1use anyhow::{Result, anyhow};
2use convert_case::{Case, Casing};
3use futures::{FutureExt, StreamExt, future, future::BoxFuture};
4use gpui::{AnyView, App, AsyncApp, Context, Entity, SharedString, Task, Window};
5use http_client::HttpClient;
6use language_model::{
7 AuthenticateError, LanguageModel, LanguageModelCompletionError, LanguageModelCompletionEvent,
8 LanguageModelId, LanguageModelName, LanguageModelProvider, LanguageModelProviderId,
9 LanguageModelProviderName, LanguageModelProviderState, LanguageModelRequest,
10 LanguageModelToolChoice, LanguageModelToolSchemaFormat, RateLimiter,
11};
12use menu;
13use open_ai::{ResponseStreamEvent, stream_completion};
14use settings::{Settings, SettingsStore};
15use std::sync::Arc;
16use ui::{ElevationIndex, Tooltip, prelude::*};
17use ui_input::InputField;
18use util::ResultExt;
19use zed_env_vars::EnvVar;
20
21use crate::api_key::ApiKeyState;
22use crate::provider::open_ai::{OpenAiEventMapper, into_open_ai};
23pub use settings::OpenAiCompatibleAvailableModel as AvailableModel;
24pub use settings::OpenAiCompatibleModelCapabilities as ModelCapabilities;
25
26#[derive(Default, Clone, Debug, PartialEq)]
27pub struct OpenAiCompatibleSettings {
28 pub api_url: String,
29 pub available_models: Vec<AvailableModel>,
30}
31
32pub struct OpenAiCompatibleLanguageModelProvider {
33 id: LanguageModelProviderId,
34 name: LanguageModelProviderName,
35 http_client: Arc<dyn HttpClient>,
36 state: Entity<State>,
37}
38
39pub struct State {
40 id: Arc<str>,
41 api_key_env_var: EnvVar,
42 api_key_state: ApiKeyState,
43 settings: OpenAiCompatibleSettings,
44}
45
46impl State {
47 fn is_authenticated(&self) -> bool {
48 self.api_key_state.has_key()
49 }
50
51 fn set_api_key(&mut self, api_key: Option<String>, cx: &mut Context<Self>) -> Task<Result<()>> {
52 let api_url = SharedString::new(self.settings.api_url.as_str());
53 self.api_key_state
54 .store(api_url, api_key, |this| &mut this.api_key_state, cx)
55 }
56
57 fn authenticate(&mut self, cx: &mut Context<Self>) -> Task<Result<(), AuthenticateError>> {
58 let api_url = SharedString::new(self.settings.api_url.clone());
59 self.api_key_state.load_if_needed(
60 api_url,
61 &self.api_key_env_var,
62 |this| &mut this.api_key_state,
63 cx,
64 )
65 }
66}
67
68impl OpenAiCompatibleLanguageModelProvider {
69 pub fn new(id: Arc<str>, http_client: Arc<dyn HttpClient>, cx: &mut App) -> Self {
70 fn resolve_settings<'a>(id: &'a str, cx: &'a App) -> Option<&'a OpenAiCompatibleSettings> {
71 crate::AllLanguageModelSettings::get_global(cx)
72 .openai_compatible
73 .get(id)
74 }
75
76 let api_key_env_var_name = format!("{}_API_KEY", id).to_case(Case::UpperSnake).into();
77 let state = cx.new(|cx| {
78 cx.observe_global::<SettingsStore>(|this: &mut State, cx| {
79 let Some(settings) = resolve_settings(&this.id, cx).cloned() else {
80 return;
81 };
82 if &this.settings != &settings {
83 let api_url = SharedString::new(settings.api_url.as_str());
84 this.api_key_state.handle_url_change(
85 api_url,
86 &this.api_key_env_var,
87 |this| &mut this.api_key_state,
88 cx,
89 );
90 this.settings = settings;
91 cx.notify();
92 }
93 })
94 .detach();
95 let settings = resolve_settings(&id, cx).cloned().unwrap_or_default();
96 State {
97 id: id.clone(),
98 api_key_env_var: EnvVar::new(api_key_env_var_name),
99 api_key_state: ApiKeyState::new(SharedString::new(settings.api_url.as_str())),
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) -> IconName {
143 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 Ok((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 }) else {
224 return future::ready(Err(anyhow!("App state dropped").into())).boxed();
225 };
226
227 let provider = self.provider_name.clone();
228 let future = self.request_limiter.stream(async move {
229 let Some(api_key) = api_key else {
230 return Err(LanguageModelCompletionError::NoApiKey { provider });
231 };
232 let request = stream_completion(
233 http_client.as_ref(),
234 provider.0.as_str(),
235 &api_url,
236 &api_key,
237 request,
238 );
239 let response = request.await?;
240 Ok(response)
241 });
242
243 async move { Ok(future.await?.boxed()) }.boxed()
244 }
245}
246
247impl LanguageModel for OpenAiCompatibleLanguageModel {
248 fn id(&self) -> LanguageModelId {
249 self.id.clone()
250 }
251
252 fn name(&self) -> LanguageModelName {
253 LanguageModelName::from(
254 self.model
255 .display_name
256 .clone()
257 .unwrap_or_else(|| self.model.name.clone()),
258 )
259 }
260
261 fn provider_id(&self) -> LanguageModelProviderId {
262 self.provider_id.clone()
263 }
264
265 fn provider_name(&self) -> LanguageModelProviderName {
266 self.provider_name.clone()
267 }
268
269 fn supports_tools(&self) -> bool {
270 self.model.capabilities.tools
271 }
272
273 fn tool_input_format(&self) -> LanguageModelToolSchemaFormat {
274 LanguageModelToolSchemaFormat::JsonSchemaSubset
275 }
276
277 fn supports_images(&self) -> bool {
278 self.model.capabilities.images
279 }
280
281 fn supports_tool_choice(&self, choice: LanguageModelToolChoice) -> bool {
282 match choice {
283 LanguageModelToolChoice::Auto => self.model.capabilities.tools,
284 LanguageModelToolChoice::Any => self.model.capabilities.tools,
285 LanguageModelToolChoice::None => true,
286 }
287 }
288
289 fn telemetry_id(&self) -> String {
290 format!("openai/{}", self.model.name)
291 }
292
293 fn max_token_count(&self) -> u64 {
294 self.model.max_tokens
295 }
296
297 fn max_output_tokens(&self) -> Option<u64> {
298 self.model.max_output_tokens
299 }
300
301 fn count_tokens(
302 &self,
303 request: LanguageModelRequest,
304 cx: &App,
305 ) -> BoxFuture<'static, Result<u64>> {
306 let max_token_count = self.max_token_count();
307 cx.background_spawn(async move {
308 let messages = super::open_ai::collect_tiktoken_messages(request);
309 let model = if max_token_count >= 100_000 {
310 // If the max tokens is 100k or more, it is likely the o200k_base tokenizer from gpt4o
311 "gpt-4o"
312 } else {
313 // Otherwise fallback to gpt-4, since only cl100k_base and o200k_base are
314 // supported with this tiktoken method
315 "gpt-4"
316 };
317 tiktoken_rs::num_tokens_from_messages(model, &messages).map(|tokens| tokens as u64)
318 })
319 .boxed()
320 }
321
322 fn stream_completion(
323 &self,
324 request: LanguageModelRequest,
325 cx: &AsyncApp,
326 ) -> BoxFuture<
327 'static,
328 Result<
329 futures::stream::BoxStream<
330 'static,
331 Result<LanguageModelCompletionEvent, LanguageModelCompletionError>,
332 >,
333 LanguageModelCompletionError,
334 >,
335 > {
336 let request = into_open_ai(
337 request,
338 &self.model.name,
339 self.model.capabilities.parallel_tool_calls,
340 self.model.capabilities.prompt_cache_key,
341 self.max_output_tokens(),
342 None,
343 );
344 let completions = self.stream_completion(request, cx);
345 async move {
346 let mapper = OpenAiEventMapper::new();
347 Ok(mapper.map_stream(completions.await?).boxed())
348 }
349 .boxed()
350 }
351}
352
353struct ConfigurationView {
354 api_key_editor: Entity<InputField>,
355 state: Entity<State>,
356 load_credentials_task: Option<Task<()>>,
357}
358
359impl ConfigurationView {
360 fn new(state: Entity<State>, window: &mut Window, cx: &mut Context<Self>) -> Self {
361 let api_key_editor = cx.new(|cx| {
362 InputField::new(
363 window,
364 cx,
365 "000000000000000000000000000000000000000000000000000",
366 )
367 });
368
369 cx.observe(&state, |_, _, cx| {
370 cx.notify();
371 })
372 .detach();
373
374 let load_credentials_task = Some(cx.spawn_in(window, {
375 let state = state.clone();
376 async move |this, cx| {
377 if let Some(task) = state
378 .update(cx, |state, cx| state.authenticate(cx))
379 .log_err()
380 {
381 // We don't log an error, because "not signed in" is also an error.
382 let _ = task.await;
383 }
384 this.update(cx, |this, cx| {
385 this.load_credentials_task = None;
386 cx.notify();
387 })
388 .log_err();
389 }
390 }));
391
392 Self {
393 api_key_editor,
394 state,
395 load_credentials_task,
396 }
397 }
398
399 fn save_api_key(&mut self, _: &menu::Confirm, window: &mut Window, cx: &mut Context<Self>) {
400 let api_key = self.api_key_editor.read(cx).text(cx).trim().to_string();
401 if api_key.is_empty() {
402 return;
403 }
404
405 // url changes can cause the editor to be displayed again
406 self.api_key_editor
407 .update(cx, |input, cx| input.set_text("", window, cx));
408
409 let state = self.state.clone();
410 cx.spawn_in(window, async move |_, cx| {
411 state
412 .update(cx, |state, cx| state.set_api_key(Some(api_key), cx))?
413 .await
414 })
415 .detach_and_log_err(cx);
416 }
417
418 fn reset_api_key(&mut self, window: &mut Window, cx: &mut Context<Self>) {
419 self.api_key_editor
420 .update(cx, |input, cx| input.set_text("", window, cx));
421
422 let state = self.state.clone();
423 cx.spawn_in(window, async move |_, cx| {
424 state
425 .update(cx, |state, cx| state.set_api_key(None, cx))?
426 .await
427 })
428 .detach_and_log_err(cx);
429 }
430
431 fn should_render_editor(&self, cx: &Context<Self>) -> bool {
432 !self.state.read(cx).is_authenticated()
433 }
434}
435
436impl Render for ConfigurationView {
437 fn render(&mut self, _: &mut Window, cx: &mut Context<Self>) -> impl IntoElement {
438 let state = self.state.read(cx);
439 let env_var_set = state.api_key_state.is_from_env_var();
440 let env_var_name = &state.api_key_env_var.name;
441
442 let api_key_section = if self.should_render_editor(cx) {
443 v_flex()
444 .on_action(cx.listener(Self::save_api_key))
445 .child(Label::new("To use Zed's agent with an OpenAI-compatible provider, you need to add an API key."))
446 .child(
447 div()
448 .pt(DynamicSpacing::Base04.rems(cx))
449 .child(self.api_key_editor.clone())
450 )
451 .child(
452 Label::new(
453 format!("You can also assign the {env_var_name} environment variable and restart Zed."),
454 )
455 .size(LabelSize::Small).color(Color::Muted),
456 )
457 .into_any()
458 } else {
459 h_flex()
460 .mt_1()
461 .p_1()
462 .justify_between()
463 .rounded_md()
464 .border_1()
465 .border_color(cx.theme().colors().border)
466 .bg(cx.theme().colors().background)
467 .child(
468 h_flex()
469 .flex_1()
470 .min_w_0()
471 .gap_1()
472 .child(Icon::new(IconName::Check).color(Color::Success))
473 .child(
474 div()
475 .w_full()
476 .overflow_x_hidden()
477 .text_ellipsis()
478 .child(Label::new(
479 if env_var_set {
480 format!("API key set in {env_var_name} environment variable")
481 } else {
482 format!("API key configured for {}", &state.settings.api_url)
483 }
484 ))
485 ),
486 )
487 .child(
488 h_flex()
489 .flex_shrink_0()
490 .child(
491 Button::new("reset-api-key", "Reset API Key")
492 .label_size(LabelSize::Small)
493 .icon(IconName::Undo)
494 .icon_size(IconSize::Small)
495 .icon_position(IconPosition::Start)
496 .layer(ElevationIndex::ModalSurface)
497 .when(env_var_set, |this| {
498 this.tooltip(Tooltip::text(format!("To reset your API key, unset the {env_var_name} environment variable.")))
499 })
500 .on_click(cx.listener(|this, _, window, cx| this.reset_api_key(window, cx))),
501 ),
502 )
503 .into_any()
504 };
505
506 if self.load_credentials_task.is_some() {
507 div().child(Label::new("Loading credentials…")).into_any()
508 } else {
509 v_flex().size_full().child(api_key_section).into_any()
510 }
511 }
512}