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    /// Path within the repository that contains the input excerpt.
160    #[serde(skip_serializing_if = "Option::is_none", default)]
161    pub input_path: Option<String>,
162    /// SHA of git HEAD commit at time of prediction.
163    #[serde(skip_serializing_if = "Option::is_none", default)]
164    pub head_sha: Option<String>,
165    /// URL of the remote called `origin`.
166    #[serde(skip_serializing_if = "Option::is_none", default)]
167    pub remote_origin_url: Option<String>,
168    /// URL of the remote called `upstream`.
169    #[serde(skip_serializing_if = "Option::is_none", default)]
170    pub remote_upstream_url: Option<String>,
171    /// Recently active files that may be within this repository.
172    #[serde(skip_serializing_if = "Option::is_none", default)]
173    pub recent_files: Option<Vec<PredictEditsRecentFile>>,
174}
175
176#[derive(Debug, Clone, Serialize, Deserialize)]
177pub struct PredictEditsRecentFile {
178    /// Path to a file within the repository.
179    pub path: String,
180    /// Milliseconds between the editor for this file being active and the request time.
181    pub active_to_now_ms: u32,
182}
183
184#[derive(Debug, Clone, Serialize, Deserialize)]
185pub struct PredictEditsResponse {
186    pub request_id: Uuid,
187    pub output_excerpt: String,
188}
189
190#[derive(Debug, Clone, Serialize, Deserialize)]
191pub struct AcceptEditPredictionBody {
192    pub request_id: Uuid,
193}
194
195#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Clone, Copy, Serialize, Deserialize)]
196#[serde(rename_all = "snake_case")]
197pub enum CompletionMode {
198    Normal,
199    Max,
200}
201
202#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Clone, Copy, Serialize, Deserialize)]
203#[serde(rename_all = "snake_case")]
204pub enum CompletionIntent {
205    UserPrompt,
206    ToolResults,
207    ThreadSummarization,
208    ThreadContextSummarization,
209    CreateFile,
210    EditFile,
211    InlineAssist,
212    TerminalInlineAssist,
213    GenerateGitCommitMessage,
214}
215
216#[derive(Debug, Serialize, Deserialize)]
217pub struct CompletionBody {
218    #[serde(skip_serializing_if = "Option::is_none", default)]
219    pub thread_id: Option<String>,
220    #[serde(skip_serializing_if = "Option::is_none", default)]
221    pub prompt_id: Option<String>,
222    #[serde(skip_serializing_if = "Option::is_none", default)]
223    pub intent: Option<CompletionIntent>,
224    #[serde(skip_serializing_if = "Option::is_none", default)]
225    pub mode: Option<CompletionMode>,
226    pub provider: LanguageModelProvider,
227    pub model: String,
228    pub provider_request: serde_json::Value,
229}
230
231#[derive(Debug, PartialEq, Clone, Serialize, Deserialize)]
232#[serde(rename_all = "snake_case")]
233pub enum CompletionRequestStatus {
234    Queued {
235        position: usize,
236    },
237    Started,
238    Failed {
239        code: String,
240        message: String,
241        request_id: Uuid,
242        /// Retry duration in seconds.
243        retry_after: Option<f64>,
244    },
245    UsageUpdated {
246        amount: usize,
247        limit: UsageLimit,
248    },
249    ToolUseLimitReached,
250}
251
252#[derive(Serialize, Deserialize)]
253#[serde(rename_all = "snake_case")]
254pub enum CompletionEvent<T> {
255    Status(CompletionRequestStatus),
256    Event(T),
257}
258
259impl<T> CompletionEvent<T> {
260    pub fn into_status(self) -> Option<CompletionRequestStatus> {
261        match self {
262            Self::Status(status) => Some(status),
263            Self::Event(_) => None,
264        }
265    }
266
267    pub fn into_event(self) -> Option<T> {
268        match self {
269            Self::Event(event) => Some(event),
270            Self::Status(_) => None,
271        }
272    }
273}
274
275#[derive(Serialize, Deserialize)]
276pub struct WebSearchBody {
277    pub query: String,
278}
279
280#[derive(Debug, Serialize, Deserialize, Clone)]
281pub struct WebSearchResponse {
282    pub results: Vec<WebSearchResult>,
283}
284
285#[derive(Debug, Serialize, Deserialize, Clone)]
286pub struct WebSearchResult {
287    pub title: String,
288    pub url: String,
289    pub text: String,
290}
291
292#[derive(Serialize, Deserialize)]
293pub struct CountTokensBody {
294    pub provider: LanguageModelProvider,
295    pub model: String,
296    pub provider_request: serde_json::Value,
297}
298
299#[derive(Serialize, Deserialize)]
300pub struct CountTokensResponse {
301    pub tokens: usize,
302}
303
304#[derive(Debug, PartialEq, Eq, Hash, Clone, Serialize, Deserialize)]
305pub struct LanguageModelId(pub Arc<str>);
306
307impl std::fmt::Display for LanguageModelId {
308    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
309        write!(f, "{}", self.0)
310    }
311}
312
313#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
314pub struct LanguageModel {
315    pub provider: LanguageModelProvider,
316    pub id: LanguageModelId,
317    pub display_name: String,
318    pub max_token_count: usize,
319    pub max_token_count_in_max_mode: Option<usize>,
320    pub max_output_tokens: usize,
321    pub supports_tools: bool,
322    pub supports_images: bool,
323    pub supports_thinking: bool,
324    pub supports_max_mode: bool,
325}
326
327#[derive(Debug, Serialize, Deserialize)]
328pub struct ListModelsResponse {
329    pub models: Vec<LanguageModel>,
330    pub default_model: LanguageModelId,
331    pub default_fast_model: LanguageModelId,
332    pub recommended_models: Vec<LanguageModelId>,
333}
334
335#[derive(Debug, Serialize, Deserialize)]
336pub struct GetSubscriptionResponse {
337    pub plan: Plan,
338    pub usage: Option<CurrentUsage>,
339}
340
341#[derive(Debug, PartialEq, Serialize, Deserialize)]
342pub struct CurrentUsage {
343    pub model_requests: UsageData,
344    pub edit_predictions: UsageData,
345}
346
347#[derive(Debug, PartialEq, Serialize, Deserialize)]
348pub struct UsageData {
349    pub used: u32,
350    pub limit: UsageLimit,
351}
352
353#[cfg(test)]
354mod tests {
355    use pretty_assertions::assert_eq;
356    use serde_json::json;
357
358    use super::*;
359
360    #[test]
361    fn test_plan_deserialize_snake_case() {
362        let plan = serde_json::from_value::<Plan>(json!("zed_free")).unwrap();
363        assert_eq!(plan, Plan::ZedFree);
364
365        let plan = serde_json::from_value::<Plan>(json!("zed_pro")).unwrap();
366        assert_eq!(plan, Plan::ZedPro);
367
368        let plan = serde_json::from_value::<Plan>(json!("zed_pro_trial")).unwrap();
369        assert_eq!(plan, Plan::ZedProTrial);
370    }
371
372    #[test]
373    fn test_plan_deserialize_aliases() {
374        let plan = serde_json::from_value::<Plan>(json!("Free")).unwrap();
375        assert_eq!(plan, Plan::ZedFree);
376
377        let plan = serde_json::from_value::<Plan>(json!("ZedPro")).unwrap();
378        assert_eq!(plan, Plan::ZedPro);
379
380        let plan = serde_json::from_value::<Plan>(json!("ZedProTrial")).unwrap();
381        assert_eq!(plan, Plan::ZedProTrial);
382    }
383
384    #[test]
385    fn test_usage_limit_from_str() {
386        let limit = UsageLimit::from_str("unlimited").unwrap();
387        assert!(matches!(limit, UsageLimit::Unlimited));
388
389        let limit = UsageLimit::from_str(&0.to_string()).unwrap();
390        assert!(matches!(limit, UsageLimit::Limited(0)));
391
392        let limit = UsageLimit::from_str(&50.to_string()).unwrap();
393        assert!(matches!(limit, UsageLimit::Limited(50)));
394
395        for value in ["not_a_number", "50xyz"] {
396            let limit = UsageLimit::from_str(value);
397            assert!(limit.is_err());
398        }
399    }
400}