cloud_llm_client.rs

  1use std::str::FromStr;
  2use std::sync::Arc;
  3
  4use anyhow::Context as _;
  5use serde::{Deserialize, Serialize};
  6use strum::{Display, EnumIter, EnumString};
  7use uuid::Uuid;
  8
  9/// The name of the header used to indicate which version of Zed the client is running.
 10pub const ZED_VERSION_HEADER_NAME: &str = "x-zed-version";
 11
 12/// The name of the header used to indicate when a request failed due to an
 13/// expired LLM token.
 14///
 15/// The client may use this as a signal to refresh the token.
 16pub const EXPIRED_LLM_TOKEN_HEADER_NAME: &str = "x-zed-expired-token";
 17
 18/// The name of the header used to indicate what plan the user is currently on.
 19pub const CURRENT_PLAN_HEADER_NAME: &str = "x-zed-plan";
 20
 21/// The name of the header used to indicate the usage limit for model requests.
 22pub const MODEL_REQUESTS_USAGE_LIMIT_HEADER_NAME: &str = "x-zed-model-requests-usage-limit";
 23
 24/// The name of the header used to indicate the usage amount for model requests.
 25pub const MODEL_REQUESTS_USAGE_AMOUNT_HEADER_NAME: &str = "x-zed-model-requests-usage-amount";
 26
 27/// The name of the header used to indicate the usage limit for edit predictions.
 28pub const EDIT_PREDICTIONS_USAGE_LIMIT_HEADER_NAME: &str = "x-zed-edit-predictions-usage-limit";
 29
 30/// The name of the header used to indicate the usage amount for edit predictions.
 31pub const EDIT_PREDICTIONS_USAGE_AMOUNT_HEADER_NAME: &str = "x-zed-edit-predictions-usage-amount";
 32
 33/// The name of the header used to indicate the resource for which the subscription limit has been reached.
 34pub const SUBSCRIPTION_LIMIT_RESOURCE_HEADER_NAME: &str = "x-zed-subscription-limit-resource";
 35
 36pub const MODEL_REQUESTS_RESOURCE_HEADER_VALUE: &str = "model_requests";
 37pub const EDIT_PREDICTIONS_RESOURCE_HEADER_VALUE: &str = "edit_predictions";
 38
 39/// The name of the header used to indicate that the maximum number of consecutive tool uses has been reached.
 40pub const TOOL_USE_LIMIT_REACHED_HEADER_NAME: &str = "x-zed-tool-use-limit-reached";
 41
 42/// The name of the header used to indicate the the minimum required Zed version.
 43///
 44/// This can be used to force a Zed upgrade in order to continue communicating
 45/// with the LLM service.
 46pub const MINIMUM_REQUIRED_VERSION_HEADER_NAME: &str = "x-zed-minimum-required-version";
 47
 48/// The name of the header used by the client to indicate to the server that it supports receiving status messages.
 49pub const CLIENT_SUPPORTS_STATUS_MESSAGES_HEADER_NAME: &str =
 50    "x-zed-client-supports-status-messages";
 51
 52/// The name of the header used by the server to indicate to the client that it supports sending status messages.
 53pub const SERVER_SUPPORTS_STATUS_MESSAGES_HEADER_NAME: &str =
 54    "x-zed-server-supports-status-messages";
 55
 56#[derive(Debug, PartialEq, Clone, Copy, Serialize, Deserialize)]
 57#[serde(rename_all = "snake_case")]
 58pub enum UsageLimit {
 59    Limited(i32),
 60    Unlimited,
 61}
 62
 63impl FromStr for UsageLimit {
 64    type Err = anyhow::Error;
 65
 66    fn from_str(value: &str) -> Result<Self, Self::Err> {
 67        match value {
 68            "unlimited" => Ok(Self::Unlimited),
 69            limit => limit
 70                .parse::<i32>()
 71                .map(Self::Limited)
 72                .context("failed to parse limit"),
 73        }
 74    }
 75}
 76
 77#[derive(Debug, Clone, Copy, Default, PartialEq, Serialize, Deserialize)]
 78#[serde(rename_all = "snake_case")]
 79pub enum Plan {
 80    #[default]
 81    #[serde(alias = "Free")]
 82    ZedFree,
 83    #[serde(alias = "ZedPro")]
 84    ZedPro,
 85    #[serde(alias = "ZedProTrial")]
 86    ZedProTrial,
 87}
 88
 89impl Plan {
 90    pub fn as_str(&self) -> &'static str {
 91        match self {
 92            Plan::ZedFree => "zed_free",
 93            Plan::ZedPro => "zed_pro",
 94            Plan::ZedProTrial => "zed_pro_trial",
 95        }
 96    }
 97
 98    pub fn model_requests_limit(&self) -> UsageLimit {
 99        match self {
100            Plan::ZedPro => UsageLimit::Limited(500),
101            Plan::ZedProTrial => UsageLimit::Limited(150),
102            Plan::ZedFree => UsageLimit::Limited(50),
103        }
104    }
105
106    pub fn edit_predictions_limit(&self) -> UsageLimit {
107        match self {
108            Plan::ZedPro => UsageLimit::Unlimited,
109            Plan::ZedProTrial => UsageLimit::Unlimited,
110            Plan::ZedFree => UsageLimit::Limited(2_000),
111        }
112    }
113}
114
115impl FromStr for Plan {
116    type Err = anyhow::Error;
117
118    fn from_str(value: &str) -> Result<Self, Self::Err> {
119        match value {
120            "zed_free" => Ok(Plan::ZedFree),
121            "zed_pro" => Ok(Plan::ZedPro),
122            "zed_pro_trial" => Ok(Plan::ZedProTrial),
123            plan => Err(anyhow::anyhow!("invalid plan: {plan:?}")),
124        }
125    }
126}
127
128#[derive(
129    Debug, PartialEq, Eq, Hash, Clone, Copy, Serialize, Deserialize, EnumString, EnumIter, Display,
130)]
131#[serde(rename_all = "snake_case")]
132#[strum(serialize_all = "snake_case")]
133pub enum LanguageModelProvider {
134    Anthropic,
135    OpenAi,
136    Google,
137}
138
139#[derive(Debug, Clone, Serialize, Deserialize)]
140pub struct PredictEditsBody {
141    #[serde(skip_serializing_if = "Option::is_none", default)]
142    pub outline: Option<String>,
143    pub input_events: String,
144    pub input_excerpt: String,
145    #[serde(skip_serializing_if = "Option::is_none", default)]
146    pub speculated_output: Option<String>,
147    /// Whether the user provided consent for sampling this interaction.
148    #[serde(default, alias = "data_collection_permission")]
149    pub can_collect_data: bool,
150    #[serde(skip_serializing_if = "Option::is_none", default)]
151    pub diagnostic_groups: Option<Vec<(String, serde_json::Value)>>,
152    /// Info about the git repository state, only present when can_collect_data is true.
153    #[serde(skip_serializing_if = "Option::is_none", default)]
154    pub git_info: Option<PredictEditsGitInfo>,
155}
156
157#[derive(Debug, Clone, Serialize, Deserialize)]
158pub struct PredictEditsGitInfo {
159    /// SHA of git HEAD commit at time of prediction.
160    #[serde(skip_serializing_if = "Option::is_none", default)]
161    pub head_sha: Option<String>,
162    /// URL of the remote called `origin`.
163    #[serde(skip_serializing_if = "Option::is_none", default)]
164    pub remote_origin_url: Option<String>,
165    /// URL of the remote called `upstream`.
166    #[serde(skip_serializing_if = "Option::is_none", default)]
167    pub remote_upstream_url: Option<String>,
168}
169
170#[derive(Debug, Clone, Serialize, Deserialize)]
171pub struct PredictEditsResponse {
172    pub request_id: Uuid,
173    pub output_excerpt: String,
174}
175
176#[derive(Debug, Clone, Serialize, Deserialize)]
177pub struct AcceptEditPredictionBody {
178    pub request_id: Uuid,
179}
180
181#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Clone, Copy, Serialize, Deserialize)]
182#[serde(rename_all = "snake_case")]
183pub enum CompletionMode {
184    Normal,
185    Max,
186}
187
188#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Clone, Copy, Serialize, Deserialize)]
189#[serde(rename_all = "snake_case")]
190pub enum CompletionIntent {
191    UserPrompt,
192    ToolResults,
193    ThreadSummarization,
194    ThreadContextSummarization,
195    CreateFile,
196    EditFile,
197    InlineAssist,
198    TerminalInlineAssist,
199    GenerateGitCommitMessage,
200}
201
202#[derive(Debug, Serialize, Deserialize)]
203pub struct CompletionBody {
204    #[serde(skip_serializing_if = "Option::is_none", default)]
205    pub thread_id: Option<String>,
206    #[serde(skip_serializing_if = "Option::is_none", default)]
207    pub prompt_id: Option<String>,
208    #[serde(skip_serializing_if = "Option::is_none", default)]
209    pub intent: Option<CompletionIntent>,
210    #[serde(skip_serializing_if = "Option::is_none", default)]
211    pub mode: Option<CompletionMode>,
212    pub provider: LanguageModelProvider,
213    pub model: String,
214    pub provider_request: serde_json::Value,
215}
216
217#[derive(Debug, PartialEq, Clone, Serialize, Deserialize)]
218#[serde(rename_all = "snake_case")]
219pub enum CompletionRequestStatus {
220    Queued {
221        position: usize,
222    },
223    Started,
224    Failed {
225        code: String,
226        message: String,
227        request_id: Uuid,
228        /// Retry duration in seconds.
229        retry_after: Option<f64>,
230    },
231    UsageUpdated {
232        amount: usize,
233        limit: UsageLimit,
234    },
235    ToolUseLimitReached,
236}
237
238#[derive(Serialize, Deserialize)]
239#[serde(rename_all = "snake_case")]
240pub enum CompletionEvent<T> {
241    Status(CompletionRequestStatus),
242    Event(T),
243}
244
245impl<T> CompletionEvent<T> {
246    pub fn into_status(self) -> Option<CompletionRequestStatus> {
247        match self {
248            Self::Status(status) => Some(status),
249            Self::Event(_) => None,
250        }
251    }
252
253    pub fn into_event(self) -> Option<T> {
254        match self {
255            Self::Event(event) => Some(event),
256            Self::Status(_) => None,
257        }
258    }
259}
260
261#[derive(Serialize, Deserialize)]
262pub struct WebSearchBody {
263    pub query: String,
264}
265
266#[derive(Debug, Serialize, Deserialize, Clone)]
267pub struct WebSearchResponse {
268    pub results: Vec<WebSearchResult>,
269}
270
271#[derive(Debug, Serialize, Deserialize, Clone)]
272pub struct WebSearchResult {
273    pub title: String,
274    pub url: String,
275    pub text: String,
276}
277
278#[derive(Serialize, Deserialize)]
279pub struct CountTokensBody {
280    pub provider: LanguageModelProvider,
281    pub model: String,
282    pub provider_request: serde_json::Value,
283}
284
285#[derive(Serialize, Deserialize)]
286pub struct CountTokensResponse {
287    pub tokens: usize,
288}
289
290#[derive(Debug, PartialEq, Eq, Hash, Clone, Serialize, Deserialize)]
291pub struct LanguageModelId(pub Arc<str>);
292
293impl std::fmt::Display for LanguageModelId {
294    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
295        write!(f, "{}", self.0)
296    }
297}
298
299#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
300pub struct LanguageModel {
301    pub provider: LanguageModelProvider,
302    pub id: LanguageModelId,
303    pub display_name: String,
304    pub max_token_count: usize,
305    pub max_token_count_in_max_mode: Option<usize>,
306    pub max_output_tokens: usize,
307    pub supports_tools: bool,
308    pub supports_images: bool,
309    pub supports_thinking: bool,
310    pub supports_max_mode: bool,
311}
312
313#[derive(Debug, Serialize, Deserialize)]
314pub struct ListModelsResponse {
315    pub models: Vec<LanguageModel>,
316    pub default_model: LanguageModelId,
317    pub default_fast_model: LanguageModelId,
318    pub recommended_models: Vec<LanguageModelId>,
319}
320
321#[derive(Debug, Serialize, Deserialize)]
322pub struct GetSubscriptionResponse {
323    pub plan: Plan,
324    pub usage: Option<CurrentUsage>,
325}
326
327#[derive(Debug, PartialEq, Serialize, Deserialize)]
328pub struct CurrentUsage {
329    pub model_requests: UsageData,
330    pub edit_predictions: UsageData,
331}
332
333#[derive(Debug, PartialEq, Serialize, Deserialize)]
334pub struct UsageData {
335    pub used: u32,
336    pub limit: UsageLimit,
337}
338
339#[cfg(test)]
340mod tests {
341    use pretty_assertions::assert_eq;
342    use serde_json::json;
343
344    use super::*;
345
346    #[test]
347    fn test_plan_deserialize_snake_case() {
348        let plan = serde_json::from_value::<Plan>(json!("zed_free")).unwrap();
349        assert_eq!(plan, Plan::ZedFree);
350
351        let plan = serde_json::from_value::<Plan>(json!("zed_pro")).unwrap();
352        assert_eq!(plan, Plan::ZedPro);
353
354        let plan = serde_json::from_value::<Plan>(json!("zed_pro_trial")).unwrap();
355        assert_eq!(plan, Plan::ZedProTrial);
356    }
357
358    #[test]
359    fn test_plan_deserialize_aliases() {
360        let plan = serde_json::from_value::<Plan>(json!("Free")).unwrap();
361        assert_eq!(plan, Plan::ZedFree);
362
363        let plan = serde_json::from_value::<Plan>(json!("ZedPro")).unwrap();
364        assert_eq!(plan, Plan::ZedPro);
365
366        let plan = serde_json::from_value::<Plan>(json!("ZedProTrial")).unwrap();
367        assert_eq!(plan, Plan::ZedProTrial);
368    }
369
370    #[test]
371    fn test_usage_limit_from_str() {
372        let limit = UsageLimit::from_str("unlimited").unwrap();
373        assert!(matches!(limit, UsageLimit::Unlimited));
374
375        let limit = UsageLimit::from_str(&0.to_string()).unwrap();
376        assert!(matches!(limit, UsageLimit::Limited(0)));
377
378        let limit = UsageLimit::from_str(&50.to_string()).unwrap();
379        assert!(matches!(limit, UsageLimit::Limited(50)));
380
381        for value in ["not_a_number", "50xyz"] {
382            let limit = UsageLimit::from_str(value);
383            assert!(limit.is_err());
384        }
385    }
386}