1use anyhow::Result;
2use convert_case::{Case, Casing};
3use futures::{FutureExt, StreamExt, future::BoxFuture};
4use gpui::{AnyView, App, AsyncApp, Context, Entity, SharedString, Task, Window};
5use http_client::HttpClient;
6use language_model::{
7 ApiKeyState, AuthenticateError, EnvVar, IconOrSvg, LanguageModel, LanguageModelCompletionError,
8 LanguageModelCompletionEvent, LanguageModelId, LanguageModelName, LanguageModelProvider,
9 LanguageModelProviderId, LanguageModelProviderName, LanguageModelProviderState,
10 LanguageModelRequest, LanguageModelToolChoice, LanguageModelToolSchemaFormat, RateLimiter,
11};
12use menu;
13use open_ai::{
14 ResponseStreamEvent,
15 responses::{Request as ResponseRequest, StreamEvent as ResponsesStreamEvent, stream_response},
16 stream_completion,
17};
18use settings::{Settings, SettingsStore};
19use std::sync::Arc;
20use ui::{ElevationIndex, Tooltip, prelude::*};
21use ui_input::InputField;
22use util::ResultExt;
23
24use crate::provider::open_ai::{
25 OpenAiEventMapper, OpenAiResponseEventMapper, into_open_ai, into_open_ai_response,
26};
27pub use settings::OpenAiCompatibleAvailableModel as AvailableModel;
28pub use settings::OpenAiCompatibleModelCapabilities as ModelCapabilities;
29
30#[derive(Default, Clone, Debug, PartialEq)]
31pub struct OpenAiCompatibleSettings {
32 pub api_url: String,
33 pub available_models: Vec<AvailableModel>,
34}
35
36pub struct OpenAiCompatibleLanguageModelProvider {
37 id: LanguageModelProviderId,
38 name: LanguageModelProviderName,
39 http_client: Arc<dyn HttpClient>,
40 state: Entity<State>,
41}
42
43pub struct State {
44 id: Arc<str>,
45 api_key_state: ApiKeyState,
46 settings: OpenAiCompatibleSettings,
47}
48
49impl State {
50 fn is_authenticated(&self) -> bool {
51 self.api_key_state.has_key()
52 }
53
54 fn set_api_key(&mut self, api_key: Option<String>, cx: &mut Context<Self>) -> Task<Result<()>> {
55 let api_url = SharedString::new(self.settings.api_url.as_str());
56 self.api_key_state
57 .store(api_url, api_key, |this| &mut this.api_key_state, cx)
58 }
59
60 fn authenticate(&mut self, cx: &mut Context<Self>) -> Task<Result<(), AuthenticateError>> {
61 let api_url = SharedString::new(self.settings.api_url.clone());
62 self.api_key_state
63 .load_if_needed(api_url, |this| &mut this.api_key_state, cx)
64 }
65}
66
67impl OpenAiCompatibleLanguageModelProvider {
68 pub fn new(id: Arc<str>, http_client: Arc<dyn HttpClient>, cx: &mut App) -> Self {
69 fn resolve_settings<'a>(id: &'a str, cx: &'a App) -> Option<&'a OpenAiCompatibleSettings> {
70 crate::AllLanguageModelSettings::get_global(cx)
71 .openai_compatible
72 .get(id)
73 }
74
75 let api_key_env_var_name = format!("{}_API_KEY", id).to_case(Case::UpperSnake).into();
76 let state = cx.new(|cx| {
77 cx.observe_global::<SettingsStore>(|this: &mut State, cx| {
78 let Some(settings) = resolve_settings(&this.id, cx).cloned() else {
79 return;
80 };
81 if &this.settings != &settings {
82 let api_url = SharedString::new(settings.api_url.as_str());
83 this.api_key_state.handle_url_change(
84 api_url,
85 |this| &mut this.api_key_state,
86 cx,
87 );
88 this.settings = settings;
89 cx.notify();
90 }
91 })
92 .detach();
93 let settings = resolve_settings(&id, cx).cloned().unwrap_or_default();
94 State {
95 id: id.clone(),
96 api_key_state: ApiKeyState::new(
97 SharedString::new(settings.api_url.as_str()),
98 EnvVar::new(api_key_env_var_name),
99 ),
100 settings,
101 }
102 });
103
104 Self {
105 id: id.clone().into(),
106 name: id.into(),
107 http_client,
108 state,
109 }
110 }
111
112 fn create_language_model(&self, model: AvailableModel) -> Arc<dyn LanguageModel> {
113 Arc::new(OpenAiCompatibleLanguageModel {
114 id: LanguageModelId::from(model.name.clone()),
115 provider_id: self.id.clone(),
116 provider_name: self.name.clone(),
117 model,
118 state: self.state.clone(),
119 http_client: self.http_client.clone(),
120 request_limiter: RateLimiter::new(4),
121 })
122 }
123}
124
125impl LanguageModelProviderState for OpenAiCompatibleLanguageModelProvider {
126 type ObservableEntity = State;
127
128 fn observable_entity(&self) -> Option<Entity<Self::ObservableEntity>> {
129 Some(self.state.clone())
130 }
131}
132
133impl LanguageModelProvider for OpenAiCompatibleLanguageModelProvider {
134 fn id(&self) -> LanguageModelProviderId {
135 self.id.clone()
136 }
137
138 fn name(&self) -> LanguageModelProviderName {
139 self.name.clone()
140 }
141
142 fn icon(&self) -> IconOrSvg {
143 IconOrSvg::Icon(IconName::AiOpenAiCompat)
144 }
145
146 fn default_model(&self, cx: &App) -> Option<Arc<dyn LanguageModel>> {
147 self.state
148 .read(cx)
149 .settings
150 .available_models
151 .first()
152 .map(|model| self.create_language_model(model.clone()))
153 }
154
155 fn default_fast_model(&self, _cx: &App) -> Option<Arc<dyn LanguageModel>> {
156 None
157 }
158
159 fn provided_models(&self, cx: &App) -> Vec<Arc<dyn LanguageModel>> {
160 self.state
161 .read(cx)
162 .settings
163 .available_models
164 .iter()
165 .map(|model| self.create_language_model(model.clone()))
166 .collect()
167 }
168
169 fn is_authenticated(&self, cx: &App) -> bool {
170 self.state.read(cx).is_authenticated()
171 }
172
173 fn authenticate(&self, cx: &mut App) -> Task<Result<(), AuthenticateError>> {
174 self.state.update(cx, |state, cx| state.authenticate(cx))
175 }
176
177 fn configuration_view(
178 &self,
179 _target_agent: language_model::ConfigurationViewTargetAgent,
180 window: &mut Window,
181 cx: &mut App,
182 ) -> AnyView {
183 cx.new(|cx| ConfigurationView::new(self.state.clone(), window, cx))
184 .into()
185 }
186
187 fn reset_credentials(&self, cx: &mut App) -> Task<Result<()>> {
188 self.state
189 .update(cx, |state, cx| state.set_api_key(None, cx))
190 }
191}
192
193pub struct OpenAiCompatibleLanguageModel {
194 id: LanguageModelId,
195 provider_id: LanguageModelProviderId,
196 provider_name: LanguageModelProviderName,
197 model: AvailableModel,
198 state: Entity<State>,
199 http_client: Arc<dyn HttpClient>,
200 request_limiter: RateLimiter,
201}
202
203impl OpenAiCompatibleLanguageModel {
204 fn stream_completion(
205 &self,
206 request: open_ai::Request,
207 cx: &AsyncApp,
208 ) -> BoxFuture<
209 'static,
210 Result<
211 futures::stream::BoxStream<'static, Result<ResponseStreamEvent>>,
212 LanguageModelCompletionError,
213 >,
214 > {
215 let http_client = self.http_client.clone();
216
217 let (api_key, api_url) = self.state.read_with(cx, |state, _cx| {
218 let api_url = &state.settings.api_url;
219 (
220 state.api_key_state.key(api_url),
221 state.settings.api_url.clone(),
222 )
223 });
224
225 let provider = self.provider_name.clone();
226 let future = self.request_limiter.stream(async move {
227 let Some(api_key) = api_key else {
228 return Err(LanguageModelCompletionError::NoApiKey { provider });
229 };
230 let request = stream_completion(
231 http_client.as_ref(),
232 provider.0.as_str(),
233 &api_url,
234 &api_key,
235 request,
236 );
237 let response = request.await?;
238 Ok(response)
239 });
240
241 async move { Ok(future.await?.boxed()) }.boxed()
242 }
243
244 fn stream_response(
245 &self,
246 request: ResponseRequest,
247 cx: &AsyncApp,
248 ) -> BoxFuture<'static, Result<futures::stream::BoxStream<'static, Result<ResponsesStreamEvent>>>>
249 {
250 let http_client = self.http_client.clone();
251
252 let (api_key, api_url) = self.state.read_with(cx, |state, _cx| {
253 let api_url = &state.settings.api_url;
254 (
255 state.api_key_state.key(api_url),
256 state.settings.api_url.clone(),
257 )
258 });
259
260 let provider = self.provider_name.clone();
261 let future = self.request_limiter.stream(async move {
262 let Some(api_key) = api_key else {
263 return Err(LanguageModelCompletionError::NoApiKey { provider });
264 };
265 let request = stream_response(
266 http_client.as_ref(),
267 provider.0.as_str(),
268 &api_url,
269 &api_key,
270 request,
271 );
272 let response = request.await?;
273 Ok(response)
274 });
275
276 async move { Ok(future.await?.boxed()) }.boxed()
277 }
278}
279
280impl LanguageModel for OpenAiCompatibleLanguageModel {
281 fn id(&self) -> LanguageModelId {
282 self.id.clone()
283 }
284
285 fn name(&self) -> LanguageModelName {
286 LanguageModelName::from(
287 self.model
288 .display_name
289 .clone()
290 .unwrap_or_else(|| self.model.name.clone()),
291 )
292 }
293
294 fn provider_id(&self) -> LanguageModelProviderId {
295 self.provider_id.clone()
296 }
297
298 fn provider_name(&self) -> LanguageModelProviderName {
299 self.provider_name.clone()
300 }
301
302 fn supports_tools(&self) -> bool {
303 self.model.capabilities.tools
304 }
305
306 fn tool_input_format(&self) -> LanguageModelToolSchemaFormat {
307 LanguageModelToolSchemaFormat::JsonSchemaSubset
308 }
309
310 fn supports_images(&self) -> bool {
311 self.model.capabilities.images
312 }
313
314 fn supports_tool_choice(&self, choice: LanguageModelToolChoice) -> bool {
315 match choice {
316 LanguageModelToolChoice::Auto => self.model.capabilities.tools,
317 LanguageModelToolChoice::Any => self.model.capabilities.tools,
318 LanguageModelToolChoice::None => true,
319 }
320 }
321
322 fn supports_streaming_tools(&self) -> bool {
323 true
324 }
325
326 fn supports_split_token_display(&self) -> bool {
327 true
328 }
329
330 fn telemetry_id(&self) -> String {
331 format!("openai/{}", self.model.name)
332 }
333
334 fn max_token_count(&self) -> u64 {
335 self.model.max_tokens
336 }
337
338 fn max_output_tokens(&self) -> Option<u64> {
339 self.model.max_output_tokens
340 }
341
342 fn count_tokens(
343 &self,
344 request: LanguageModelRequest,
345 cx: &App,
346 ) -> BoxFuture<'static, Result<u64>> {
347 let max_token_count = self.max_token_count();
348 cx.background_spawn(async move {
349 let messages = super::open_ai::collect_tiktoken_messages(request);
350 let model = if max_token_count >= 100_000 {
351 // If the max tokens is 100k or more, it is likely the o200k_base tokenizer from gpt4o
352 "gpt-4o"
353 } else {
354 // Otherwise fallback to gpt-4, since only cl100k_base and o200k_base are
355 // supported with this tiktoken method
356 "gpt-4"
357 };
358 tiktoken_rs::num_tokens_from_messages(model, &messages).map(|tokens| tokens as u64)
359 })
360 .boxed()
361 }
362
363 fn stream_completion(
364 &self,
365 request: LanguageModelRequest,
366 cx: &AsyncApp,
367 ) -> BoxFuture<
368 'static,
369 Result<
370 futures::stream::BoxStream<
371 'static,
372 Result<LanguageModelCompletionEvent, LanguageModelCompletionError>,
373 >,
374 LanguageModelCompletionError,
375 >,
376 > {
377 if self.model.capabilities.chat_completions {
378 let request = into_open_ai(
379 request,
380 &self.model.name,
381 self.model.capabilities.parallel_tool_calls,
382 self.model.capabilities.prompt_cache_key,
383 self.max_output_tokens(),
384 None,
385 );
386 let completions = self.stream_completion(request, cx);
387 async move {
388 let mapper = OpenAiEventMapper::new();
389 Ok(mapper.map_stream(completions.await?).boxed())
390 }
391 .boxed()
392 } else {
393 let request = into_open_ai_response(
394 request,
395 &self.model.name,
396 self.model.capabilities.parallel_tool_calls,
397 self.model.capabilities.prompt_cache_key,
398 self.max_output_tokens(),
399 None,
400 );
401 let completions = self.stream_response(request, cx);
402 async move {
403 let mapper = OpenAiResponseEventMapper::new();
404 Ok(mapper.map_stream(completions.await?).boxed())
405 }
406 .boxed()
407 }
408 }
409}
410
411struct ConfigurationView {
412 api_key_editor: Entity<InputField>,
413 state: Entity<State>,
414 load_credentials_task: Option<Task<()>>,
415}
416
417impl ConfigurationView {
418 fn new(state: Entity<State>, window: &mut Window, cx: &mut Context<Self>) -> Self {
419 let api_key_editor = cx.new(|cx| {
420 InputField::new(
421 window,
422 cx,
423 "000000000000000000000000000000000000000000000000000",
424 )
425 });
426
427 cx.observe(&state, |_, _, cx| {
428 cx.notify();
429 })
430 .detach();
431
432 let load_credentials_task = Some(cx.spawn_in(window, {
433 let state = state.clone();
434 async move |this, cx| {
435 if let Some(task) = Some(state.update(cx, |state, cx| state.authenticate(cx))) {
436 // We don't log an error, because "not signed in" is also an error.
437 let _ = task.await;
438 }
439 this.update(cx, |this, cx| {
440 this.load_credentials_task = None;
441 cx.notify();
442 })
443 .log_err();
444 }
445 }));
446
447 Self {
448 api_key_editor,
449 state,
450 load_credentials_task,
451 }
452 }
453
454 fn save_api_key(&mut self, _: &menu::Confirm, window: &mut Window, cx: &mut Context<Self>) {
455 let api_key = self.api_key_editor.read(cx).text(cx).trim().to_string();
456 if api_key.is_empty() {
457 return;
458 }
459
460 // url changes can cause the editor to be displayed again
461 self.api_key_editor
462 .update(cx, |input, cx| input.set_text("", window, cx));
463
464 let state = self.state.clone();
465 cx.spawn_in(window, async move |_, cx| {
466 state
467 .update(cx, |state, cx| state.set_api_key(Some(api_key), cx))
468 .await
469 })
470 .detach_and_log_err(cx);
471 }
472
473 fn reset_api_key(&mut self, window: &mut Window, cx: &mut Context<Self>) {
474 self.api_key_editor
475 .update(cx, |input, cx| input.set_text("", window, cx));
476
477 let state = self.state.clone();
478 cx.spawn_in(window, async move |_, cx| {
479 state
480 .update(cx, |state, cx| state.set_api_key(None, cx))
481 .await
482 })
483 .detach_and_log_err(cx);
484 }
485
486 fn should_render_editor(&self, cx: &Context<Self>) -> bool {
487 !self.state.read(cx).is_authenticated()
488 }
489}
490
491impl Render for ConfigurationView {
492 fn render(&mut self, _: &mut Window, cx: &mut Context<Self>) -> impl IntoElement {
493 let state = self.state.read(cx);
494 let env_var_set = state.api_key_state.is_from_env_var();
495 let env_var_name = state.api_key_state.env_var_name();
496
497 let api_key_section = if self.should_render_editor(cx) {
498 v_flex()
499 .on_action(cx.listener(Self::save_api_key))
500 .child(Label::new("To use Zed's agent with an OpenAI-compatible provider, you need to add an API key."))
501 .child(
502 div()
503 .pt(DynamicSpacing::Base04.rems(cx))
504 .child(self.api_key_editor.clone())
505 )
506 .child(
507 Label::new(
508 format!("You can also set the {env_var_name} environment variable and restart Zed."),
509 )
510 .size(LabelSize::Small).color(Color::Muted),
511 )
512 .into_any()
513 } else {
514 h_flex()
515 .mt_1()
516 .p_1()
517 .justify_between()
518 .rounded_md()
519 .border_1()
520 .border_color(cx.theme().colors().border)
521 .bg(cx.theme().colors().background)
522 .child(
523 h_flex()
524 .flex_1()
525 .min_w_0()
526 .gap_1()
527 .child(Icon::new(IconName::Check).color(Color::Success))
528 .child(
529 div()
530 .w_full()
531 .overflow_x_hidden()
532 .text_ellipsis()
533 .child(Label::new(
534 if env_var_set {
535 format!("API key set in {env_var_name} environment variable")
536 } else {
537 format!("API key configured for {}", &state.settings.api_url)
538 }
539 ))
540 ),
541 )
542 .child(
543 h_flex()
544 .flex_shrink_0()
545 .child(
546 Button::new("reset-api-key", "Reset API Key")
547 .label_size(LabelSize::Small)
548 .icon(IconName::Undo)
549 .icon_size(IconSize::Small)
550 .icon_position(IconPosition::Start)
551 .layer(ElevationIndex::ModalSurface)
552 .when(env_var_set, |this| {
553 this.tooltip(Tooltip::text(format!("To reset your API key, unset the {env_var_name} environment variable.")))
554 })
555 .on_click(cx.listener(|this, _, window, cx| this.reset_api_key(window, cx))),
556 ),
557 )
558 .into_any()
559 };
560
561 if self.load_credentials_task.is_some() {
562 div().child(Label::new("Loading credentials…")).into_any()
563 } else {
564 v_flex().size_full().child(api_key_section).into_any()
565 }
566 }
567}