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::FourPointOne
87 | open_ai::Model::FourPointOneMini
88 | open_ai::Model::FourPointOneNano
89 | open_ai::Model::O1Mini
90 | open_ai::Model::O1Preview
91 | open_ai::Model::O1
92 | open_ai::Model::O3Mini
93 | open_ai::Model::O3
94 | open_ai::Model::O4Mini
95 | open_ai::Model::Custom { .. } => {
96 LanguageModelAvailability::RequiresPlan(Plan::ZedPro)
97 }
98 },
99 Self::Google(model) => match model {
100 google_ai::Model::Gemini15Pro
101 | google_ai::Model::Gemini15Flash
102 | google_ai::Model::Gemini20Pro
103 | google_ai::Model::Gemini20Flash
104 | google_ai::Model::Gemini20FlashThinking
105 | google_ai::Model::Gemini20FlashLite
106 | google_ai::Model::Gemini25ProExp0325
107 | google_ai::Model::Gemini25ProPreview0325
108 | google_ai::Model::Custom { .. } => {
109 LanguageModelAvailability::RequiresPlan(Plan::ZedPro)
110 }
111 },
112 }
113 }
114
115 pub fn tool_input_format(&self) -> LanguageModelToolSchemaFormat {
116 match self {
117 Self::Anthropic(_) | Self::OpenAi(_) => LanguageModelToolSchemaFormat::JsonSchema,
118 Self::Google(_) => LanguageModelToolSchemaFormat::JsonSchemaSubset,
119 }
120 }
121}
122
123#[derive(Error, Debug)]
124pub struct PaymentRequiredError;
125
126impl fmt::Display for PaymentRequiredError {
127 fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
128 write!(
129 f,
130 "Payment required to use this language model. Please upgrade your account."
131 )
132 }
133}
134
135#[derive(Error, Debug)]
136pub struct MaxMonthlySpendReachedError;
137
138impl fmt::Display for MaxMonthlySpendReachedError {
139 fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
140 write!(
141 f,
142 "Maximum spending limit reached for this month. For more usage, increase your spending limit."
143 )
144 }
145}
146
147#[derive(Clone, Default)]
148pub struct LlmApiToken(Arc<RwLock<Option<String>>>);
149
150impl LlmApiToken {
151 pub async fn acquire(&self, client: &Arc<Client>) -> Result<String> {
152 let lock = self.0.upgradable_read().await;
153 if let Some(token) = lock.as_ref() {
154 Ok(token.to_string())
155 } else {
156 Self::fetch(RwLockUpgradableReadGuard::upgrade(lock).await, client).await
157 }
158 }
159
160 pub async fn refresh(&self, client: &Arc<Client>) -> Result<String> {
161 Self::fetch(self.0.write().await, client).await
162 }
163
164 async fn fetch(
165 mut lock: RwLockWriteGuard<'_, Option<String>>,
166 client: &Arc<Client>,
167 ) -> Result<String> {
168 let response = client.request(proto::GetLlmToken {}).await?;
169 *lock = Some(response.token.clone());
170 Ok(response.token.clone())
171 }
172}
173
174struct GlobalRefreshLlmTokenListener(Entity<RefreshLlmTokenListener>);
175
176impl Global for GlobalRefreshLlmTokenListener {}
177
178pub struct RefreshLlmTokenEvent;
179
180pub struct RefreshLlmTokenListener {
181 _llm_token_subscription: client::Subscription,
182}
183
184impl EventEmitter<RefreshLlmTokenEvent> for RefreshLlmTokenListener {}
185
186impl RefreshLlmTokenListener {
187 pub fn register(client: Arc<Client>, cx: &mut App) {
188 let listener = cx.new(|cx| RefreshLlmTokenListener::new(client, cx));
189 cx.set_global(GlobalRefreshLlmTokenListener(listener));
190 }
191
192 pub fn global(cx: &App) -> Entity<Self> {
193 GlobalRefreshLlmTokenListener::global(cx).0.clone()
194 }
195
196 fn new(client: Arc<Client>, cx: &mut Context<Self>) -> Self {
197 Self {
198 _llm_token_subscription: client
199 .add_message_handler(cx.weak_entity(), Self::handle_refresh_llm_token),
200 }
201 }
202
203 async fn handle_refresh_llm_token(
204 this: Entity<Self>,
205 _: TypedEnvelope<proto::RefreshLlmToken>,
206 mut cx: AsyncApp,
207 ) -> Result<()> {
208 this.update(&mut cx, |_this, cx| cx.emit(RefreshLlmTokenEvent))
209 }
210}