cloud_model.rs

  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::Custom { .. } => {
 94                    LanguageModelAvailability::RequiresPlan(Plan::ZedPro)
 95                }
 96            },
 97            Self::Google(model) => match model {
 98                google_ai::Model::Gemini15Pro
 99                | google_ai::Model::Gemini15Flash
100                | google_ai::Model::Gemini20Pro
101                | google_ai::Model::Gemini20Flash
102                | google_ai::Model::Gemini20FlashThinking
103                | google_ai::Model::Gemini20FlashLite
104                | google_ai::Model::Gemini25ProExp0325
105                | google_ai::Model::Gemini25ProPreview0325
106                | google_ai::Model::Custom { .. } => {
107                    LanguageModelAvailability::RequiresPlan(Plan::ZedPro)
108                }
109            },
110        }
111    }
112
113    pub fn tool_input_format(&self) -> LanguageModelToolSchemaFormat {
114        match self {
115            Self::Anthropic(_) | Self::OpenAi(_) => LanguageModelToolSchemaFormat::JsonSchema,
116            Self::Google(_) => LanguageModelToolSchemaFormat::JsonSchemaSubset,
117        }
118    }
119}
120
121#[derive(Error, Debug)]
122pub struct PaymentRequiredError;
123
124impl fmt::Display for PaymentRequiredError {
125    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
126        write!(
127            f,
128            "Payment required to use this language model. Please upgrade your account."
129        )
130    }
131}
132
133#[derive(Error, Debug)]
134pub struct MaxMonthlySpendReachedError;
135
136impl fmt::Display for MaxMonthlySpendReachedError {
137    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
138        write!(
139            f,
140            "Maximum spending limit reached for this month. For more usage, increase your spending limit."
141        )
142    }
143}
144
145#[derive(Error, Debug)]
146pub struct ModelRequestLimitReachedError {
147    pub plan: Plan,
148}
149
150impl fmt::Display for ModelRequestLimitReachedError {
151    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
152        let message = match self.plan {
153            Plan::Free => "Model request limit reached. Upgrade to Zed Pro for more requests.",
154            Plan::ZedPro => {
155                "Model request limit reached. Upgrade to usage-based billing for more requests."
156            }
157        };
158
159        write!(f, "{message}")
160    }
161}
162
163#[derive(Clone, Default)]
164pub struct LlmApiToken(Arc<RwLock<Option<String>>>);
165
166impl LlmApiToken {
167    pub async fn acquire(&self, client: &Arc<Client>) -> Result<String> {
168        let lock = self.0.upgradable_read().await;
169        if let Some(token) = lock.as_ref() {
170            Ok(token.to_string())
171        } else {
172            Self::fetch(RwLockUpgradableReadGuard::upgrade(lock).await, client).await
173        }
174    }
175
176    pub async fn refresh(&self, client: &Arc<Client>) -> Result<String> {
177        Self::fetch(self.0.write().await, client).await
178    }
179
180    async fn fetch(
181        mut lock: RwLockWriteGuard<'_, Option<String>>,
182        client: &Arc<Client>,
183    ) -> Result<String> {
184        let response = client.request(proto::GetLlmToken {}).await?;
185        *lock = Some(response.token.clone());
186        Ok(response.token.clone())
187    }
188}
189
190struct GlobalRefreshLlmTokenListener(Entity<RefreshLlmTokenListener>);
191
192impl Global for GlobalRefreshLlmTokenListener {}
193
194pub struct RefreshLlmTokenEvent;
195
196pub struct RefreshLlmTokenListener {
197    _llm_token_subscription: client::Subscription,
198}
199
200impl EventEmitter<RefreshLlmTokenEvent> for RefreshLlmTokenListener {}
201
202impl RefreshLlmTokenListener {
203    pub fn register(client: Arc<Client>, cx: &mut App) {
204        let listener = cx.new(|cx| RefreshLlmTokenListener::new(client, cx));
205        cx.set_global(GlobalRefreshLlmTokenListener(listener));
206    }
207
208    pub fn global(cx: &App) -> Entity<Self> {
209        GlobalRefreshLlmTokenListener::global(cx).0.clone()
210    }
211
212    fn new(client: Arc<Client>, cx: &mut Context<Self>) -> Self {
213        Self {
214            _llm_token_subscription: client
215                .add_message_handler(cx.weak_entity(), Self::handle_refresh_llm_token),
216        }
217    }
218
219    async fn handle_refresh_llm_token(
220        this: Entity<Self>,
221        _: TypedEnvelope<proto::RefreshLlmToken>,
222        mut cx: AsyncApp,
223    ) -> Result<()> {
224        this.update(&mut cx, |_this, cx| cx.emit(RefreshLlmTokenEvent))
225    }
226}