use ai_onboarding::YoungAccountBanner;use anthropic::AnthropicModelMode;use anyhow::{Context as _, Result, anyhow};use chrono::{DateTime, Utc};use client::{Client, ModelRequestUsage, UserStore, zed_urls};use cloud_llm_client::{    CLIENT_SUPPORTS_STATUS_MESSAGES_HEADER_NAME, CURRENT_PLAN_HEADER_NAME, CompletionBody,    CompletionEvent, CompletionRequestStatus, CountTokensBody, CountTokensResponse,    EXPIRED_LLM_TOKEN_HEADER_NAME, ListModelsResponse, MODEL_REQUESTS_RESOURCE_HEADER_VALUE, Plan,    SERVER_SUPPORTS_STATUS_MESSAGES_HEADER_NAME, SUBSCRIPTION_LIMIT_RESOURCE_HEADER_NAME,    TOOL_USE_LIMIT_REACHED_HEADER_NAME, ZED_VERSION_HEADER_NAME,};use futures::{    AsyncBufReadExt, FutureExt, Stream, StreamExt, future::BoxFuture, stream::BoxStream,};use google_ai::GoogleModelMode;use gpui::{    AnyElement, AnyView, App, AsyncApp, Context, Entity, SemanticVersion, Subscription, Task,};use http_client::http::{HeaderMap, HeaderValue};use http_client::{AsyncBody, HttpClient, Method, Response, StatusCode};use language_model::{    AuthenticateError, LanguageModel, LanguageModelCacheConfiguration,    LanguageModelCompletionError, LanguageModelCompletionEvent, LanguageModelId, LanguageModelName,    LanguageModelProvider, LanguageModelProviderId, LanguageModelProviderName,    LanguageModelProviderState, LanguageModelProviderTosView, LanguageModelRequest,    LanguageModelToolChoice, LanguageModelToolSchemaFormat, LlmApiToken,    ModelRequestLimitReachedError, PaymentRequiredError, RateLimiter, RefreshLlmTokenListener,};use release_channel::AppVersion;use schemars::JsonSchema;use serde::{Deserialize, Serialize, de::DeserializeOwned};use settings::SettingsStore;use smol::io::{AsyncReadExt, BufReader};use std::pin::Pin;use std::str::FromStr as _;use std::sync::Arc;use std::time::Duration;use thiserror::Error;use ui::{TintColor, prelude::*};use util::{ResultExt as _, maybe}; use crate::provider::anthropic::{AnthropicEventMapper, count_anthropic_tokens, into_anthropic};use crate::provider::google::{GoogleEventMapper, into_google};use crate::provider::open_ai::{OpenAiEventMapper, count_open_ai_tokens, into_open_ai}; const PROVIDER_ID: LanguageModelProviderId = language_model::ZED_CLOUD_PROVIDER_ID;const PROVIDER_NAME: LanguageModelProviderName = language_model::ZED_CLOUD_PROVIDER_NAME; #[derive(Default, Clone, Debug, PartialEq)]pub struct ZedDotDevSettings {    pub available_models: Vec<AvailableModel>,} #[derive(Clone, Debug, PartialEq, Serialize, Deserialize, JsonSchema)]#[serde(rename_all = "lowercase")]pub enum AvailableProvider {    Anthropic,    OpenAi,    Google,} #[derive(Clone, Debug, PartialEq, Serialize, Deserialize, JsonSchema)]pub struct AvailableModel {    /// The provider of the language model.    pub provider: AvailableProvider,    /// The model's name in the provider's API. e.g. claude-3-5-sonnet-20240620    pub name: String,    /// The name displayed in the UI, such as in the assistant panel model dropdown menu.    pub display_name: Option<String>,    /// The size of the context window, indicating the maximum number of tokens the model can process.    pub max_tokens: usize,    /// The maximum number of output tokens allowed by the model.    pub max_output_tokens: Option<u64>,    /// The maximum number of completion tokens allowed by the model (o1-* only)    pub max_completion_tokens: Option<u64>,    /// Override this model with a different Anthropic model for tool calls.    pub tool_override: Option<String>,    /// Indicates whether this custom model supports caching.    pub cache_configuration: Option<LanguageModelCacheConfiguration>,    /// The default temperature to use for this model.    pub default_temperature: Option<f32>,    /// Any extra beta headers to provide when using the model.    #[serde(default)]    pub extra_beta_headers: Vec<String>,    /// The model's mode (e.g. thinking)    pub mode: Option<ModelMode>,} #[derive(Default, Clone, Debug, PartialEq, Serialize, Deserialize, JsonSchema)]#[serde(tag = "type", rename_all = "lowercase")]pub enum ModelMode {    #[default]    Default,    Thinking {        /// The maximum number of tokens to use for reasoning. Must be lower than the model's `max_output_tokens`.        budget_tokens: Option<u32>,    },} impl From<ModelMode> for AnthropicModelMode {    fn from(value: ModelMode) -> Self {        match value {            ModelMode::Default => AnthropicModelMode::Default,            ModelMode::Thinking { budget_tokens } => AnthropicModelMode::Thinking { budget_tokens },        }    }} pub struct CloudLanguageModelProvider {    client: Arc<Client>,    state: gpui::Entity<State>,    _maintain_client_status: Task<()>,} pub struct State {    client: Arc<Client>,    llm_api_token: LlmApiToken,    user_store: Entity<UserStore>,    status: client::Status,    accept_terms_of_service_task: Option<Task<Result<()>>>,    models: Vec<Arc<cloud_llm_client::LanguageModel>>,    default_model: Option<Arc<cloud_llm_client::LanguageModel>>,    default_fast_model: Option<Arc<cloud_llm_client::LanguageModel>>,    recommended_models: Vec<Arc<cloud_llm_client::LanguageModel>>,    _fetch_models_task: Task<()>,    _settings_subscription: Subscription,    _llm_token_subscription: Subscription,} impl State {    fn new(        client: Arc<Client>,        user_store: Entity<UserStore>,        status: client::Status,        cx: &mut Context<Self>,    ) -> Self {        let refresh_llm_token_listener = RefreshLlmTokenListener::global(cx);        Self {            client: client.clone(),            llm_api_token: LlmApiToken::default(),            user_store: user_store.clone(),            status,            accept_terms_of_service_task: None,            models: Vec::new(),            default_model: None,            default_fast_model: None,            recommended_models: Vec::new(),            _fetch_models_task: cx.spawn(async move |this, cx| {                maybe!(async move {                    let (client, llm_api_token) = this                        .read_with(cx, |this, _cx| (client.clone(), this.llm_api_token.clone()))?;                     loop {                        let is_authenticated = user_store                            .read_with(cx, |user_store, _cx| user_store.current_user().is_some())?;                        if is_authenticated {                            break;                        }                         cx.background_executor()                            .timer(Duration::from_millis(100))                            .await;                    }                     let response = Self::fetch_models(client, llm_api_token).await?;                    this.update(cx, |this, cx| {                        this.update_models(response, cx);                    })                })                .await                .context("failed to fetch Zed models")                .log_err();            }),            _settings_subscription: cx.observe_global::<SettingsStore>(|_, cx| {                cx.notify();            }),            _llm_token_subscription: cx.subscribe(                &refresh_llm_token_listener,                move |this, _listener, _event, cx| {                    let client = this.client.clone();                    let llm_api_token = this.llm_api_token.clone();                    cx.spawn(async move |this, cx| {                        llm_api_token.refresh(&client).await?;                        let response = Self::fetch_models(client, llm_api_token).await?;                        this.update(cx, |this, cx| {                            this.update_models(response, cx);                        })                    })                    .detach_and_log_err(cx);                },            ),        }    }     fn is_signed_out(&self, cx: &App) -> bool {        self.user_store.read(cx).current_user().is_none()    }     fn authenticate(&self, cx: &mut Context<Self>) -> Task<Result<()>> {        let client = self.client.clone();        cx.spawn(async move |state, cx| {            client.sign_in_with_optional_connect(true, &cx).await?;            state.update(cx, |_, cx| cx.notify())        })    }     fn has_accepted_terms_of_service(&self, cx: &App) -> bool {        self.user_store.read(cx).has_accepted_terms_of_service()    }     fn accept_terms_of_service(&mut self, cx: &mut Context<Self>) {        let user_store = self.user_store.clone();        self.accept_terms_of_service_task = Some(cx.spawn(async move |this, cx| {            let _ = user_store                .update(cx, |store, cx| store.accept_terms_of_service(cx))?                .await;            this.update(cx, |this, cx| {                this.accept_terms_of_service_task = None;                cx.notify()            })        }));    }     fn update_models(&mut self, response: ListModelsResponse, cx: &mut Context<Self>) {        let mut models = Vec::new();         for model in response.models {            models.push(Arc::new(model.clone()));             // Right now we represent thinking variants of models as separate models on the client,            // so we need to insert variants for any model that supports thinking.            if model.supports_thinking {                models.push(Arc::new(cloud_llm_client::LanguageModel {                    id: cloud_llm_client::LanguageModelId(format!("{}-thinking", model.id).into()),                    display_name: format!("{} Thinking", model.display_name),                    ..model                }));            }        }         self.default_model = models            .iter()            .find(|model| model.id == response.default_model)            .cloned();        self.default_fast_model = models            .iter()            .find(|model| model.id == response.default_fast_model)            .cloned();        self.recommended_models = response            .recommended_models            .iter()            .filter_map(|id| models.iter().find(|model| &model.id == id))            .cloned()            .collect();        self.models = models;        cx.notify();    }     async fn fetch_models(        client: Arc<Client>,        llm_api_token: LlmApiToken,    ) -> Result<ListModelsResponse> {        let http_client = &client.http_client();        let token = llm_api_token.acquire(&client).await?;         let request = http_client::Request::builder()            .method(Method::GET)            .uri(http_client.build_zed_llm_url("/models", &[])?.as_ref())            .header("Authorization", format!("Bearer {token}"))            .body(AsyncBody::empty())?;        let mut response = http_client            .send(request)            .await            .context("failed to send list models request")?;         if response.status().is_success() {            let mut body = String::new();            response.body_mut().read_to_string(&mut body).await?;            return Ok(serde_json::from_str(&body)?);        } else {            let mut body = String::new();            response.body_mut().read_to_string(&mut body).await?;            anyhow::bail!(                "error listing models.\nStatus: {:?}\nBody: {body}",                response.status(),            );        }    }} impl CloudLanguageModelProvider {    pub fn new(user_store: Entity<UserStore>, client: Arc<Client>, cx: &mut App) -> Self {        let mut status_rx = client.status();        let status = *status_rx.borrow();         let state = cx.new(|cx| State::new(client.clone(), user_store.clone(), status, cx));         let state_ref = state.downgrade();        let maintain_client_status = cx.spawn(async move |cx| {            while let Some(status) = status_rx.next().await {                if let Some(this) = state_ref.upgrade() {                    _ = this.update(cx, |this, cx| {                        if this.status != status {                            this.status = status;                            cx.notify();                        }                    });                } else {                    break;                }            }        });         Self {            client,            state: state.clone(),            _maintain_client_status: maintain_client_status,        }    }     fn create_language_model(        &self,        model: Arc<cloud_llm_client::LanguageModel>,        llm_api_token: LlmApiToken,    ) -> Arc<dyn LanguageModel> {        Arc::new(CloudLanguageModel {            id: LanguageModelId(SharedString::from(model.id.0.clone())),            model,            llm_api_token: llm_api_token.clone(),            client: self.client.clone(),            request_limiter: RateLimiter::new(4),        })    }} impl LanguageModelProviderState for CloudLanguageModelProvider {    type ObservableEntity = State;     fn observable_entity(&self) -> Option<gpui::Entity<Self::ObservableEntity>> {        Some(self.state.clone())    }} impl LanguageModelProvider for CloudLanguageModelProvider {    fn id(&self) -> LanguageModelProviderId {        PROVIDER_ID    }     fn name(&self) -> LanguageModelProviderName {        PROVIDER_NAME    }     fn icon(&self) -> IconName {        IconName::AiZed    }     fn default_model(&self, cx: &App) -> Option<Arc<dyn LanguageModel>> {        let default_model = self.state.read(cx).default_model.clone()?;        let llm_api_token = self.state.read(cx).llm_api_token.clone();        Some(self.create_language_model(default_model, llm_api_token))    }     fn default_fast_model(&self, cx: &App) -> Option<Arc<dyn LanguageModel>> {        let default_fast_model = self.state.read(cx).default_fast_model.clone()?;        let llm_api_token = self.state.read(cx).llm_api_token.clone();        Some(self.create_language_model(default_fast_model, llm_api_token))    }     fn recommended_models(&self, cx: &App) -> Vec<Arc<dyn LanguageModel>> {        let llm_api_token = self.state.read(cx).llm_api_token.clone();        self.state            .read(cx)            .recommended_models            .iter()            .cloned()            .map(|model| self.create_language_model(model, llm_api_token.clone()))            .collect()    }     fn provided_models(&self, cx: &App) -> Vec<Arc<dyn LanguageModel>> {        let llm_api_token = self.state.read(cx).llm_api_token.clone();        self.state            .read(cx)            .models            .iter()            .cloned()            .map(|model| self.create_language_model(model, llm_api_token.clone()))            .collect()    }     fn is_authenticated(&self, cx: &App) -> bool {        let state = self.state.read(cx);        !state.is_signed_out(cx) && state.has_accepted_terms_of_service(cx)    }     fn authenticate(&self, _cx: &mut App) -> Task<Result<(), AuthenticateError>> {        Task::ready(Ok(()))    }     fn configuration_view(&self, _: &mut Window, cx: &mut App) -> AnyView {        cx.new(|_| ConfigurationView::new(self.state.clone()))            .into()    }     fn must_accept_terms(&self, cx: &App) -> bool {        !self.state.read(cx).has_accepted_terms_of_service(cx)    }     fn render_accept_terms(        &self,        view: LanguageModelProviderTosView,        cx: &mut App,    ) -> Option<AnyElement> {        let state = self.state.read(cx);        if state.has_accepted_terms_of_service(cx) {            return None;        }        Some(            render_accept_terms(view, state.accept_terms_of_service_task.is_some(), {                let state = self.state.clone();                move |_window, cx| {                    state.update(cx, |state, cx| state.accept_terms_of_service(cx));                }            })            .into_any_element(),        )    }     fn reset_credentials(&self, _cx: &mut App) -> Task<Result<()>> {        Task::ready(Ok(()))    }} fn render_accept_terms(    view_kind: LanguageModelProviderTosView,    accept_terms_of_service_in_progress: bool,    accept_terms_callback: impl Fn(&mut Window, &mut App) + 'static,) -> impl IntoElement {    let thread_fresh_start = matches!(view_kind, LanguageModelProviderTosView::ThreadFreshStart);    let thread_empty_state = matches!(view_kind, LanguageModelProviderTosView::ThreadEmptyState);     let terms_button = Button::new("terms_of_service", "Terms of Service")        .style(ButtonStyle::Subtle)        .icon(IconName::ArrowUpRight)        .icon_color(Color::Muted)        .icon_size(IconSize::XSmall)        .when(thread_empty_state, |this| this.label_size(LabelSize::Small))        .on_click(move |_, _window, cx| cx.open_url("https://zed.dev/terms-of-service"));     let button_container = h_flex().child(        Button::new("accept_terms", "I accept the Terms of Service")            .when(!thread_empty_state, |this| {                this.full_width()                    .style(ButtonStyle::Tinted(TintColor::Accent))                    .icon(IconName::Check)                    .icon_position(IconPosition::Start)                    .icon_size(IconSize::Small)            })            .when(thread_empty_state, |this| {                this.style(ButtonStyle::Tinted(TintColor::Warning))                    .label_size(LabelSize::Small)            })            .disabled(accept_terms_of_service_in_progress)            .on_click(move |_, window, cx| (accept_terms_callback)(window, cx)),    );     if thread_empty_state {        h_flex()            .w_full()            .flex_wrap()            .justify_between()            .child(                h_flex()                    .child(                        Label::new("To start using Zed AI, please read and accept the")                            .size(LabelSize::Small),                    )                    .child(terms_button),            )            .child(button_container)    } else {        v_flex()            .w_full()            .gap_2()            .child(                h_flex()                    .flex_wrap()                    .when(thread_fresh_start, |this| this.justify_center())                    .child(Label::new(                        "To start using Zed AI, please read and accept the",                    ))                    .child(terms_button),            )            .child({                match view_kind {                    LanguageModelProviderTosView::TextThreadPopup => {                        button_container.w_full().justify_end()                    }                    LanguageModelProviderTosView::Configuration => {

Lines 1–500 of 1487. Load more