1use std::fmt;
2use std::sync::Arc;
3
4use anyhow::Result;
5use client::Client;
6use gpui::{
7 App, AppContext as _, AsyncApp, Context, Entity, EventEmitter, Global, ReadGlobal as _,
8};
9use proto::{Plan, TypedEnvelope};
10use schemars::JsonSchema;
11use serde::{Deserialize, Serialize};
12use smol::lock::{RwLock, RwLockUpgradableReadGuard, RwLockWriteGuard};
13use strum::EnumIter;
14use thiserror::Error;
15
16use crate::{LanguageModelAvailability, LanguageModelToolSchemaFormat};
17
18#[derive(Clone, Debug, PartialEq, Serialize, Deserialize, JsonSchema)]
19#[serde(tag = "provider", rename_all = "lowercase")]
20pub enum CloudModel {
21 Anthropic(anthropic::Model),
22 OpenAi(open_ai::Model),
23 Google(google_ai::Model),
24}
25
26#[derive(Clone, Debug, PartialEq, Serialize, Deserialize, JsonSchema, EnumIter)]
27pub enum ZedModel {
28 #[serde(rename = "Qwen/Qwen2-7B-Instruct")]
29 Qwen2_7bInstruct,
30}
31
32impl Default for CloudModel {
33 fn default() -> Self {
34 Self::Anthropic(anthropic::Model::default())
35 }
36}
37
38impl CloudModel {
39 pub fn id(&self) -> &str {
40 match self {
41 Self::Anthropic(model) => model.id(),
42 Self::OpenAi(model) => model.id(),
43 Self::Google(model) => model.id(),
44 }
45 }
46
47 pub fn display_name(&self) -> &str {
48 match self {
49 Self::Anthropic(model) => model.display_name(),
50 Self::OpenAi(model) => model.display_name(),
51 Self::Google(model) => model.display_name(),
52 }
53 }
54
55 pub fn max_token_count(&self) -> usize {
56 match self {
57 Self::Anthropic(model) => model.max_token_count(),
58 Self::OpenAi(model) => model.max_token_count(),
59 Self::Google(model) => model.max_token_count(),
60 }
61 }
62
63 /// Returns the availability of this model.
64 pub fn availability(&self) -> LanguageModelAvailability {
65 match self {
66 Self::Anthropic(model) => match model {
67 anthropic::Model::Claude3_5Sonnet
68 | anthropic::Model::Claude3_7Sonnet
69 | anthropic::Model::Claude3_7SonnetThinking => {
70 LanguageModelAvailability::RequiresPlan(Plan::Free)
71 }
72 anthropic::Model::Claude3Opus
73 | anthropic::Model::Claude3Sonnet
74 | anthropic::Model::Claude3Haiku
75 | anthropic::Model::Claude3_5Haiku
76 | anthropic::Model::Custom { .. } => {
77 LanguageModelAvailability::RequiresPlan(Plan::ZedPro)
78 }
79 },
80 Self::OpenAi(model) => match model {
81 open_ai::Model::ThreePointFiveTurbo
82 | open_ai::Model::Four
83 | open_ai::Model::FourTurbo
84 | open_ai::Model::FourOmni
85 | open_ai::Model::FourOmniMini
86 | open_ai::Model::O1Mini
87 | open_ai::Model::O1Preview
88 | open_ai::Model::O1
89 | open_ai::Model::O3Mini
90 | open_ai::Model::Custom { .. } => {
91 LanguageModelAvailability::RequiresPlan(Plan::ZedPro)
92 }
93 },
94 Self::Google(model) => match model {
95 google_ai::Model::Gemini15Pro
96 | google_ai::Model::Gemini15Flash
97 | google_ai::Model::Gemini20Pro
98 | google_ai::Model::Gemini20Flash
99 | google_ai::Model::Gemini20FlashThinking
100 | google_ai::Model::Gemini20FlashLite
101 | google_ai::Model::Gemini25ProExp0325
102 | google_ai::Model::Gemini25ProPreview0325
103 | google_ai::Model::Custom { .. } => {
104 LanguageModelAvailability::RequiresPlan(Plan::ZedPro)
105 }
106 },
107 }
108 }
109
110 pub fn tool_input_format(&self) -> LanguageModelToolSchemaFormat {
111 match self {
112 Self::Anthropic(_) | Self::OpenAi(_) => LanguageModelToolSchemaFormat::JsonSchema,
113 Self::Google(_) => LanguageModelToolSchemaFormat::JsonSchemaSubset,
114 }
115 }
116}
117
118#[derive(Error, Debug)]
119pub struct PaymentRequiredError;
120
121impl fmt::Display for PaymentRequiredError {
122 fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
123 write!(
124 f,
125 "Payment required to use this language model. Please upgrade your account."
126 )
127 }
128}
129
130#[derive(Error, Debug)]
131pub struct MaxMonthlySpendReachedError;
132
133impl fmt::Display for MaxMonthlySpendReachedError {
134 fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
135 write!(
136 f,
137 "Maximum spending limit reached for this month. For more usage, increase your spending limit."
138 )
139 }
140}
141
142#[derive(Clone, Default)]
143pub struct LlmApiToken(Arc<RwLock<Option<String>>>);
144
145impl LlmApiToken {
146 pub async fn acquire(&self, client: &Arc<Client>) -> Result<String> {
147 let lock = self.0.upgradable_read().await;
148 if let Some(token) = lock.as_ref() {
149 Ok(token.to_string())
150 } else {
151 Self::fetch(RwLockUpgradableReadGuard::upgrade(lock).await, client).await
152 }
153 }
154
155 pub async fn refresh(&self, client: &Arc<Client>) -> Result<String> {
156 Self::fetch(self.0.write().await, client).await
157 }
158
159 async fn fetch(
160 mut lock: RwLockWriteGuard<'_, Option<String>>,
161 client: &Arc<Client>,
162 ) -> Result<String> {
163 let response = client.request(proto::GetLlmToken {}).await?;
164 *lock = Some(response.token.clone());
165 Ok(response.token.clone())
166 }
167}
168
169struct GlobalRefreshLlmTokenListener(Entity<RefreshLlmTokenListener>);
170
171impl Global for GlobalRefreshLlmTokenListener {}
172
173pub struct RefreshLlmTokenEvent;
174
175pub struct RefreshLlmTokenListener {
176 _llm_token_subscription: client::Subscription,
177}
178
179impl EventEmitter<RefreshLlmTokenEvent> for RefreshLlmTokenListener {}
180
181impl RefreshLlmTokenListener {
182 pub fn register(client: Arc<Client>, cx: &mut App) {
183 let listener = cx.new(|cx| RefreshLlmTokenListener::new(client, cx));
184 cx.set_global(GlobalRefreshLlmTokenListener(listener));
185 }
186
187 pub fn global(cx: &App) -> Entity<Self> {
188 GlobalRefreshLlmTokenListener::global(cx).0.clone()
189 }
190
191 fn new(client: Arc<Client>, cx: &mut Context<Self>) -> Self {
192 Self {
193 _llm_token_subscription: client
194 .add_message_handler(cx.weak_entity(), Self::handle_refresh_llm_token),
195 }
196 }
197
198 async fn handle_refresh_llm_token(
199 this: Entity<Self>,
200 _: TypedEnvelope<proto::RefreshLlmToken>,
201 mut cx: AsyncApp,
202 ) -> Result<()> {
203 this.update(&mut cx, |_this, cx| cx.emit(RefreshLlmTokenEvent))
204 }
205}