1use anyhow::{Result, anyhow};
2use credentials_provider::CredentialsProvider;
3use fs::Fs;
4use futures::{FutureExt, StreamExt, future::BoxFuture, stream::BoxStream};
5use futures::{Stream, TryFutureExt, stream};
6use gpui::{AnyView, App, AsyncApp, Context, CursorStyle, Entity, Task};
7use http_client::HttpClient;
8use language_model::{
9 ApiKeyState, AuthenticateError, EnvVar, IconOrSvg, LanguageModel, LanguageModelCompletionError,
10 LanguageModelCompletionEvent, LanguageModelId, LanguageModelName, LanguageModelProvider,
11 LanguageModelProviderId, LanguageModelProviderName, LanguageModelProviderState,
12 LanguageModelRequest, LanguageModelRequestTool, LanguageModelToolChoice, LanguageModelToolUse,
13 LanguageModelToolUseId, MessageContent, RateLimiter, Role, StopReason, TokenUsage, env_var,
14};
15use menu;
16use ollama::{
17 ChatMessage, ChatOptions, ChatRequest, ChatResponseDelta, OLLAMA_API_URL, OllamaFunctionCall,
18 OllamaFunctionTool, OllamaToolCall, get_models, show_model, stream_chat_completion,
19};
20pub use settings::OllamaAvailableModel as AvailableModel;
21use settings::{Settings, SettingsStore, update_settings_file};
22use std::pin::Pin;
23use std::sync::LazyLock;
24use std::{collections::HashMap, sync::Arc};
25use ui::{
26 ButtonLike, ButtonLink, ConfiguredApiCard, ElevationIndex, List, ListBulletItem, Tooltip,
27 prelude::*,
28};
29use ui_input::InputField;
30
31use crate::AllLanguageModelSettings;
32
33const OLLAMA_DOWNLOAD_URL: &str = "https://ollama.com/download";
34const OLLAMA_LIBRARY_URL: &str = "https://ollama.com/library";
35const OLLAMA_SITE: &str = "https://ollama.com/";
36
37const PROVIDER_ID: LanguageModelProviderId = LanguageModelProviderId::new("ollama");
38const PROVIDER_NAME: LanguageModelProviderName = LanguageModelProviderName::new("Ollama");
39
40const API_KEY_ENV_VAR_NAME: &str = "OLLAMA_API_KEY";
41static API_KEY_ENV_VAR: LazyLock<EnvVar> = env_var!(API_KEY_ENV_VAR_NAME);
42
43#[derive(Default, Debug, Clone, PartialEq)]
44pub struct OllamaSettings {
45 pub api_url: String,
46 pub auto_discover: bool,
47 pub available_models: Vec<AvailableModel>,
48 pub context_window: Option<u64>,
49}
50
51pub struct OllamaLanguageModelProvider {
52 http_client: Arc<dyn HttpClient>,
53 state: Entity<State>,
54}
55
56pub struct State {
57 api_key_state: ApiKeyState,
58 credentials_provider: Arc<dyn CredentialsProvider>,
59 http_client: Arc<dyn HttpClient>,
60 fetched_models: Vec<ollama::Model>,
61 fetch_model_task: Option<Task<Result<()>>>,
62}
63
64impl State {
65 fn is_authenticated(&self) -> bool {
66 !self.fetched_models.is_empty()
67 }
68
69 fn set_api_key(&mut self, api_key: Option<String>, cx: &mut Context<Self>) -> Task<Result<()>> {
70 let credentials_provider = self.credentials_provider.clone();
71 let api_url = OllamaLanguageModelProvider::api_url(cx);
72 let task = self.api_key_state.store(
73 api_url,
74 api_key,
75 |this| &mut this.api_key_state,
76 credentials_provider,
77 cx,
78 );
79
80 self.fetched_models.clear();
81 cx.spawn(async move |this, cx| {
82 let result = task.await;
83 this.update(cx, |this, cx| this.restart_fetch_models_task(cx))
84 .ok();
85 result
86 })
87 }
88
89 fn authenticate(&mut self, cx: &mut Context<Self>) -> Task<Result<(), AuthenticateError>> {
90 let credentials_provider = self.credentials_provider.clone();
91 let api_url = OllamaLanguageModelProvider::api_url(cx);
92 let task = self.api_key_state.load_if_needed(
93 api_url,
94 |this| &mut this.api_key_state,
95 credentials_provider,
96 cx,
97 );
98
99 // Always try to fetch models - if no API key is needed (local Ollama), it will work
100 // If API key is needed and provided, it will work
101 // If API key is needed and not provided, it will fail gracefully
102 cx.spawn(async move |this, cx| {
103 let result = task.await;
104 this.update(cx, |this, cx| this.restart_fetch_models_task(cx))
105 .ok();
106 result
107 })
108 }
109
110 fn fetch_models(&mut self, cx: &mut Context<Self>) -> Task<Result<()>> {
111 let http_client = Arc::clone(&self.http_client);
112 let api_url = OllamaLanguageModelProvider::api_url(cx);
113 let api_key = self.api_key_state.key(&api_url);
114
115 // As a proxy for the server being "authenticated", we'll check if its up by fetching the models
116 cx.spawn(async move |this, cx| {
117 let models = get_models(http_client.as_ref(), &api_url, api_key.as_deref()).await?;
118
119 let tasks = models
120 .into_iter()
121 // Since there is no metadata from the Ollama API
122 // indicating which models are embedding models,
123 // simply filter out models with "-embed" in their name
124 .filter(|model| !model.name.contains("-embed"))
125 .map(|model| {
126 let http_client = Arc::clone(&http_client);
127 let api_url = api_url.clone();
128 let api_key = api_key.clone();
129 async move {
130 let name = model.name.as_str();
131 let model =
132 show_model(http_client.as_ref(), &api_url, api_key.as_deref(), name)
133 .await?;
134 let ollama_model = ollama::Model::new(
135 name,
136 None,
137 model.context_length,
138 Some(model.supports_tools()),
139 Some(model.supports_vision()),
140 Some(model.supports_thinking()),
141 );
142 Ok(ollama_model)
143 }
144 });
145
146 // Rate-limit capability fetches
147 // since there is an arbitrary number of models available
148 let mut ollama_models: Vec<_> = futures::stream::iter(tasks)
149 .buffer_unordered(5)
150 .collect::<Vec<Result<_>>>()
151 .await
152 .into_iter()
153 .collect::<Result<Vec<_>>>()?;
154
155 ollama_models.sort_by(|a, b| a.name.cmp(&b.name));
156
157 this.update(cx, |this, cx| {
158 this.fetched_models = ollama_models;
159 cx.notify();
160 })
161 })
162 }
163
164 fn restart_fetch_models_task(&mut self, cx: &mut Context<Self>) {
165 let task = self.fetch_models(cx);
166 self.fetch_model_task.replace(task);
167 }
168}
169
170impl OllamaLanguageModelProvider {
171 pub fn new(
172 http_client: Arc<dyn HttpClient>,
173 credentials_provider: Arc<dyn CredentialsProvider>,
174 cx: &mut App,
175 ) -> Self {
176 let this = Self {
177 http_client: http_client.clone(),
178 state: cx.new(|cx| {
179 cx.observe_global::<SettingsStore>({
180 let mut last_settings = OllamaLanguageModelProvider::settings(cx).clone();
181 move |this: &mut State, cx| {
182 let current_settings = OllamaLanguageModelProvider::settings(cx);
183 let settings_changed = current_settings != &last_settings;
184 if settings_changed {
185 let url_changed = last_settings.api_url != current_settings.api_url;
186 last_settings = current_settings.clone();
187 if url_changed {
188 let credentials_provider = this.credentials_provider.clone();
189 let api_url = Self::api_url(cx);
190 this.api_key_state.handle_url_change(
191 api_url,
192 |this| &mut this.api_key_state,
193 credentials_provider,
194 cx,
195 );
196 this.fetched_models.clear();
197 this.authenticate(cx).detach();
198 }
199 cx.notify();
200 }
201 }
202 })
203 .detach();
204
205 State {
206 http_client,
207 fetched_models: Default::default(),
208 fetch_model_task: None,
209 api_key_state: ApiKeyState::new(Self::api_url(cx), (*API_KEY_ENV_VAR).clone()),
210 credentials_provider,
211 }
212 }),
213 };
214 this
215 }
216
217 fn settings(cx: &App) -> &OllamaSettings {
218 &AllLanguageModelSettings::get_global(cx).ollama
219 }
220
221 fn api_url(cx: &App) -> SharedString {
222 let api_url = &Self::settings(cx).api_url;
223 if api_url.is_empty() {
224 OLLAMA_API_URL.into()
225 } else {
226 SharedString::new(api_url.as_str())
227 }
228 }
229
230 fn has_custom_url(cx: &App) -> bool {
231 Self::settings(cx).api_url != OLLAMA_API_URL
232 }
233}
234
235impl LanguageModelProviderState for OllamaLanguageModelProvider {
236 type ObservableEntity = State;
237
238 fn observable_entity(&self) -> Option<Entity<Self::ObservableEntity>> {
239 Some(self.state.clone())
240 }
241}
242
243impl LanguageModelProvider for OllamaLanguageModelProvider {
244 fn id(&self) -> LanguageModelProviderId {
245 PROVIDER_ID
246 }
247
248 fn name(&self) -> LanguageModelProviderName {
249 PROVIDER_NAME
250 }
251
252 fn icon(&self) -> IconOrSvg {
253 IconOrSvg::Icon(IconName::AiOllama)
254 }
255
256 fn default_model(&self, _: &App) -> Option<Arc<dyn LanguageModel>> {
257 // We shouldn't try to select default model, because it might lead to a load call for an unloaded model.
258 // In a constrained environment where user might not have enough resources it'll be a bad UX to select something
259 // to load by default.
260 None
261 }
262
263 fn default_fast_model(&self, _: &App) -> Option<Arc<dyn LanguageModel>> {
264 // See explanation for default_model.
265 None
266 }
267
268 fn provided_models(&self, cx: &App) -> Vec<Arc<dyn LanguageModel>> {
269 let mut models: HashMap<String, ollama::Model> = HashMap::new();
270 let settings = OllamaLanguageModelProvider::settings(cx);
271
272 // Add models from the Ollama API
273 for model in self.state.read(cx).fetched_models.iter() {
274 let mut model = model.clone();
275 if let Some(context_window) = settings.context_window {
276 model.max_tokens = context_window;
277 }
278 models.insert(model.name.clone(), model);
279 }
280
281 // Override with available models from settings
282 merge_settings_into_models(
283 &mut models,
284 &settings.available_models,
285 settings.context_window,
286 );
287
288 let mut models = models
289 .into_values()
290 .map(|model| {
291 Arc::new(OllamaLanguageModel {
292 id: LanguageModelId::from(model.name.clone()),
293 model,
294 http_client: self.http_client.clone(),
295 request_limiter: RateLimiter::new(4),
296 state: self.state.clone(),
297 }) as Arc<dyn LanguageModel>
298 })
299 .collect::<Vec<_>>();
300 models.sort_by_key(|model| model.name());
301 models
302 }
303
304 fn is_authenticated(&self, cx: &App) -> bool {
305 self.state.read(cx).is_authenticated()
306 }
307
308 fn authenticate(&self, cx: &mut App) -> Task<Result<(), AuthenticateError>> {
309 self.state.update(cx, |state, cx| state.authenticate(cx))
310 }
311
312 fn configuration_view(
313 &self,
314 _target_agent: language_model::ConfigurationViewTargetAgent,
315 window: &mut Window,
316 cx: &mut App,
317 ) -> AnyView {
318 let state = self.state.clone();
319 cx.new(|cx| ConfigurationView::new(state, window, cx))
320 .into()
321 }
322
323 fn reset_credentials(&self, cx: &mut App) -> Task<Result<()>> {
324 self.state
325 .update(cx, |state, cx| state.set_api_key(None, cx))
326 }
327}
328
329pub struct OllamaLanguageModel {
330 id: LanguageModelId,
331 model: ollama::Model,
332 http_client: Arc<dyn HttpClient>,
333 request_limiter: RateLimiter,
334 state: Entity<State>,
335}
336
337impl OllamaLanguageModel {
338 fn to_ollama_request(&self, request: LanguageModelRequest) -> ChatRequest {
339 let supports_vision = self.model.supports_vision.unwrap_or(false);
340
341 let mut messages = Vec::with_capacity(request.messages.len());
342
343 for mut msg in request.messages.into_iter() {
344 let images = if supports_vision {
345 msg.content
346 .iter()
347 .filter_map(|content| match content {
348 MessageContent::Image(image) => Some(image.source.to_string()),
349 _ => None,
350 })
351 .collect::<Vec<String>>()
352 } else {
353 vec![]
354 };
355
356 match msg.role {
357 Role::User => {
358 for tool_result in msg
359 .content
360 .extract_if(.., |x| matches!(x, MessageContent::ToolResult(..)))
361 {
362 match tool_result {
363 MessageContent::ToolResult(tool_result) => {
364 messages.push(ChatMessage::Tool {
365 tool_name: tool_result.tool_name.to_string(),
366 content: tool_result.content.to_str().unwrap_or("").to_string(),
367 })
368 }
369 _ => unreachable!("Only tool result should be extracted"),
370 }
371 }
372 if !msg.content.is_empty() {
373 messages.push(ChatMessage::User {
374 content: msg.string_contents(),
375 images: if images.is_empty() {
376 None
377 } else {
378 Some(images)
379 },
380 })
381 }
382 }
383 Role::Assistant => {
384 let content = msg.string_contents();
385 let mut thinking = None;
386 let mut tool_calls = Vec::new();
387 for content in msg.content.into_iter() {
388 match content {
389 MessageContent::Thinking { text, .. } if !text.is_empty() => {
390 thinking = Some(text)
391 }
392 MessageContent::ToolUse(tool_use) => {
393 tool_calls.push(OllamaToolCall {
394 id: tool_use.id.to_string(),
395 function: OllamaFunctionCall {
396 name: tool_use.name.to_string(),
397 arguments: tool_use.input,
398 },
399 });
400 }
401 _ => (),
402 }
403 }
404 messages.push(ChatMessage::Assistant {
405 content,
406 tool_calls: Some(tool_calls),
407 images: if images.is_empty() {
408 None
409 } else {
410 Some(images)
411 },
412 thinking,
413 })
414 }
415 Role::System => messages.push(ChatMessage::System {
416 content: msg.string_contents(),
417 }),
418 }
419 }
420 ChatRequest {
421 model: self.model.name.clone(),
422 messages,
423 keep_alive: self.model.keep_alive.clone().unwrap_or_default(),
424 stream: true,
425 options: Some(ChatOptions {
426 num_ctx: Some(self.model.max_tokens),
427 // Only send stop tokens if explicitly provided. When empty/None,
428 // Ollama will use the model's default stop tokens from its Modelfile.
429 // Sending an empty array would override and disable the defaults.
430 stop: if request.stop.is_empty() {
431 None
432 } else {
433 Some(request.stop)
434 },
435 temperature: request.temperature.or(Some(1.0)),
436 ..Default::default()
437 }),
438 think: self
439 .model
440 .supports_thinking
441 .map(|supports_thinking| supports_thinking && request.thinking_allowed),
442 tools: if self.model.supports_tools.unwrap_or(false) {
443 request.tools.into_iter().map(tool_into_ollama).collect()
444 } else {
445 vec![]
446 },
447 }
448 }
449}
450
451impl LanguageModel for OllamaLanguageModel {
452 fn id(&self) -> LanguageModelId {
453 self.id.clone()
454 }
455
456 fn name(&self) -> LanguageModelName {
457 LanguageModelName::from(self.model.display_name().to_string())
458 }
459
460 fn provider_id(&self) -> LanguageModelProviderId {
461 PROVIDER_ID
462 }
463
464 fn provider_name(&self) -> LanguageModelProviderName {
465 PROVIDER_NAME
466 }
467
468 fn supports_tools(&self) -> bool {
469 self.model.supports_tools.unwrap_or(false)
470 }
471
472 fn supports_images(&self) -> bool {
473 self.model.supports_vision.unwrap_or(false)
474 }
475
476 fn supports_thinking(&self) -> bool {
477 self.model.supports_thinking.unwrap_or(false)
478 }
479
480 fn supports_tool_choice(&self, choice: LanguageModelToolChoice) -> bool {
481 match choice {
482 LanguageModelToolChoice::Auto => false,
483 LanguageModelToolChoice::Any => false,
484 LanguageModelToolChoice::None => false,
485 }
486 }
487
488 fn telemetry_id(&self) -> String {
489 format!("ollama/{}", self.model.id())
490 }
491
492 fn max_token_count(&self) -> u64 {
493 self.model.max_token_count()
494 }
495
496 fn stream_completion(
497 &self,
498 request: LanguageModelRequest,
499 cx: &AsyncApp,
500 ) -> BoxFuture<
501 'static,
502 Result<
503 BoxStream<'static, Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>,
504 LanguageModelCompletionError,
505 >,
506 > {
507 let request = self.to_ollama_request(request);
508
509 let http_client = self.http_client.clone();
510 let (api_key, api_url) = self.state.read_with(cx, |state, cx| {
511 let api_url = OllamaLanguageModelProvider::api_url(cx);
512 (state.api_key_state.key(&api_url), api_url)
513 });
514
515 let future = self.request_limiter.stream(async move {
516 let stream =
517 stream_chat_completion(http_client.as_ref(), &api_url, api_key.as_deref(), request)
518 .await?;
519 let stream = map_to_language_model_completion_events(stream);
520 Ok(stream)
521 });
522
523 future.map_ok(|f| f.boxed()).boxed()
524 }
525}
526
527fn map_to_language_model_completion_events(
528 stream: Pin<Box<dyn Stream<Item = anyhow::Result<ChatResponseDelta>> + Send>>,
529) -> impl Stream<Item = Result<LanguageModelCompletionEvent, LanguageModelCompletionError>> {
530 struct State {
531 stream: Pin<Box<dyn Stream<Item = anyhow::Result<ChatResponseDelta>> + Send>>,
532 used_tools: bool,
533 }
534
535 // We need to create a ToolUse and Stop event from a single
536 // response from the original stream
537 let stream = stream::unfold(
538 State {
539 stream,
540 used_tools: false,
541 },
542 async move |mut state| {
543 let response = state.stream.next().await?;
544
545 let delta = match response {
546 Ok(delta) => delta,
547 Err(e) => {
548 let event = Err(LanguageModelCompletionError::from(anyhow!(e)));
549 return Some((vec![event], state));
550 }
551 };
552
553 let mut events = Vec::new();
554
555 match delta.message {
556 ChatMessage::User { content, images: _ } => {
557 events.push(Ok(LanguageModelCompletionEvent::Text(content)));
558 }
559 ChatMessage::System { content } => {
560 events.push(Ok(LanguageModelCompletionEvent::Text(content)));
561 }
562 ChatMessage::Tool { content, .. } => {
563 events.push(Ok(LanguageModelCompletionEvent::Text(content)));
564 }
565 ChatMessage::Assistant {
566 content,
567 tool_calls,
568 images: _,
569 thinking,
570 } => {
571 if let Some(text) = thinking {
572 events.push(Ok(LanguageModelCompletionEvent::Thinking {
573 text,
574 signature: None,
575 }));
576 }
577
578 if let Some(tool_call) = tool_calls.and_then(|v| v.into_iter().next()) {
579 let OllamaToolCall { id, function } = tool_call;
580 let event = LanguageModelCompletionEvent::ToolUse(LanguageModelToolUse {
581 id: LanguageModelToolUseId::from(id),
582 name: Arc::from(function.name),
583 raw_input: function.arguments.to_string(),
584 input: function.arguments,
585 is_input_complete: true,
586 thought_signature: None,
587 });
588 events.push(Ok(event));
589 state.used_tools = true;
590 } else if !content.is_empty() {
591 events.push(Ok(LanguageModelCompletionEvent::Text(content)));
592 }
593 }
594 };
595
596 if delta.done {
597 events.push(Ok(LanguageModelCompletionEvent::UsageUpdate(TokenUsage {
598 input_tokens: delta.prompt_eval_count.unwrap_or(0),
599 output_tokens: delta.eval_count.unwrap_or(0),
600 cache_creation_input_tokens: 0,
601 cache_read_input_tokens: 0,
602 })));
603 if state.used_tools {
604 state.used_tools = false;
605 events.push(Ok(LanguageModelCompletionEvent::Stop(StopReason::ToolUse)));
606 } else {
607 events.push(Ok(LanguageModelCompletionEvent::Stop(StopReason::EndTurn)));
608 }
609 }
610
611 Some((events, state))
612 },
613 );
614
615 stream.flat_map(futures::stream::iter)
616}
617
618struct ConfigurationView {
619 api_key_editor: Entity<InputField>,
620 api_url_editor: Entity<InputField>,
621 context_window_editor: Entity<InputField>,
622 state: Entity<State>,
623}
624
625impl ConfigurationView {
626 pub fn new(state: Entity<State>, window: &mut Window, cx: &mut Context<Self>) -> Self {
627 let api_key_editor = cx.new(|cx| InputField::new(window, cx, "63e02e...").label("API key"));
628
629 let api_url_editor = cx.new(|cx| {
630 let input = InputField::new(window, cx, OLLAMA_API_URL).label("API URL");
631 input.set_text(&OllamaLanguageModelProvider::api_url(cx), window, cx);
632 input
633 });
634
635 let context_window_editor = cx.new(|cx| {
636 let input = InputField::new(window, cx, "8192").label("Context Window");
637 if let Some(context_window) = OllamaLanguageModelProvider::settings(cx).context_window {
638 input.set_text(&context_window.to_string(), window, cx);
639 }
640 input
641 });
642
643 cx.observe(&state, |_, _, cx| {
644 cx.notify();
645 })
646 .detach();
647
648 Self {
649 api_key_editor,
650 api_url_editor,
651 context_window_editor,
652 state,
653 }
654 }
655
656 fn retry_connection(&mut self, window: &mut Window, cx: &mut Context<Self>) {
657 let has_api_url = OllamaLanguageModelProvider::has_custom_url(cx);
658 let has_api_key = self
659 .state
660 .read_with(cx, |state, _| state.api_key_state.has_key());
661 if !has_api_url {
662 self.save_api_url(cx);
663 }
664 if !has_api_key {
665 self.save_api_key(&Default::default(), window, cx);
666 }
667
668 self.state.update(cx, |state, cx| {
669 state.restart_fetch_models_task(cx);
670 });
671 }
672
673 fn save_api_key(&mut self, _: &menu::Confirm, window: &mut Window, cx: &mut Context<Self>) {
674 let api_key = self.api_key_editor.read(cx).text(cx).trim().to_string();
675 if api_key.is_empty() {
676 return;
677 }
678
679 // url changes can cause the editor to be displayed again
680 self.api_key_editor
681 .update(cx, |input, cx| input.set_text("", window, cx));
682
683 let state = self.state.clone();
684 cx.spawn_in(window, async move |_, cx| {
685 state
686 .update(cx, |state, cx| state.set_api_key(Some(api_key), cx))
687 .await
688 })
689 .detach_and_log_err(cx);
690 }
691
692 fn reset_api_key(&mut self, window: &mut Window, cx: &mut Context<Self>) {
693 self.api_key_editor
694 .update(cx, |input, cx| input.set_text("", window, cx));
695
696 let state = self.state.clone();
697 cx.spawn_in(window, async move |_, cx| {
698 state
699 .update(cx, |state, cx| state.set_api_key(None, cx))
700 .await
701 })
702 .detach_and_log_err(cx);
703
704 cx.notify();
705 }
706
707 fn save_api_url(&self, cx: &mut Context<Self>) {
708 let api_url = self.api_url_editor.read(cx).text(cx).trim().to_string();
709 let current_url = OllamaLanguageModelProvider::api_url(cx);
710 if !api_url.is_empty() && &api_url != ¤t_url {
711 let fs = <dyn Fs>::global(cx);
712 update_settings_file(fs, cx, move |settings, _| {
713 settings
714 .language_models
715 .get_or_insert_default()
716 .ollama
717 .get_or_insert_default()
718 .api_url = Some(api_url);
719 });
720 }
721 }
722
723 fn reset_api_url(&mut self, window: &mut Window, cx: &mut Context<Self>) {
724 self.api_url_editor
725 .update(cx, |input, cx| input.set_text("", window, cx));
726 let fs = <dyn Fs>::global(cx);
727 update_settings_file(fs, cx, |settings, _cx| {
728 if let Some(settings) = settings
729 .language_models
730 .as_mut()
731 .and_then(|models| models.ollama.as_mut())
732 {
733 settings.api_url = Some(OLLAMA_API_URL.into());
734 }
735 });
736 cx.notify();
737 }
738
739 fn save_context_window(&mut self, cx: &mut Context<Self>) {
740 let context_window_str = self
741 .context_window_editor
742 .read(cx)
743 .text(cx)
744 .trim()
745 .to_string();
746 let current_context_window = OllamaLanguageModelProvider::settings(cx).context_window;
747
748 if let Ok(context_window) = context_window_str.parse::<u64>() {
749 if Some(context_window) != current_context_window {
750 let fs = <dyn Fs>::global(cx);
751 update_settings_file(fs, cx, move |settings, _| {
752 settings
753 .language_models
754 .get_or_insert_default()
755 .ollama
756 .get_or_insert_default()
757 .context_window = Some(context_window);
758 });
759 }
760 } else if context_window_str.is_empty() && current_context_window.is_some() {
761 let fs = <dyn Fs>::global(cx);
762 update_settings_file(fs, cx, move |settings, _| {
763 settings
764 .language_models
765 .get_or_insert_default()
766 .ollama
767 .get_or_insert_default()
768 .context_window = None;
769 });
770 }
771 }
772
773 fn reset_context_window(&mut self, window: &mut Window, cx: &mut Context<Self>) {
774 self.context_window_editor
775 .update(cx, |input, cx| input.set_text("", window, cx));
776 let fs = <dyn Fs>::global(cx);
777 update_settings_file(fs, cx, |settings, _cx| {
778 if let Some(settings) = settings
779 .language_models
780 .as_mut()
781 .and_then(|models| models.ollama.as_mut())
782 {
783 settings.context_window = None;
784 }
785 });
786 cx.notify();
787 }
788
789 fn render_instructions(cx: &App) -> Div {
790 v_flex()
791 .gap_2()
792 .child(Label::new(
793 "Run LLMs locally on your machine with Ollama, or connect to an Ollama server. \
794 Can provide access to Llama, Mistral, Gemma, and hundreds of other models.",
795 ))
796 .child(Label::new("To use local Ollama:"))
797 .child(
798 List::new()
799 .child(
800 ListBulletItem::new("")
801 .child(Label::new("Download and install Ollama from"))
802 .child(ButtonLink::new("ollama.com", "https://ollama.com/download")),
803 )
804 .child(
805 ListBulletItem::new("")
806 .child(Label::new("Start Ollama and download a model:"))
807 .child(Label::new("ollama run gpt-oss:20b").inline_code(cx)),
808 )
809 .child(ListBulletItem::new(
810 "Click 'Connect' below to start using Ollama in Zed",
811 )),
812 )
813 .child(Label::new(
814 "Alternatively, you can connect to an Ollama server by specifying its \
815 URL and API key (may not be required):",
816 ))
817 }
818
819 fn render_api_key_editor(&self, cx: &Context<Self>) -> impl IntoElement {
820 let state = self.state.read(cx);
821 let env_var_set = state.api_key_state.is_from_env_var();
822 let configured_card_label = if env_var_set {
823 format!("API key set in {API_KEY_ENV_VAR_NAME} environment variable.")
824 } else {
825 "API key configured".to_string()
826 };
827
828 if !state.api_key_state.has_key() {
829 v_flex()
830 .on_action(cx.listener(Self::save_api_key))
831 .child(self.api_key_editor.clone())
832 .child(
833 Label::new(
834 format!("You can also set the {API_KEY_ENV_VAR_NAME} environment variable and restart Zed.")
835 )
836 .size(LabelSize::Small)
837 .color(Color::Muted),
838 )
839 .into_any_element()
840 } else {
841 ConfiguredApiCard::new(configured_card_label)
842 .disabled(env_var_set)
843 .on_click(cx.listener(|this, _, window, cx| this.reset_api_key(window, cx)))
844 .when(env_var_set, |this| {
845 this.tooltip_label(format!("To reset your API key, unset the {API_KEY_ENV_VAR_NAME} environment variable."))
846 })
847 .into_any_element()
848 }
849 }
850
851 fn render_context_window_editor(&self, cx: &Context<Self>) -> Div {
852 let settings = OllamaLanguageModelProvider::settings(cx);
853 let custom_context_window_set = settings.context_window.is_some();
854
855 if custom_context_window_set {
856 h_flex()
857 .p_3()
858 .justify_between()
859 .rounded_md()
860 .border_1()
861 .border_color(cx.theme().colors().border)
862 .bg(cx.theme().colors().elevated_surface_background)
863 .child(
864 h_flex()
865 .gap_2()
866 .child(Icon::new(IconName::Check).color(Color::Success))
867 .child(v_flex().gap_1().child(Label::new(format!(
868 "Context Window: {}",
869 settings.context_window.unwrap()
870 )))),
871 )
872 .child(
873 Button::new("reset-context-window", "Reset")
874 .label_size(LabelSize::Small)
875 .start_icon(Icon::new(IconName::Undo).size(IconSize::Small))
876 .layer(ElevationIndex::ModalSurface)
877 .on_click(
878 cx.listener(|this, _, window, cx| {
879 this.reset_context_window(window, cx)
880 }),
881 ),
882 )
883 } else {
884 v_flex()
885 .on_action(
886 cx.listener(|this, _: &menu::Confirm, _window, cx| {
887 this.save_context_window(cx)
888 }),
889 )
890 .child(self.context_window_editor.clone())
891 .child(
892 Label::new("Default: Model specific")
893 .size(LabelSize::Small)
894 .color(Color::Muted),
895 )
896 }
897 }
898
899 fn render_api_url_editor(&self, cx: &Context<Self>) -> Div {
900 let api_url = OllamaLanguageModelProvider::api_url(cx);
901 let custom_api_url_set = api_url != OLLAMA_API_URL;
902
903 if custom_api_url_set {
904 h_flex()
905 .p_3()
906 .justify_between()
907 .rounded_md()
908 .border_1()
909 .border_color(cx.theme().colors().border)
910 .bg(cx.theme().colors().elevated_surface_background)
911 .child(
912 h_flex()
913 .gap_2()
914 .child(Icon::new(IconName::Check).color(Color::Success))
915 .child(v_flex().gap_1().child(Label::new(api_url))),
916 )
917 .child(
918 Button::new("reset-api-url", "Reset API URL")
919 .label_size(LabelSize::Small)
920 .start_icon(Icon::new(IconName::Undo).size(IconSize::Small))
921 .layer(ElevationIndex::ModalSurface)
922 .on_click(
923 cx.listener(|this, _, window, cx| this.reset_api_url(window, cx)),
924 ),
925 )
926 } else {
927 v_flex()
928 .on_action(cx.listener(|this, _: &menu::Confirm, _window, cx| {
929 this.save_api_url(cx);
930 cx.notify();
931 }))
932 .gap_2()
933 .child(self.api_url_editor.clone())
934 }
935 }
936}
937
938impl Render for ConfigurationView {
939 fn render(&mut self, _: &mut Window, cx: &mut Context<Self>) -> impl IntoElement {
940 let is_authenticated = self.state.read(cx).is_authenticated();
941
942 v_flex()
943 .gap_2()
944 .child(Self::render_instructions(cx))
945 .child(self.render_api_url_editor(cx))
946 .child(self.render_context_window_editor(cx))
947 .child(self.render_api_key_editor(cx))
948 .child(
949 h_flex()
950 .w_full()
951 .justify_between()
952 .gap_2()
953 .child(
954 h_flex()
955 .w_full()
956 .gap_2()
957 .map(|this| {
958 if is_authenticated {
959 this.child(
960 Button::new("ollama-site", "Ollama")
961 .style(ButtonStyle::Subtle)
962 .end_icon(
963 Icon::new(IconName::ArrowUpRight)
964 .size(IconSize::XSmall)
965 .color(Color::Muted),
966 )
967 .on_click(move |_, _, cx| cx.open_url(OLLAMA_SITE))
968 .into_any_element(),
969 )
970 } else {
971 this.child(
972 Button::new("download_ollama_button", "Download Ollama")
973 .style(ButtonStyle::Subtle)
974 .end_icon(
975 Icon::new(IconName::ArrowUpRight)
976 .size(IconSize::XSmall)
977 .color(Color::Muted),
978 )
979 .on_click(move |_, _, cx| {
980 cx.open_url(OLLAMA_DOWNLOAD_URL)
981 })
982 .into_any_element(),
983 )
984 }
985 })
986 .child(
987 Button::new("view-models", "View All Models")
988 .style(ButtonStyle::Subtle)
989 .end_icon(
990 Icon::new(IconName::ArrowUpRight)
991 .size(IconSize::XSmall)
992 .color(Color::Muted),
993 )
994 .on_click(move |_, _, cx| cx.open_url(OLLAMA_LIBRARY_URL)),
995 ),
996 )
997 .map(|this| {
998 if is_authenticated {
999 this.child(
1000 ButtonLike::new("connected")
1001 .disabled(true)
1002 .cursor_style(CursorStyle::Arrow)
1003 .child(
1004 h_flex()
1005 .gap_2()
1006 .child(Icon::new(IconName::Check).color(Color::Success))
1007 .child(Label::new("Connected"))
1008 .into_any_element(),
1009 )
1010 .child(
1011 IconButton::new("refresh-models", IconName::RotateCcw)
1012 .tooltip(Tooltip::text("Refresh Models"))
1013 .on_click(cx.listener(|this, _, window, cx| {
1014 this.state.update(cx, |state, _| {
1015 state.fetched_models.clear();
1016 });
1017 this.retry_connection(window, cx);
1018 })),
1019 ),
1020 )
1021 } else {
1022 this.child(
1023 Button::new("retry_ollama_models", "Connect")
1024 .start_icon(
1025 Icon::new(IconName::PlayOutlined).size(IconSize::XSmall),
1026 )
1027 .on_click(cx.listener(move |this, _, window, cx| {
1028 this.retry_connection(window, cx)
1029 })),
1030 )
1031 }
1032 }),
1033 )
1034 }
1035}
1036
1037fn merge_settings_into_models(
1038 models: &mut HashMap<String, ollama::Model>,
1039 available_models: &[AvailableModel],
1040 context_window: Option<u64>,
1041) {
1042 for setting_model in available_models {
1043 if let Some(model) = models.get_mut(&setting_model.name) {
1044 if context_window.is_none() {
1045 model.max_tokens = setting_model.max_tokens;
1046 }
1047 model.display_name = setting_model.display_name.clone();
1048 model.keep_alive = setting_model.keep_alive.clone();
1049 model.supports_tools = setting_model.supports_tools;
1050 model.supports_vision = setting_model.supports_images;
1051 model.supports_thinking = setting_model.supports_thinking;
1052 } else {
1053 models.insert(
1054 setting_model.name.clone(),
1055 ollama::Model {
1056 name: setting_model.name.clone(),
1057 display_name: setting_model.display_name.clone(),
1058 max_tokens: context_window.unwrap_or(setting_model.max_tokens),
1059 keep_alive: setting_model.keep_alive.clone(),
1060 supports_tools: setting_model.supports_tools,
1061 supports_vision: setting_model.supports_images,
1062 supports_thinking: setting_model.supports_thinking,
1063 },
1064 );
1065 }
1066 }
1067}
1068
1069fn tool_into_ollama(tool: LanguageModelRequestTool) -> ollama::OllamaTool {
1070 ollama::OllamaTool::Function {
1071 function: OllamaFunctionTool {
1072 name: tool.name,
1073 description: Some(tool.description),
1074 parameters: Some(tool.input_schema),
1075 },
1076 }
1077}
1078
1079#[cfg(test)]
1080mod tests {
1081 use super::*;
1082
1083 #[test]
1084 fn test_merge_settings_preserves_display_names_for_similar_models() {
1085 // Regression test for https://github.com/zed-industries/zed/issues/43646
1086 // When multiple models share the same base name (e.g., qwen2.5-coder:1.5b and qwen2.5-coder:3b),
1087 // each model should get its own display_name from settings, not a random one.
1088
1089 let mut models: HashMap<String, ollama::Model> = HashMap::new();
1090 models.insert(
1091 "qwen2.5-coder:1.5b".to_string(),
1092 ollama::Model {
1093 name: "qwen2.5-coder:1.5b".to_string(),
1094 display_name: None,
1095 max_tokens: 4096,
1096 keep_alive: None,
1097 supports_tools: None,
1098 supports_vision: None,
1099 supports_thinking: None,
1100 },
1101 );
1102 models.insert(
1103 "qwen2.5-coder:3b".to_string(),
1104 ollama::Model {
1105 name: "qwen2.5-coder:3b".to_string(),
1106 display_name: None,
1107 max_tokens: 4096,
1108 keep_alive: None,
1109 supports_tools: None,
1110 supports_vision: None,
1111 supports_thinking: None,
1112 },
1113 );
1114
1115 let available_models = vec![
1116 AvailableModel {
1117 name: "qwen2.5-coder:1.5b".to_string(),
1118 display_name: Some("QWEN2.5 Coder 1.5B".to_string()),
1119 max_tokens: 5000,
1120 keep_alive: None,
1121 supports_tools: Some(true),
1122 supports_images: None,
1123 supports_thinking: None,
1124 },
1125 AvailableModel {
1126 name: "qwen2.5-coder:3b".to_string(),
1127 display_name: Some("QWEN2.5 Coder 3B".to_string()),
1128 max_tokens: 6000,
1129 keep_alive: None,
1130 supports_tools: Some(true),
1131 supports_images: None,
1132 supports_thinking: None,
1133 },
1134 ];
1135
1136 merge_settings_into_models(&mut models, &available_models, None);
1137
1138 let model_1_5b = models
1139 .get("qwen2.5-coder:1.5b")
1140 .expect("1.5b model missing");
1141 let model_3b = models.get("qwen2.5-coder:3b").expect("3b model missing");
1142
1143 assert_eq!(
1144 model_1_5b.display_name,
1145 Some("QWEN2.5 Coder 1.5B".to_string()),
1146 "1.5b model should have its own display_name"
1147 );
1148 assert_eq!(model_1_5b.max_tokens, 5000);
1149
1150 assert_eq!(
1151 model_3b.display_name,
1152 Some("QWEN2.5 Coder 3B".to_string()),
1153 "3b model should have its own display_name"
1154 );
1155 assert_eq!(model_3b.max_tokens, 6000);
1156 }
1157}