ZIm/crates/language_model/src/provider/cloud.rs

use super::open_ai::count_open_ai_tokens;
use crate::{
    settings::AllLanguageModelSettings, CloudModel, LanguageModel, LanguageModelId,
    LanguageModelName, LanguageModelProviderId, LanguageModelProviderName,
    LanguageModelProviderState, LanguageModelRequest, RateLimiter, ZedModel,
};
use anyhow::{anyhow, Context as _, Result};
use client::{Client, PerformCompletionParams, UserStore, EXPIRED_LLM_TOKEN_HEADER_NAME};
use collections::BTreeMap;
use feature_flags::{FeatureFlag, FeatureFlagAppExt};
use futures::{future::BoxFuture, stream::BoxStream, AsyncBufReadExt, FutureExt, StreamExt};
use gpui::{AnyView, AppContext, AsyncAppContext, Model, ModelContext, Subscription, Task};
use http_client::{HttpClient, Method};
use schemars::JsonSchema;
use serde::{Deserialize, Serialize};
use serde_json::value::RawValue;
use settings::{Settings, SettingsStore};
use smol::{
    io::BufReader,
    lock::{RwLock, RwLockUpgradableReadGuard, RwLockWriteGuard},
};
use std::{future, sync::Arc};
use strum::IntoEnumIterator;
use ui::prelude::*;

use crate::{LanguageModelAvailability, LanguageModelProvider};

use super::anthropic::count_anthropic_tokens;

pub const PROVIDER_ID: &str = "zed.dev";
pub const PROVIDER_NAME: &str = "Zed";

#[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 {
    provider: AvailableProvider,
    name: String,
    max_tokens: usize,
    tool_override: Option<String>,
}

pub struct CloudLanguageModelProvider {
    client: Arc<Client>,
    llm_api_token: LlmApiToken,
    state: gpui::Model<State>,
    _maintain_client_status: Task<()>,
}

pub struct State {
    client: Arc<Client>,
    user_store: Model<UserStore>,
    status: client::Status,
    _subscription: Subscription,
}

impl State {
    fn is_signed_out(&self) -> bool {
        self.status.is_signed_out()
    }

    fn authenticate(&self, cx: &mut ModelContext<Self>) -> Task<Result<()>> {
        let client = self.client.clone();
        cx.spawn(move |this, mut cx| async move {
            client.authenticate_and_connect(true, &cx).await?;
            this.update(&mut cx, |_, cx| cx.notify())
        })
    }
}

impl CloudLanguageModelProvider {
    pub fn new(user_store: Model<UserStore>, client: Arc<Client>, cx: &mut AppContext) -> Self {
        let mut status_rx = client.status();
        let status = *status_rx.borrow();

        let state = cx.new_model(|cx| State {
            client: client.clone(),
            user_store,
            status,
            _subscription: cx.observe_global::<SettingsStore>(|_, cx| {
                cx.notify();
            }),
        });

        let state_ref = state.downgrade();
        let maintain_client_status = cx.spawn(|mut cx| async move {
            while let Some(status) = status_rx.next().await {
                if let Some(this) = state_ref.upgrade() {
                    _ = this.update(&mut cx, |this, cx| {
                        if this.status != status {
                            this.status = status;
                            cx.notify();
                        }
                    });
                } else {
                    break;
                }
            }
        });

        Self {
            client,
            state,
            llm_api_token: LlmApiToken::default(),
            _maintain_client_status: maintain_client_status,
        }
    }
}

impl LanguageModelProviderState for CloudLanguageModelProvider {
    type ObservableEntity = State;

    fn observable_entity(&self) -> Option<gpui::Model<Self::ObservableEntity>> {
        Some(self.state.clone())
    }
}

impl LanguageModelProvider for CloudLanguageModelProvider {
    fn id(&self) -> LanguageModelProviderId {
        LanguageModelProviderId(PROVIDER_ID.into())
    }

    fn name(&self) -> LanguageModelProviderName {
        LanguageModelProviderName(PROVIDER_NAME.into())
    }

    fn icon(&self) -> IconName {
        IconName::AiZed
    }

    fn provided_models(&self, cx: &AppContext) -> Vec<Arc<dyn LanguageModel>> {
        let mut models = BTreeMap::default();

        for model in anthropic::Model::iter() {
            if !matches!(model, anthropic::Model::Custom { .. }) {
                models.insert(model.id().to_string(), CloudModel::Anthropic(model));
            }
        }
        for model in open_ai::Model::iter() {
            if !matches!(model, open_ai::Model::Custom { .. }) {
                models.insert(model.id().to_string(), CloudModel::OpenAi(model));
            }
        }
        for model in google_ai::Model::iter() {
            if !matches!(model, google_ai::Model::Custom { .. }) {
                models.insert(model.id().to_string(), CloudModel::Google(model));
            }
        }
        for model in ZedModel::iter() {
            models.insert(model.id().to_string(), CloudModel::Zed(model));
        }

        // Override with available models from settings
        for model in &AllLanguageModelSettings::get_global(cx)
            .zed_dot_dev
            .available_models
        {
            let model = match model.provider {
                AvailableProvider::Anthropic => CloudModel::Anthropic(anthropic::Model::Custom {
                    name: model.name.clone(),
                    max_tokens: model.max_tokens,
                    tool_override: model.tool_override.clone(),
                }),
                AvailableProvider::OpenAi => CloudModel::OpenAi(open_ai::Model::Custom {
                    name: model.name.clone(),
                    max_tokens: model.max_tokens,
                }),
                AvailableProvider::Google => CloudModel::Google(google_ai::Model::Custom {
                    name: model.name.clone(),
                    max_tokens: model.max_tokens,
                }),
            };
            models.insert(model.id().to_string(), model.clone());
        }

        models
            .into_values()
            .map(|model| {
                Arc::new(CloudLanguageModel {
                    id: LanguageModelId::from(model.id().to_string()),
                    model,
                    llm_api_token: self.llm_api_token.clone(),
                    client: self.client.clone(),
                    request_limiter: RateLimiter::new(4),
                }) as Arc<dyn LanguageModel>
            })
            .collect()
    }

    fn is_authenticated(&self, cx: &AppContext) -> bool {
        !self.state.read(cx).is_signed_out()
    }

    fn authenticate(&self, _cx: &mut AppContext) -> Task<Result<()>> {
        Task::ready(Ok(()))
    }

    fn configuration_view(&self, cx: &mut WindowContext) -> AnyView {
        cx.new_view(|_cx| ConfigurationView {
            state: self.state.clone(),
        })
        .into()
    }

    fn reset_credentials(&self, _cx: &mut AppContext) -> Task<Result<()>> {
        Task::ready(Ok(()))
    }
}

struct LlmServiceFeatureFlag;

impl FeatureFlag for LlmServiceFeatureFlag {
    const NAME: &'static str = "llm-service";

    fn enabled_for_staff() -> bool {
        false
    }
}

pub struct CloudLanguageModel {
    id: LanguageModelId,
    model: CloudModel,
    llm_api_token: LlmApiToken,
    client: Arc<Client>,
    request_limiter: RateLimiter,
}

#[derive(Clone, Default)]
struct LlmApiToken(Arc<RwLock<Option<String>>>);

impl LanguageModel for CloudLanguageModel {
    fn id(&self) -> LanguageModelId {
        self.id.clone()
    }

    fn name(&self) -> LanguageModelName {
        LanguageModelName::from(self.model.display_name().to_string())
    }

    fn provider_id(&self) -> LanguageModelProviderId {
        LanguageModelProviderId(PROVIDER_ID.into())
    }

    fn provider_name(&self) -> LanguageModelProviderName {
        LanguageModelProviderName(PROVIDER_NAME.into())
    }

    fn telemetry_id(&self) -> String {
        format!("zed.dev/{}", self.model.id())
    }

    fn availability(&self) -> LanguageModelAvailability {
        self.model.availability()
    }

    fn max_token_count(&self) -> usize {
        self.model.max_token_count()
    }

    fn count_tokens(
        &self,
        request: LanguageModelRequest,
        cx: &AppContext,
    ) -> BoxFuture<'static, Result<usize>> {
        match self.model.clone() {
            CloudModel::Anthropic(_) => count_anthropic_tokens(request, cx),
            CloudModel::OpenAi(model) => count_open_ai_tokens(request, model, cx),
            CloudModel::Google(model) => {
                let client = self.client.clone();
                let request = request.into_google(model.id().into());
                let request = google_ai::CountTokensRequest {
                    contents: request.contents,
                };
                async move {
                    let request = serde_json::to_string(&request)?;
                    let response = client
                        .request(proto::CountLanguageModelTokens {
                            provider: proto::LanguageModelProvider::Google as i32,
                            request,
                        })
                        .await?;
                    Ok(response.token_count as usize)
                }
                .boxed()
            }
            CloudModel::Zed(_) => {
                count_open_ai_tokens(request, open_ai::Model::ThreePointFiveTurbo, cx)
            }
        }
    }

    fn stream_completion(
        &self,
        request: LanguageModelRequest,
        cx: &AsyncAppContext,
    ) -> BoxFuture<'static, Result<BoxStream<'static, Result<String>>>> {
        match &self.model {
            CloudModel::Anthropic(model) => {
                let request = request.into_anthropic(model.id().into());
                let client = self.client.clone();

                if cx
                    .update(|cx| cx.has_flag::<LlmServiceFeatureFlag>())
                    .unwrap_or(false)
                {
                    let http_client = self.client.http_client();
                    let llm_api_token = self.llm_api_token.clone();
                    let future = self.request_limiter.stream(async move {
                        let request = serde_json::to_string(&request)?;
                        let mut token = llm_api_token.acquire(&client).await?;
                        let mut did_retry = false;

                        let response = loop {
                            let request = http_client::Request::builder()
                                .method(Method::POST)
                                .uri(http_client.build_zed_llm_url("/completion", &[])?.as_ref())
                                .header("Content-Type", "application/json")
                                .header("Authorization", format!("Bearer {token}"))
                                .body(
                                    serde_json::to_string(&PerformCompletionParams {
                                        provider_request: RawValue::from_string(request.clone())?,
                                    })?
                                    .into(),
                                )?;
                            let response = http_client.send(request).await?;
                            if response.status().is_success() {
                                break response;
                            } else if !did_retry
                                && response
                                    .headers()
                                    .get(EXPIRED_LLM_TOKEN_HEADER_NAME)
                                    .is_some()
                            {
                                did_retry = true;
                                token = llm_api_token.refresh(&client).await?;
                            } else {
                                break Err(anyhow!(
                                    "cloud language model completion failed with status {}",
                                    response.status()
                                ))?;
                            }
                        };

                        let body = BufReader::new(response.into_body());

                        let stream =
                            futures::stream::try_unfold(body, move |mut body| async move {
                                let mut buffer = String::new();
                                match body.read_line(&mut buffer).await {
                                    Ok(0) => Ok(None),
                                    Ok(_) => {
                                        let event: anthropic::Event =
                                            serde_json::from_str(&buffer)?;
                                        Ok(Some((event, body)))
                                    }
                                    Err(e) => Err(e.into()),
                                }
                            });

                        Ok(anthropic::extract_text_from_events(stream))
                    });
                    async move { Ok(future.await?.boxed()) }.boxed()
                } else {
                    let future = self.request_limiter.stream(async move {
                        let request = serde_json::to_string(&request)?;
                        let stream = client
                            .request_stream(proto::StreamCompleteWithLanguageModel {
                                provider: proto::LanguageModelProvider::Anthropic as i32,
                                request,
                            })
                            .await?
                            .map(|event| Ok(serde_json::from_str(&event?.event)?));
                        Ok(anthropic::extract_text_from_events(stream))
                    });
                    async move { Ok(future.await?.boxed()) }.boxed()
                }
            }
            CloudModel::OpenAi(model) => {
                let client = self.client.clone();
                let request = request.into_open_ai(model.id().into());
                let future = self.request_limiter.stream(async move {
                    let request = serde_json::to_string(&request)?;
                    let stream = client
                        .request_stream(proto::StreamCompleteWithLanguageModel {
                            provider: proto::LanguageModelProvider::OpenAi as i32,
                            request,
                        })
                        .await?;
                    Ok(open_ai::extract_text_from_events(
                        stream.map(|item| Ok(serde_json::from_str(&item?.event)?)),
                    ))
                });
                async move { Ok(future.await?.boxed()) }.boxed()
            }
            CloudModel::Google(model) => {
                let client = self.client.clone();
                let request = request.into_google(model.id().into());
                let future = self.request_limiter.stream(async move {
                    let request = serde_json::to_string(&request)?;
                    let stream = client
                        .request_stream(proto::StreamCompleteWithLanguageModel {
                            provider: proto::LanguageModelProvider::Google as i32,
                            request,
                        })
                        .await?;
                    Ok(google_ai::extract_text_from_events(
                        stream.map(|item| Ok(serde_json::from_str(&item?.event)?)),
                    ))
                });
                async move { Ok(future.await?.boxed()) }.boxed()
            }
            CloudModel::Zed(model) => {
                let client = self.client.clone();
                let mut request = request.into_open_ai(model.id().into());
                request.max_tokens = Some(4000);
                let future = self.request_limiter.stream(async move {
                    let request = serde_json::to_string(&request)?;
                    let stream = client
                        .request_stream(proto::StreamCompleteWithLanguageModel {
                            provider: proto::LanguageModelProvider::Zed as i32,
                            request,
                        })
                        .await?;
                    Ok(open_ai::extract_text_from_events(
                        stream.map(|item| Ok(serde_json::from_str(&item?.event)?)),
                    ))
                });
                async move { Ok(future.await?.boxed()) }.boxed()
            }
        }
    }

    fn use_any_tool(
        &self,
        request: LanguageModelRequest,
        tool_name: String,
        tool_description: String,
        input_schema: serde_json::Value,
        _cx: &AsyncAppContext,
    ) -> BoxFuture<'static, Result<serde_json::Value>> {
        match &self.model {
            CloudModel::Anthropic(model) => {
                let client = self.client.clone();
                let mut request = request.into_anthropic(model.tool_model_id().into());
                request.tool_choice = Some(anthropic::ToolChoice::Tool {
                    name: tool_name.clone(),
                });
                request.tools = vec![anthropic::Tool {
                    name: tool_name.clone(),
                    description: tool_description,
                    input_schema,
                }];

                self.request_limiter
                    .run(async move {
                        let request = serde_json::to_string(&request)?;
                        let response = client
                            .request(proto::CompleteWithLanguageModel {
                                provider: proto::LanguageModelProvider::Anthropic as i32,
                                request,
                            })
                            .await?;
                        let response: anthropic::Response =
                            serde_json::from_str(&response.completion)?;
                        response
                            .content
                            .into_iter()
                            .find_map(|content| {
                                if let anthropic::Content::ToolUse { name, input, .. } = content {
                                    if name == tool_name {
                                        Some(input)
                                    } else {
                                        None
                                    }
                                } else {
                                    None
                                }
                            })
                            .context("tool not used")
                    })
                    .boxed()
            }
            CloudModel::OpenAi(_) => {
                future::ready(Err(anyhow!("tool use not implemented for OpenAI"))).boxed()
            }
            CloudModel::Google(_) => {
                future::ready(Err(anyhow!("tool use not implemented for Google AI"))).boxed()
            }
            CloudModel::Zed(_) => {
                future::ready(Err(anyhow!("tool use not implemented for Zed models"))).boxed()
            }
        }
    }
}

impl LlmApiToken {
    async fn acquire(&self, client: &Arc<Client>) -> Result<String> {
        let lock = self.0.upgradable_read().await;
        if let Some(token) = lock.as_ref() {
            Ok(token.to_string())
        } else {
            Self::fetch(RwLockUpgradableReadGuard::upgrade(lock).await, &client).await
        }
    }

    async fn refresh(&self, client: &Arc<Client>) -> Result<String> {
        Self::fetch(self.0.write().await, &client).await
    }

    async fn fetch<'a>(
        mut lock: RwLockWriteGuard<'a, Option<String>>,
        client: &Arc<Client>,
    ) -> Result<String> {
        let response = client.request(proto::GetLlmToken {}).await?;
        *lock = Some(response.token.clone());
        Ok(response.token.clone())
    }
}

struct ConfigurationView {
    state: gpui::Model<State>,
}

impl ConfigurationView {
    fn authenticate(&mut self, cx: &mut ViewContext<Self>) {
        self.state.update(cx, |state, cx| {
            state.authenticate(cx).detach_and_log_err(cx);
        });
        cx.notify();
    }
}

impl Render for ConfigurationView {
    fn render(&mut self, cx: &mut ViewContext<Self>) -> impl IntoElement {
        const ZED_AI_URL: &str = "https://zed.dev/ai";
        const ACCOUNT_SETTINGS_URL: &str = "https://zed.dev/account";

        let is_connected = !self.state.read(cx).is_signed_out();
        let plan = self.state.read(cx).user_store.read(cx).current_plan();

        let is_pro = plan == Some(proto::Plan::ZedPro);

        if is_connected {
            v_flex()
                .gap_3()
                .max_w_4_5()
                .child(Label::new(
                    if is_pro {
                        "You have full access to Zed's hosted models from Anthropic, OpenAI, Google with faster speeds and higher limits through Zed Pro."
                    } else {
                        "You have basic access to models from Anthropic, OpenAI, Google and more through the Zed AI Free plan."
                    }))
                .child(
                    if is_pro {
                        h_flex().child(
                        Button::new("manage_settings", "Manage Subscription")
                            .style(ButtonStyle::Filled)
                            .on_click(cx.listener(|_, _, cx| {
                                cx.open_url(ACCOUNT_SETTINGS_URL)
                            })))
                    } else {
                        h_flex()
                            .gap_2()
                            .child(
                        Button::new("learn_more", "Learn more")
                            .style(ButtonStyle::Subtle)
                            .on_click(cx.listener(|_, _, cx| {
                                cx.open_url(ZED_AI_URL)
                            })))
                            .child(
                        Button::new("upgrade", "Upgrade")
                            .style(ButtonStyle::Subtle)
                            .color(Color::Accent)
                            .on_click(cx.listener(|_, _, cx| {
                                cx.open_url(ACCOUNT_SETTINGS_URL)
                            })))
                    },
                )
        } else {
            v_flex()
                .gap_6()
                .child(Label::new("Use the zed.dev to access language models."))
                .child(
                    v_flex()
                        .gap_2()
                        .child(
                            Button::new("sign_in", "Sign in")
                                .icon_color(Color::Muted)
                                .icon(IconName::Github)
                                .icon_position(IconPosition::Start)
                                .style(ButtonStyle::Filled)
                                .full_width()
                                .on_click(cx.listener(move |this, _, cx| this.authenticate(cx))),
                        )
                        .child(
                            div().flex().w_full().items_center().child(
                                Label::new("Sign in to enable collaboration.")
                                    .color(Color::Muted)
                                    .size(LabelSize::Small),
                            ),
                        ),
                )
        }
    }
}