cloud_llm_client.rs

  1pub mod predict_edits_v3;
  2
  3use std::str::FromStr;
  4use std::sync::Arc;
  5
  6use anyhow::Context as _;
  7use serde::{Deserialize, Serialize};
  8use strum::{Display, EnumIter, EnumString};
  9use uuid::Uuid;
 10
 11/// The name of the header used to indicate which version of Zed the client is running.
 12pub const ZED_VERSION_HEADER_NAME: &str = "x-zed-version";
 13
 14/// The name of the header used to indicate when a request failed due to an
 15/// expired LLM token.
 16///
 17/// The client may use this as a signal to refresh the token.
 18pub const EXPIRED_LLM_TOKEN_HEADER_NAME: &str = "x-zed-expired-token";
 19
 20/// The name of the header used to indicate what plan the user is currently on.
 21pub const CURRENT_PLAN_HEADER_NAME: &str = "x-zed-plan";
 22
 23/// The name of the header used to indicate the usage limit for model requests.
 24pub const MODEL_REQUESTS_USAGE_LIMIT_HEADER_NAME: &str = "x-zed-model-requests-usage-limit";
 25
 26/// The name of the header used to indicate the usage amount for model requests.
 27pub const MODEL_REQUESTS_USAGE_AMOUNT_HEADER_NAME: &str = "x-zed-model-requests-usage-amount";
 28
 29/// The name of the header used to indicate the usage limit for edit predictions.
 30pub const EDIT_PREDICTIONS_USAGE_LIMIT_HEADER_NAME: &str = "x-zed-edit-predictions-usage-limit";
 31
 32/// The name of the header used to indicate the usage amount for edit predictions.
 33pub const EDIT_PREDICTIONS_USAGE_AMOUNT_HEADER_NAME: &str = "x-zed-edit-predictions-usage-amount";
 34
 35/// The name of the header used to indicate the resource for which the subscription limit has been reached.
 36pub const SUBSCRIPTION_LIMIT_RESOURCE_HEADER_NAME: &str = "x-zed-subscription-limit-resource";
 37
 38pub const MODEL_REQUESTS_RESOURCE_HEADER_VALUE: &str = "model_requests";
 39pub const EDIT_PREDICTIONS_RESOURCE_HEADER_VALUE: &str = "edit_predictions";
 40
 41/// The name of the header used to indicate that the maximum number of consecutive tool uses has been reached.
 42pub const TOOL_USE_LIMIT_REACHED_HEADER_NAME: &str = "x-zed-tool-use-limit-reached";
 43
 44/// The name of the header used to indicate the minimum required Zed version.
 45///
 46/// This can be used to force a Zed upgrade in order to continue communicating
 47/// with the LLM service.
 48pub const MINIMUM_REQUIRED_VERSION_HEADER_NAME: &str = "x-zed-minimum-required-version";
 49
 50/// The name of the header used by the client to indicate to the server that it supports receiving status messages.
 51pub const CLIENT_SUPPORTS_STATUS_MESSAGES_HEADER_NAME: &str =
 52    "x-zed-client-supports-status-messages";
 53
 54/// The name of the header used by the server to indicate to the client that it supports sending status messages.
 55pub const SERVER_SUPPORTS_STATUS_MESSAGES_HEADER_NAME: &str =
 56    "x-zed-server-supports-status-messages";
 57
 58/// The name of the header used by the client to indicate that it supports receiving xAI models.
 59pub const CLIENT_SUPPORTS_X_AI_HEADER_NAME: &str = "x-zed-client-supports-x-ai";
 60
 61/// The maximum number of edit predictions that can be rejected per request.
 62pub const MAX_EDIT_PREDICTION_REJECTIONS_PER_REQUEST: usize = 100;
 63
 64#[derive(Debug, PartialEq, Clone, Copy, Serialize, Deserialize)]
 65#[serde(rename_all = "snake_case")]
 66pub enum UsageLimit {
 67    Limited(i32),
 68    Unlimited,
 69}
 70
 71impl FromStr for UsageLimit {
 72    type Err = anyhow::Error;
 73
 74    fn from_str(value: &str) -> Result<Self, Self::Err> {
 75        match value {
 76            "unlimited" => Ok(Self::Unlimited),
 77            limit => limit
 78                .parse::<i32>()
 79                .map(Self::Limited)
 80                .context("failed to parse limit"),
 81        }
 82    }
 83}
 84
 85#[derive(Debug, Clone, Copy, PartialEq)]
 86pub enum Plan {
 87    V1(PlanV1),
 88    V2(PlanV2),
 89}
 90
 91impl Plan {
 92    pub fn is_v2(&self) -> bool {
 93        matches!(self, Self::V2(_))
 94    }
 95}
 96
 97#[derive(Debug, Clone, Copy, Default, PartialEq, Serialize, Deserialize)]
 98#[serde(rename_all = "snake_case")]
 99pub enum PlanV1 {
100    #[default]
101    #[serde(alias = "Free")]
102    ZedFree,
103    #[serde(alias = "ZedPro")]
104    ZedPro,
105    #[serde(alias = "ZedProTrial")]
106    ZedProTrial,
107}
108
109impl FromStr for PlanV1 {
110    type Err = anyhow::Error;
111
112    fn from_str(value: &str) -> Result<Self, Self::Err> {
113        match value {
114            "zed_free" => Ok(Self::ZedFree),
115            "zed_pro" => Ok(Self::ZedPro),
116            "zed_pro_trial" => Ok(Self::ZedProTrial),
117            plan => Err(anyhow::anyhow!("invalid plan: {plan:?}")),
118        }
119    }
120}
121
122#[derive(Debug, Clone, Copy, Default, PartialEq, Serialize, Deserialize)]
123#[serde(rename_all = "snake_case")]
124pub enum PlanV2 {
125    #[default]
126    ZedFree,
127    ZedPro,
128    ZedProTrial,
129}
130
131impl FromStr for PlanV2 {
132    type Err = anyhow::Error;
133
134    fn from_str(value: &str) -> Result<Self, Self::Err> {
135        match value {
136            "zed_free" => Ok(Self::ZedFree),
137            "zed_pro" => Ok(Self::ZedPro),
138            "zed_pro_trial" => Ok(Self::ZedProTrial),
139            plan => Err(anyhow::anyhow!("invalid plan: {plan:?}")),
140        }
141    }
142}
143
144#[derive(
145    Debug, PartialEq, Eq, Hash, Clone, Copy, Serialize, Deserialize, EnumString, EnumIter, Display,
146)]
147#[serde(rename_all = "snake_case")]
148#[strum(serialize_all = "snake_case")]
149pub enum LanguageModelProvider {
150    Anthropic,
151    OpenAi,
152    Google,
153    XAi,
154}
155
156#[derive(Debug, Clone, Serialize, Deserialize)]
157pub struct PredictEditsBody {
158    #[serde(skip_serializing_if = "Option::is_none", default)]
159    pub outline: Option<String>,
160    pub input_events: String,
161    pub input_excerpt: String,
162    #[serde(skip_serializing_if = "Option::is_none", default)]
163    pub speculated_output: Option<String>,
164    /// Whether the user provided consent for sampling this interaction.
165    #[serde(default, alias = "data_collection_permission")]
166    pub can_collect_data: bool,
167    #[serde(skip_serializing_if = "Option::is_none", default)]
168    pub diagnostic_groups: Option<Vec<(String, serde_json::Value)>>,
169    /// Info about the git repository state, only present when can_collect_data is true.
170    #[serde(skip_serializing_if = "Option::is_none", default)]
171    pub git_info: Option<PredictEditsGitInfo>,
172    /// The trigger for this request.
173    #[serde(default)]
174    pub trigger: PredictEditsRequestTrigger,
175}
176
177#[derive(Default, Debug, Clone, Copy, Serialize, Deserialize)]
178pub enum PredictEditsRequestTrigger {
179    Diagnostics,
180    Cli,
181    #[default]
182    Other,
183}
184
185#[derive(Debug, Clone, Serialize, Deserialize)]
186pub struct PredictEditsGitInfo {
187    /// SHA of git HEAD commit at time of prediction.
188    #[serde(skip_serializing_if = "Option::is_none", default)]
189    pub head_sha: Option<String>,
190    /// URL of the remote called `origin`.
191    #[serde(skip_serializing_if = "Option::is_none", default)]
192    pub remote_origin_url: Option<String>,
193    /// URL of the remote called `upstream`.
194    #[serde(skip_serializing_if = "Option::is_none", default)]
195    pub remote_upstream_url: Option<String>,
196}
197
198#[derive(Debug, Clone, Serialize, Deserialize)]
199pub struct PredictEditsResponse {
200    pub request_id: String,
201    pub output_excerpt: String,
202}
203
204#[derive(Debug, Clone, Serialize, Deserialize)]
205pub struct AcceptEditPredictionBody {
206    pub request_id: String,
207}
208
209#[derive(Debug, Clone, Deserialize)]
210pub struct RejectEditPredictionsBody {
211    pub rejections: Vec<EditPredictionRejection>,
212}
213
214#[derive(Debug, Clone, Serialize)]
215pub struct RejectEditPredictionsBodyRef<'a> {
216    pub rejections: &'a [EditPredictionRejection],
217}
218
219#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
220pub struct EditPredictionRejection {
221    pub request_id: String,
222    #[serde(default)]
223    pub reason: EditPredictionRejectReason,
224    pub was_shown: bool,
225}
226
227#[derive(Default, Debug, Clone, Copy, Serialize, Deserialize, PartialEq)]
228pub enum EditPredictionRejectReason {
229    /// New requests were triggered before this one completed
230    Canceled,
231    /// No edits returned
232    Empty,
233    /// Edits returned, but none remained after interpolation
234    InterpolatedEmpty,
235    /// The new prediction was preferred over the current one
236    Replaced,
237    /// The current prediction was preferred over the new one
238    CurrentPreferred,
239    /// The current prediction was discarded
240    #[default]
241    Discarded,
242}
243
244#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Clone, Copy, Serialize, Deserialize)]
245#[serde(rename_all = "snake_case")]
246pub enum CompletionMode {
247    Normal,
248    Max,
249}
250
251#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Clone, Copy, Serialize, Deserialize)]
252#[serde(rename_all = "snake_case")]
253pub enum CompletionIntent {
254    UserPrompt,
255    ToolResults,
256    ThreadSummarization,
257    ThreadContextSummarization,
258    CreateFile,
259    EditFile,
260    InlineAssist,
261    TerminalInlineAssist,
262    GenerateGitCommitMessage,
263}
264
265#[derive(Debug, Serialize, Deserialize)]
266pub struct CompletionBody {
267    #[serde(skip_serializing_if = "Option::is_none", default)]
268    pub thread_id: Option<String>,
269    #[serde(skip_serializing_if = "Option::is_none", default)]
270    pub prompt_id: Option<String>,
271    #[serde(skip_serializing_if = "Option::is_none", default)]
272    pub intent: Option<CompletionIntent>,
273    #[serde(skip_serializing_if = "Option::is_none", default)]
274    pub mode: Option<CompletionMode>,
275    pub provider: LanguageModelProvider,
276    pub model: String,
277    pub provider_request: serde_json::Value,
278}
279
280#[derive(Debug, PartialEq, Clone, Serialize, Deserialize)]
281#[serde(rename_all = "snake_case")]
282pub enum CompletionRequestStatus {
283    Queued {
284        position: usize,
285    },
286    Started,
287    Failed {
288        code: String,
289        message: String,
290        request_id: Uuid,
291        /// Retry duration in seconds.
292        retry_after: Option<f64>,
293    },
294    UsageUpdated {
295        amount: usize,
296        limit: UsageLimit,
297    },
298    ToolUseLimitReached,
299}
300
301#[derive(Serialize, Deserialize)]
302#[serde(rename_all = "snake_case")]
303pub enum CompletionEvent<T> {
304    Status(CompletionRequestStatus),
305    Event(T),
306}
307
308impl<T> CompletionEvent<T> {
309    pub fn into_status(self) -> Option<CompletionRequestStatus> {
310        match self {
311            Self::Status(status) => Some(status),
312            Self::Event(_) => None,
313        }
314    }
315
316    pub fn into_event(self) -> Option<T> {
317        match self {
318            Self::Event(event) => Some(event),
319            Self::Status(_) => None,
320        }
321    }
322}
323
324#[derive(Serialize, Deserialize)]
325pub struct WebSearchBody {
326    pub query: String,
327}
328
329#[derive(Debug, Serialize, Deserialize, Clone)]
330pub struct WebSearchResponse {
331    pub results: Vec<WebSearchResult>,
332}
333
334#[derive(Debug, Serialize, Deserialize, Clone)]
335pub struct WebSearchResult {
336    pub title: String,
337    pub url: String,
338    pub text: String,
339}
340
341#[derive(Serialize, Deserialize)]
342pub struct CountTokensBody {
343    pub provider: LanguageModelProvider,
344    pub model: String,
345    pub provider_request: serde_json::Value,
346}
347
348#[derive(Serialize, Deserialize)]
349pub struct CountTokensResponse {
350    pub tokens: usize,
351}
352
353#[derive(Debug, PartialEq, Eq, Hash, Clone, Serialize, Deserialize)]
354pub struct LanguageModelId(pub Arc<str>);
355
356impl std::fmt::Display for LanguageModelId {
357    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
358        write!(f, "{}", self.0)
359    }
360}
361
362#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
363pub struct LanguageModel {
364    pub provider: LanguageModelProvider,
365    pub id: LanguageModelId,
366    pub display_name: String,
367    pub max_token_count: usize,
368    pub max_token_count_in_max_mode: Option<usize>,
369    pub max_output_tokens: usize,
370    pub supports_tools: bool,
371    pub supports_images: bool,
372    pub supports_thinking: bool,
373    pub supports_max_mode: bool,
374    // only used by OpenAI and xAI
375    #[serde(default)]
376    pub supports_parallel_tool_calls: bool,
377}
378
379#[derive(Debug, Serialize, Deserialize)]
380pub struct ListModelsResponse {
381    pub models: Vec<LanguageModel>,
382    pub default_model: Option<LanguageModelId>,
383    pub default_fast_model: Option<LanguageModelId>,
384    pub recommended_models: Vec<LanguageModelId>,
385}
386
387#[derive(Debug, Serialize, Deserialize)]
388pub struct GetSubscriptionResponse {
389    pub plan: PlanV1,
390    pub usage: Option<CurrentUsage>,
391}
392
393#[derive(Debug, PartialEq, Serialize, Deserialize)]
394pub struct CurrentUsage {
395    pub model_requests: UsageData,
396    pub edit_predictions: UsageData,
397}
398
399#[derive(Debug, PartialEq, Serialize, Deserialize)]
400pub struct UsageData {
401    pub used: u32,
402    pub limit: UsageLimit,
403}
404
405#[cfg(test)]
406mod tests {
407    use pretty_assertions::assert_eq;
408    use serde_json::json;
409
410    use super::*;
411
412    #[test]
413    fn test_plan_v1_deserialize_snake_case() {
414        let plan = serde_json::from_value::<PlanV1>(json!("zed_free")).unwrap();
415        assert_eq!(plan, PlanV1::ZedFree);
416
417        let plan = serde_json::from_value::<PlanV1>(json!("zed_pro")).unwrap();
418        assert_eq!(plan, PlanV1::ZedPro);
419
420        let plan = serde_json::from_value::<PlanV1>(json!("zed_pro_trial")).unwrap();
421        assert_eq!(plan, PlanV1::ZedProTrial);
422    }
423
424    #[test]
425    fn test_plan_v1_deserialize_aliases() {
426        let plan = serde_json::from_value::<PlanV1>(json!("Free")).unwrap();
427        assert_eq!(plan, PlanV1::ZedFree);
428
429        let plan = serde_json::from_value::<PlanV1>(json!("ZedPro")).unwrap();
430        assert_eq!(plan, PlanV1::ZedPro);
431
432        let plan = serde_json::from_value::<PlanV1>(json!("ZedProTrial")).unwrap();
433        assert_eq!(plan, PlanV1::ZedProTrial);
434    }
435
436    #[test]
437    fn test_plan_v2_deserialize_snake_case() {
438        let plan = serde_json::from_value::<PlanV2>(json!("zed_free")).unwrap();
439        assert_eq!(plan, PlanV2::ZedFree);
440
441        let plan = serde_json::from_value::<PlanV2>(json!("zed_pro")).unwrap();
442        assert_eq!(plan, PlanV2::ZedPro);
443
444        let plan = serde_json::from_value::<PlanV2>(json!("zed_pro_trial")).unwrap();
445        assert_eq!(plan, PlanV2::ZedProTrial);
446    }
447
448    #[test]
449    fn test_usage_limit_from_str() {
450        let limit = UsageLimit::from_str("unlimited").unwrap();
451        assert!(matches!(limit, UsageLimit::Unlimited));
452
453        let limit = UsageLimit::from_str(&0.to_string()).unwrap();
454        assert!(matches!(limit, UsageLimit::Limited(0)));
455
456        let limit = UsageLimit::from_str(&50.to_string()).unwrap();
457        assert!(matches!(limit, UsageLimit::Limited(50)));
458
459        for value in ["not_a_number", "50xyz"] {
460            let limit = UsageLimit::from_str(value);
461            assert!(limit.is_err());
462        }
463    }
464}