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, CLIENT_SUPPORTS_X_AI_HEADER_NAME,    CURRENT_PLAN_HEADER_NAME, CompletionBody, CompletionEvent, CompletionRequestStatus,    CountTokensBody, CountTokensResponse, EXPIRED_LLM_TOKEN_HEADER_NAME, ListModelsResponse,    MODEL_REQUESTS_RESOURCE_HEADER_VALUE, Plan, PlanV1, PlanV2,    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, Subscription, Task};use http_client::http::{HeaderMap, HeaderValue};use http_client::{AsyncBody, HttpClient, HttpRequestExt, Method, Response, StatusCode};use language_model::{    AuthenticateError, IconOrSvg, LanguageModel, LanguageModelCacheConfiguration,    LanguageModelCompletionError, LanguageModelCompletionEvent, LanguageModelId, LanguageModelName,    LanguageModelProvider, LanguageModelProviderId, LanguageModelProviderName,    LanguageModelProviderState, LanguageModelRequest, LanguageModelToolChoice,    LanguageModelToolSchemaFormat, LlmApiToken, ModelRequestLimitReachedError,    PaymentRequiredError, RateLimiter, RefreshLlmTokenListener,};use release_channel::AppVersion;use schemars::JsonSchema;use semver::Version;use serde::{Deserialize, Serialize, de::DeserializeOwned};use settings::SettingsStore;pub use settings::ZedDotDevAvailableModel as AvailableModel;pub use settings::ZedDotDevAvailableProvider as AvailableProvider;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_with_tiktoken, into_anthropic,};use crate::provider::google::{GoogleEventMapper, into_google};use crate::provider::open_ai::{OpenAiEventMapper, count_open_ai_tokens, into_open_ai};use crate::provider::x_ai::count_xai_tokens; 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(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: Entity<State>,    _maintain_client_status: Task<()>,} pub struct State {    client: Arc<Client>,    llm_api_token: LlmApiToken,    user_store: Entity<UserStore>,    status: client::Status,    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);        let mut current_user = user_store.read(cx).watch_current_user();        Self {            client: client.clone(),            llm_api_token: LlmApiToken::default(),            user_store,            status,            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()))?;                     while current_user.borrow().is_none() {                        current_user.next().await;                    }                     let response =                        Self::fetch_models(client.clone(), llm_api_token.clone()).await?;                    this.update(cx, |this, cx| this.update_models(response, cx))?;                    anyhow::Ok(())                })                .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 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| {                response                    .default_model                    .as_ref()                    .is_some_and(|default_model_id| &model.id == default_model_id)            })            .cloned();        self.default_fast_model = models            .iter()            .find(|model| {                response                    .default_fast_model                    .as_ref()                    .is_some_and(|default_fast_model_id| &model.id == default_fast_model_id)            })            .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)            .header(CLIENT_SUPPORTS_X_AI_HEADER_NAME, "true")            .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?;            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,            _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,            client: self.client.clone(),            request_limiter: RateLimiter::new(4),        })    }} impl LanguageModelProviderState for CloudLanguageModelProvider {    type ObservableEntity = State;     fn observable_entity(&self) -> Option<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) -> IconOrSvg {        IconOrSvg::Icon(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)    }     fn authenticate(&self, _cx: &mut App) -> Task<Result<(), AuthenticateError>> {        Task::ready(Ok(()))    }     fn configuration_view(        &self,        _target_agent: language_model::ConfigurationViewTargetAgent,        _: &mut Window,        cx: &mut App,    ) -> AnyView {        cx.new(|_| ConfigurationView::new(self.state.clone()))            .into()    }     fn reset_credentials(&self, _cx: &mut App) -> Task<Result<()>> {        Task::ready(Ok(()))    }} pub struct CloudLanguageModel {    id: LanguageModelId,    model: Arc<cloud_llm_client::LanguageModel>,    llm_api_token: LlmApiToken,    client: Arc<Client>,    request_limiter: RateLimiter,} struct PerformLlmCompletionResponse {    response: Response<AsyncBody>,    usage: Option<ModelRequestUsage>,    tool_use_limit_reached: bool,    includes_status_messages: bool,} impl CloudLanguageModel {    async fn perform_llm_completion(        client: Arc<Client>,        llm_api_token: LlmApiToken,        app_version: Option<Version>,        body: CompletionBody,    ) -> Result<PerformLlmCompletionResponse> {        let http_client = &client.http_client();         let mut token = llm_api_token.acquire(&client).await?;        let mut refreshed_token = false;         loop {            let request = http_client::Request::builder()                .method(Method::POST)                .uri(http_client.build_zed_llm_url("/completions", &[])?.as_ref())                .when_some(app_version.as_ref(), |builder, app_version| {                    builder.header(ZED_VERSION_HEADER_NAME, app_version.to_string())                })                .header("Content-Type", "application/json")                .header("Authorization", format!("Bearer {token}"))                .header(CLIENT_SUPPORTS_STATUS_MESSAGES_HEADER_NAME, "true")                .body(serde_json::to_string(&body)?.into())?;             let mut response = http_client.send(request).await?;            let status = response.status();            if status.is_success() {                let includes_status_messages = response                    .headers()                    .get(SERVER_SUPPORTS_STATUS_MESSAGES_HEADER_NAME)                    .is_some();                 let tool_use_limit_reached = response                    .headers()                    .get(TOOL_USE_LIMIT_REACHED_HEADER_NAME)                    .is_some();                 let usage = if includes_status_messages {                    None                } else {                    ModelRequestUsage::from_headers(response.headers()).ok()                };                 return Ok(PerformLlmCompletionResponse {                    response,                    usage,                    includes_status_messages,                    tool_use_limit_reached,                });            }             if !refreshed_token                && response                    .headers()                    .get(EXPIRED_LLM_TOKEN_HEADER_NAME)                    .is_some()            {                token = llm_api_token.refresh(&client).await?;                refreshed_token = true;                continue;            }             if status == StatusCode::FORBIDDEN                && response                    .headers()                    .get(SUBSCRIPTION_LIMIT_RESOURCE_HEADER_NAME)                    .is_some()            {                if let Some(MODEL_REQUESTS_RESOURCE_HEADER_VALUE) = response                    .headers()                    .get(SUBSCRIPTION_LIMIT_RESOURCE_HEADER_NAME)                    .and_then(|resource| resource.to_str().ok())                    && let Some(plan) = response                        .headers()                        .get(CURRENT_PLAN_HEADER_NAME)                        .and_then(|plan| plan.to_str().ok())                        .and_then(|plan| cloud_llm_client::PlanV1::from_str(plan).ok())                        .map(Plan::V1)                {                    return Err(anyhow!(ModelRequestLimitReachedError { plan }));                }            } else if status == StatusCode::PAYMENT_REQUIRED {                return Err(anyhow!(PaymentRequiredError));            }             let mut body = String::new();            let headers = response.headers().clone();            response.body_mut().read_to_string(&mut body).await?;            return Err(anyhow!(ApiError {                status,                body,                headers            }));        }    }} #[derive(Debug, Error)]#[error("cloud language model request failed with status {status}: {body}")]struct ApiError {    status: StatusCode,    body: String,    headers: HeaderMap<HeaderValue>,} /// Represents error responses from Zed's cloud API.////// Example JSON for an upstream HTTP error:/// ```json/// {///   "code": "upstream_http_error",///   "message": "Received an error from the Anthropic API: upstream connect error or disconnect/reset before headers, reset reason: connection timeout",///   "upstream_status": 503/// }/// ```#[derive(Debug, serde::Deserialize)]struct CloudApiError {

Lines 1–500 of 1402. Load more