>,
738}
739
740impl ConfigurationView {
741 fn new(
742 state: Entity<State>,
743 target_agent: language_model::ConfigurationViewTargetAgent,
744 window: &mut Window,
745 cx: &mut Context<Self>,
746 ) -> Self {
747 cx.observe(&state, |_, _, cx| {
748 cx.notify();
749 })
750 .detach();
751
752 let load_credentials_task = Some(cx.spawn_in(window, {
753 let state = state.clone();
754 async move |this, cx| {
755 if let Some(task) = Some(state.update(cx, |state, cx| state.authenticate(cx))) {
756 // We don't log an error, because "not signed in" is also an error.
757 let _ = task.await;
758 }
759 this.update(cx, |this, cx| {
760 this.load_credentials_task = None;
761 cx.notify();
762 })
763 .log_err();
764 }
765 }));
766
767 Self {
768 api_key_editor: cx.new(|cx| InputField::new(window, cx, "AIzaSy...")),
769 target_agent,
770 state,
771 load_credentials_task,
772 }
773 }
774
775 fn save_api_key(&mut self, _: &menu::Confirm, window: &mut Window, cx: &mut Context<Self>) {
776 let api_key = self.api_key_editor.read(cx).text(cx).trim().to_string();
777 if api_key.is_empty() {
778 return;
779 }
780
781 // url changes can cause the editor to be displayed again
782 self.api_key_editor
783 .update(cx, |editor, cx| editor.set_text("", window, cx));
784
785 let state = self.state.clone();
786 cx.spawn_in(window, async move |_, cx| {
787 state
788 .update(cx, |state, cx| state.set_api_key(Some(api_key), cx))
789 .await
790 })
791 .detach_and_log_err(cx);
792 }
793
794 fn reset_api_key(&mut self, window: &mut Window, cx: &mut Context<Self>) {
795 self.api_key_editor
796 .update(cx, |editor, cx| editor.set_text("", window, cx));
797
798 let state = self.state.clone();
799 cx.spawn_in(window, async move |_, cx| {
800 state
801 .update(cx, |state, cx| state.set_api_key(None, cx))
802 .await
803 })
804 .detach_and_log_err(cx);
805 }
806
807 fn should_render_editor(&self, cx: &mut Context<Self>) -> bool {
808 !self.state.read(cx).is_authenticated()
809 }
810}
811
812impl Render for ConfigurationView {
813 fn render(&mut self, _: &mut Window, cx: &mut Context<Self>) -> impl IntoElement {
814 let env_var_set = self.state.read(cx).api_key_state.is_from_env_var();
815 let configured_card_label = if env_var_set {
816 format!(
817 "API key set in {} environment variable",
818 API_KEY_ENV_VAR.name
819 )
820 } else {
821 let api_url = GoogleLanguageModelProvider::api_url(cx);
822 if api_url == google_ai::API_URL {
823 "API key configured".to_string()
824 } else {
825 format!("API key configured for {}", api_url)
826 }
827 };
828
829 if self.load_credentials_task.is_some() {
830 div()
831 .child(Label::new("Loading credentials..."))
832 .into_any_element()
833 } else if self.should_render_editor(cx) {
834 v_flex()
835 .size_full()
836 .on_action(cx.listener(Self::save_api_key))
837 .child(Label::new(format!("To use {}, you need to add an API key. Follow these steps:", match &self.target_agent {
838 ConfigurationViewTargetAgent::ZedAgent => "Zed's agent with Google AI".into(),
839 ConfigurationViewTargetAgent::Other(agent) => agent.clone(),
840 })))
841 .child(
842 List::new()
843 .child(
844 ListBulletItem::new("")
845 .child(Label::new("Create one by visiting"))
846 .child(ButtonLink::new("Google AI's console", "https://aistudio.google.com/app/apikey"))
847 )
848 .child(
849 ListBulletItem::new("Paste your API key below and hit enter to start using the agent")
850 )
851 )
852 .child(self.api_key_editor.clone())
853 .child(
854 Label::new(
855 format!("You can also set the {GEMINI_API_KEY_VAR_NAME} environment variable and restart Zed."),
856 )
857 .size(LabelSize::Small).color(Color::Muted),
858 )
859 .into_any_element()
860 } else {
861 ConfiguredApiCard::new(configured_card_label)
862 .disabled(env_var_set)
863 .on_click(cx.listener(|this, _, window, cx| this.reset_api_key(window, cx)))
864 .when(env_var_set, |this| {
865 this.tooltip_label(format!("To reset your API key, make sure {GEMINI_API_KEY_VAR_NAME} and {GOOGLE_AI_API_KEY_VAR_NAME} environment variables are unset."))
866 })
867 .into_any_element()
868 }
869 }
870}
871
872#[cfg(test)]
873mod tests {
874 use super::*;
875 use google_ai::{
876 Content, FunctionCall, FunctionCallPart, GenerateContentCandidate, GenerateContentResponse,
877 Part, Role as GoogleRole, TextPart,
878 };
879 use language_model::{LanguageModelToolUseId, MessageContent, Role};
880 use serde_json::json;
881
882 #[test]
883 fn test_function_call_with_signature_creates_tool_use_with_signature() {
884 let mut mapper = GoogleEventMapper::new();
885
886 let response = GenerateContentResponse {
887 candidates: Some(vec![GenerateContentCandidate {
888 index: Some(0),
889 content: Content {
890 parts: vec![Part::FunctionCallPart(FunctionCallPart {
891 function_call: FunctionCall {
892 name: "test_function".to_string(),
893 args: json!({"arg": "value"}),
894 },
895 thought_signature: Some("test_signature_123".to_string()),
896 })],
897 role: GoogleRole::Model,
898 },
899 finish_reason: None,
900 finish_message: None,
901 safety_ratings: None,
902 citation_metadata: None,
903 }]),
904 prompt_feedback: None,
905 usage_metadata: None,
906 };
907
908 let events = mapper.map_event(response);
909
910 assert_eq!(events.len(), 2); // ToolUse event + Stop event
911
912 if let Ok(LanguageModelCompletionEvent::ToolUse(tool_use)) = &events[0] {
913 assert_eq!(tool_use.name.as_ref(), "test_function");
914 assert_eq!(
915 tool_use.thought_signature.as_deref(),
916 Some("test_signature_123")
917 );
918 } else {
919 panic!("Expected ToolUse event");
920 }
921 }
922
923 #[test]
924 fn test_function_call_without_signature_has_none() {
925 let mut mapper = GoogleEventMapper::new();
926
927 let response = GenerateContentResponse {
928 candidates: Some(vec![GenerateContentCandidate {
929 index: Some(0),
930 content: Content {
931 parts: vec![Part::FunctionCallPart(FunctionCallPart {
932 function_call: FunctionCall {
933 name: "test_function".to_string(),
934 args: json!({"arg": "value"}),
935 },
936 thought_signature: None,
937 })],
938 role: GoogleRole::Model,
939 },
940 finish_reason: None,
941 finish_message: None,
942 safety_ratings: None,
943 citation_metadata: None,
944 }]),
945 prompt_feedback: None,
946 usage_metadata: None,
947 };
948
949 let events = mapper.map_event(response);
950
951 if let Ok(LanguageModelCompletionEvent::ToolUse(tool_use)) = &events[0] {
952 assert_eq!(tool_use.thought_signature, None);
953 } else {
954 panic!("Expected ToolUse event");
955 }
956 }
957
958 #[test]
959 fn test_empty_string_signature_normalized_to_none() {
960 let mut mapper = GoogleEventMapper::new();
961
962 let response = GenerateContentResponse {
963 candidates: Some(vec![GenerateContentCandidate {
964 index: Some(0),
965 content: Content {
966 parts: vec![Part::FunctionCallPart(FunctionCallPart {
967 function_call: FunctionCall {
968 name: "test_function".to_string(),
969 args: json!({"arg": "value"}),
970 },
971 thought_signature: Some("".to_string()),
972 })],
973 role: GoogleRole::Model,
974 },
975 finish_reason: None,
976 finish_message: None,
977 safety_ratings: None,
978 citation_metadata: None,
979 }]),
980 prompt_feedback: None,
981 usage_metadata: None,
982 };
983
984 let events = mapper.map_event(response);
985
986 if let Ok(LanguageModelCompletionEvent::ToolUse(tool_use)) = &events[0] {
987 assert_eq!(tool_use.thought_signature, None);
988 } else {
989 panic!("Expected ToolUse event");
990 }
991 }
992
993 #[test]
994 fn test_parallel_function_calls_preserve_signatures() {
995 let mut mapper = GoogleEventMapper::new();
996
997 let response = GenerateContentResponse {
998 candidates: Some(vec![GenerateContentCandidate {
999 index: Some(0),
1000 content: Content {
1001 parts: vec![
1002 Part::FunctionCallPart(FunctionCallPart {
1003 function_call: FunctionCall {
1004 name: "function_1".to_string(),
1005 args: json!({"arg": "value1"}),
1006 },
1007 thought_signature: Some("signature_1".to_string()),
1008 }),
1009 Part::FunctionCallPart(FunctionCallPart {
1010 function_call: FunctionCall {
1011 name: "function_2".to_string(),
1012 args: json!({"arg": "value2"}),
1013 },
1014 thought_signature: None,
1015 }),
1016 ],
1017 role: GoogleRole::Model,
1018 },
1019 finish_reason: None,
1020 finish_message: None,
1021 safety_ratings: None,
1022 citation_metadata: None,
1023 }]),
1024 prompt_feedback: None,
1025 usage_metadata: None,
1026 };
1027
1028 let events = mapper.map_event(response);
1029
1030 assert_eq!(events.len(), 3); // 2 ToolUse events + Stop event
1031
1032 if let Ok(LanguageModelCompletionEvent::ToolUse(tool_use)) = &events[0] {
1033 assert_eq!(tool_use.name.as_ref(), "function_1");
1034 assert_eq!(tool_use.thought_signature.as_deref(), Some("signature_1"));
1035 } else {
1036 panic!("Expected ToolUse event for function_1");
1037 }
1038
1039 if let Ok(LanguageModelCompletionEvent::ToolUse(tool_use)) = &events[1] {
1040 assert_eq!(tool_use.name.as_ref(), "function_2");
1041 assert_eq!(tool_use.thought_signature, None);
1042 } else {
1043 panic!("Expected ToolUse event for function_2");
1044 }
1045 }
1046
1047 #[test]
1048 fn test_tool_use_with_signature_converts_to_function_call_part() {
1049 let tool_use = language_model::LanguageModelToolUse {
1050 id: LanguageModelToolUseId::from("test_id"),
1051 name: "test_function".into(),
1052 raw_input: json!({"arg": "value"}).to_string(),
1053 input: json!({"arg": "value"}),
1054 is_input_complete: true,
1055 thought_signature: Some("test_signature_456".to_string()),
1056 };
1057
1058 let request = super::into_google(
1059 LanguageModelRequest {
1060 messages: vec![language_model::LanguageModelRequestMessage {
1061 role: Role::Assistant,
1062 content: vec![MessageContent::ToolUse(tool_use)],
1063 cache: false,
1064 reasoning_details: None,
1065 }],
1066 ..Default::default()
1067 },
1068 "gemini-2.5-flash".to_string(),
1069 GoogleModelMode::Default,
1070 );
1071
1072 assert_eq!(request.contents[0].parts.len(), 1);
1073 if let Part::FunctionCallPart(fc_part) = &request.contents[0].parts[0] {
1074 assert_eq!(fc_part.function_call.name, "test_function");
1075 assert_eq!(
1076 fc_part.thought_signature.as_deref(),
1077 Some("test_signature_456")
1078 );
1079 } else {
1080 panic!("Expected FunctionCallPart");
1081 }
1082 }
1083
1084 #[test]
1085 fn test_tool_use_without_signature_omits_field() {
1086 let tool_use = language_model::LanguageModelToolUse {
1087 id: LanguageModelToolUseId::from("test_id"),
1088 name: "test_function".into(),
1089 raw_input: json!({"arg": "value"}).to_string(),
1090 input: json!({"arg": "value"}),
1091 is_input_complete: true,
1092 thought_signature: None,
1093 };
1094
1095 let request = super::into_google(
1096 LanguageModelRequest {
1097 messages: vec![language_model::LanguageModelRequestMessage {
1098 role: Role::Assistant,
1099 content: vec![MessageContent::ToolUse(tool_use)],
1100 cache: false,
1101 reasoning_details: None,
1102 }],
1103 ..Default::default()
1104 },
1105 "gemini-2.5-flash".to_string(),
1106 GoogleModelMode::Default,
1107 );
1108
1109 assert_eq!(request.contents[0].parts.len(), 1);
1110 if let Part::FunctionCallPart(fc_part) = &request.contents[0].parts[0] {
1111 assert_eq!(fc_part.thought_signature, None);
1112 } else {
1113 panic!("Expected FunctionCallPart");
1114 }
1115 }
1116
1117 #[test]
1118 fn test_empty_signature_in_tool_use_normalized_to_none() {
1119 let tool_use = language_model::LanguageModelToolUse {
1120 id: LanguageModelToolUseId::from("test_id"),
1121 name: "test_function".into(),
1122 raw_input: json!({"arg": "value"}).to_string(),
1123 input: json!({"arg": "value"}),
1124 is_input_complete: true,
1125 thought_signature: Some("".to_string()),
1126 };
1127
1128 let request = super::into_google(
1129 LanguageModelRequest {
1130 messages: vec![language_model::LanguageModelRequestMessage {
1131 role: Role::Assistant,
1132 content: vec![MessageContent::ToolUse(tool_use)],
1133 cache: false,
1134 reasoning_details: None,
1135 }],
1136 ..Default::default()
1137 },
1138 "gemini-2.5-flash".to_string(),
1139 GoogleModelMode::Default,
1140 );
1141
1142 if let Part::FunctionCallPart(fc_part) = &request.contents[0].parts[0] {
1143 assert_eq!(fc_part.thought_signature, None);
1144 } else {
1145 panic!("Expected FunctionCallPart");
1146 }
1147 }
1148
1149 #[test]
1150 fn test_round_trip_preserves_signature() {
1151 let mut mapper = GoogleEventMapper::new();
1152
1153 // Simulate receiving a response from Google with a signature
1154 let response = GenerateContentResponse {
1155 candidates: Some(vec![GenerateContentCandidate {
1156 index: Some(0),
1157 content: Content {
1158 parts: vec![Part::FunctionCallPart(FunctionCallPart {
1159 function_call: FunctionCall {
1160 name: "test_function".to_string(),
1161 args: json!({"arg": "value"}),
1162 },
1163 thought_signature: Some("round_trip_sig".to_string()),
1164 })],
1165 role: GoogleRole::Model,
1166 },
1167 finish_reason: None,
1168 finish_message: None,
1169 safety_ratings: None,
1170 citation_metadata: None,
1171 }]),
1172 prompt_feedback: None,
1173 usage_metadata: None,
1174 };
1175
1176 let events = mapper.map_event(response);
1177
1178 let tool_use = if let Ok(LanguageModelCompletionEvent::ToolUse(tool_use)) = &events[0] {
1179 tool_use.clone()
1180 } else {
1181 panic!("Expected ToolUse event");
1182 };
1183
1184 // Convert back to Google format
1185 let request = super::into_google(
1186 LanguageModelRequest {
1187 messages: vec![language_model::LanguageModelRequestMessage {
1188 role: Role::Assistant,
1189 content: vec![MessageContent::ToolUse(tool_use)],
1190 cache: false,
1191 reasoning_details: None,
1192 }],
1193 ..Default::default()
1194 },
1195 "gemini-2.5-flash".to_string(),
1196 GoogleModelMode::Default,
1197 );
1198
1199 // Verify signature is preserved
1200 if let Part::FunctionCallPart(fc_part) = &request.contents[0].parts[0] {
1201 assert_eq!(fc_part.thought_signature.as_deref(), Some("round_trip_sig"));
1202 } else {
1203 panic!("Expected FunctionCallPart");
1204 }
1205 }
1206
1207 #[test]
1208 fn test_mixed_text_and_function_call_with_signature() {
1209 let mut mapper = GoogleEventMapper::new();
1210
1211 let response = GenerateContentResponse {
1212 candidates: Some(vec![GenerateContentCandidate {
1213 index: Some(0),
1214 content: Content {
1215 parts: vec![
1216 Part::TextPart(TextPart {
1217 text: "I'll help with that.".to_string(),
1218 }),
1219 Part::FunctionCallPart(FunctionCallPart {
1220 function_call: FunctionCall {
1221 name: "helper_function".to_string(),
1222 args: json!({"query": "help"}),
1223 },
1224 thought_signature: Some("mixed_sig".to_string()),
1225 }),
1226 ],
1227 role: GoogleRole::Model,
1228 },
1229 finish_reason: None,
1230 finish_message: None,
1231 safety_ratings: None,
1232 citation_metadata: None,
1233 }]),
1234 prompt_feedback: None,
1235 usage_metadata: None,
1236 };
1237
1238 let events = mapper.map_event(response);
1239
1240 assert_eq!(events.len(), 3); // Text event + ToolUse event + Stop event
1241
1242 if let Ok(LanguageModelCompletionEvent::Text(text)) = &events[0] {
1243 assert_eq!(text, "I'll help with that.");
1244 } else {
1245 panic!("Expected Text event");
1246 }
1247
1248 if let Ok(LanguageModelCompletionEvent::ToolUse(tool_use)) = &events[1] {
1249 assert_eq!(tool_use.name.as_ref(), "helper_function");
1250 assert_eq!(tool_use.thought_signature.as_deref(), Some("mixed_sig"));
1251 } else {
1252 panic!("Expected ToolUse event");
1253 }
1254 }
1255
1256 #[test]
1257 fn test_special_characters_in_signature_preserved() {
1258 let mut mapper = GoogleEventMapper::new();
1259
1260 let signature_with_special_chars = "sig<>\"'&%$#@!{}[]".to_string();
1261
1262 let response = GenerateContentResponse {
1263 candidates: Some(vec![GenerateContentCandidate {
1264 index: Some(0),
1265 content: Content {
1266 parts: vec![Part::FunctionCallPart(FunctionCallPart {
1267 function_call: FunctionCall {
1268 name: "test_function".to_string(),
1269 args: json!({"arg": "value"}),
1270 },
1271 thought_signature: Some(signature_with_special_chars.clone()),
1272 })],
1273 role: GoogleRole::Model,
1274 },
1275 finish_reason: None,
1276 finish_message: None,
1277 safety_ratings: None,
1278 citation_metadata: None,
1279 }]),
1280 prompt_feedback: None,
1281 usage_metadata: None,
1282 };
1283
1284 let events = mapper.map_event(response);
1285
1286 if let Ok(LanguageModelCompletionEvent::ToolUse(tool_use)) = &events[0] {
1287 assert_eq!(
1288 tool_use.thought_signature.as_deref(),
1289 Some(signature_with_special_chars.as_str())
1290 );
1291 } else {
1292 panic!("Expected ToolUse event");
1293 }
1294 }
1295}