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 20/// The name of the header used to indicate when a request failed due to an outdated LLM token.
 21///
 22/// A token is considered "outdated" when we can't parse the claims (e.g., after adding a new required claim).
 23///
 24/// This is distinct from [`EXPIRED_LLM_TOKEN_HEADER_NAME`] which indicates the token's time-based validity has passed.
 25/// An outdated token means the token's structure is incompatible with the current server expectations.
 26pub const OUTDATED_LLM_TOKEN_HEADER_NAME: &str = "x-zed-outdated-token";
 27
 28/// The name of the header used to indicate the usage limit for edit predictions.
 29pub const EDIT_PREDICTIONS_USAGE_LIMIT_HEADER_NAME: &str = "x-zed-edit-predictions-usage-limit";
 30
 31/// The name of the header used to indicate the usage amount for edit predictions.
 32pub const EDIT_PREDICTIONS_USAGE_AMOUNT_HEADER_NAME: &str = "x-zed-edit-predictions-usage-amount";
 33
 34pub const EDIT_PREDICTIONS_RESOURCE_HEADER_VALUE: &str = "edit_predictions";
 35
 36/// The name of the header used to indicate the minimum required Zed version.
 37///
 38/// This can be used to force a Zed upgrade in order to continue communicating
 39/// with the LLM service.
 40pub const MINIMUM_REQUIRED_VERSION_HEADER_NAME: &str = "x-zed-minimum-required-version";
 41
 42/// The name of the header used by the client to indicate to the server that it supports receiving status messages.
 43pub const CLIENT_SUPPORTS_STATUS_MESSAGES_HEADER_NAME: &str =
 44    "x-zed-client-supports-status-messages";
 45
 46/// The name of the header used by the client to indicate to the server that it supports receiving a "stream_ended" request completion status.
 47pub const CLIENT_SUPPORTS_STATUS_STREAM_ENDED_HEADER_NAME: &str =
 48    "x-zed-client-supports-stream-ended-request-completion-status";
 49
 50/// The name of the header used by the server to indicate to the client that it supports sending status messages.
 51pub const SERVER_SUPPORTS_STATUS_MESSAGES_HEADER_NAME: &str =
 52    "x-zed-server-supports-status-messages";
 53
 54/// The name of the header used by the client to indicate that it supports receiving xAI models.
 55pub const CLIENT_SUPPORTS_X_AI_HEADER_NAME: &str = "x-zed-client-supports-x-ai";
 56
 57/// The maximum number of edit predictions that can be rejected per request.
 58pub const MAX_EDIT_PREDICTION_REJECTIONS_PER_REQUEST: usize = 100;
 59
 60#[derive(Debug, PartialEq, Clone, Copy, Serialize, Deserialize)]
 61#[serde(rename_all = "snake_case")]
 62pub enum UsageLimit {
 63    Limited(i32),
 64    Unlimited,
 65}
 66
 67impl FromStr for UsageLimit {
 68    type Err = anyhow::Error;
 69
 70    fn from_str(value: &str) -> Result<Self, Self::Err> {
 71        match value {
 72            "unlimited" => Ok(Self::Unlimited),
 73            limit => limit
 74                .parse::<i32>()
 75                .map(Self::Limited)
 76                .context("failed to parse limit"),
 77        }
 78    }
 79}
 80
 81#[derive(
 82    Debug, PartialEq, Eq, Hash, Clone, Copy, Serialize, Deserialize, EnumString, EnumIter, Display,
 83)]
 84#[serde(rename_all = "snake_case")]
 85#[strum(serialize_all = "snake_case")]
 86pub enum LanguageModelProvider {
 87    Anthropic,
 88    OpenAi,
 89    Google,
 90    XAi,
 91}
 92
 93#[derive(Debug, Clone, Serialize, Deserialize)]
 94pub struct PredictEditsBody {
 95    #[serde(skip_serializing_if = "Option::is_none", default)]
 96    pub outline: Option<String>,
 97    pub input_events: String,
 98    pub input_excerpt: String,
 99    #[serde(skip_serializing_if = "Option::is_none", default)]
100    pub speculated_output: Option<String>,
101    /// Whether the user provided consent for sampling this interaction.
102    #[serde(default, alias = "data_collection_permission")]
103    pub can_collect_data: bool,
104    #[serde(skip_serializing_if = "Option::is_none", default)]
105    pub diagnostic_groups: Option<Vec<(String, serde_json::Value)>>,
106    /// Info about the git repository state, only present when can_collect_data is true.
107    #[serde(skip_serializing_if = "Option::is_none", default)]
108    pub git_info: Option<PredictEditsGitInfo>,
109    /// The trigger for this request.
110    #[serde(default)]
111    pub trigger: PredictEditsRequestTrigger,
112}
113
114#[derive(Default, Debug, Clone, Copy, Serialize, Deserialize)]
115pub enum PredictEditsRequestTrigger {
116    Testing,
117    Diagnostics,
118    Cli,
119    #[default]
120    Other,
121}
122
123#[derive(Debug, Clone, Serialize, Deserialize)]
124pub struct PredictEditsGitInfo {
125    /// SHA of git HEAD commit at time of prediction.
126    #[serde(skip_serializing_if = "Option::is_none", default)]
127    pub head_sha: Option<String>,
128    /// URL of the remote called `origin`.
129    #[serde(skip_serializing_if = "Option::is_none", default)]
130    pub remote_origin_url: Option<String>,
131    /// URL of the remote called `upstream`.
132    #[serde(skip_serializing_if = "Option::is_none", default)]
133    pub remote_upstream_url: Option<String>,
134}
135
136#[derive(Debug, Clone, Serialize, Deserialize)]
137pub struct PredictEditsResponse {
138    pub request_id: String,
139    pub output_excerpt: String,
140}
141
142#[derive(Debug, Clone, Serialize, Deserialize)]
143pub struct AcceptEditPredictionBody {
144    pub request_id: String,
145    #[serde(default, skip_serializing_if = "Option::is_none")]
146    pub model_version: Option<String>,
147    #[serde(default, skip_serializing_if = "Option::is_none")]
148    pub e2e_latency_ms: Option<u128>,
149}
150
151#[derive(Debug, Clone, Deserialize)]
152pub struct RejectEditPredictionsBody {
153    pub rejections: Vec<EditPredictionRejection>,
154}
155
156#[derive(Debug, Clone, Serialize)]
157pub struct RejectEditPredictionsBodyRef<'a> {
158    pub rejections: &'a [EditPredictionRejection],
159}
160
161#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
162pub struct EditPredictionRejection {
163    pub request_id: String,
164    #[serde(default)]
165    pub reason: EditPredictionRejectReason,
166    pub was_shown: bool,
167    #[serde(default, skip_serializing_if = "Option::is_none")]
168    pub model_version: Option<String>,
169    #[serde(default, skip_serializing_if = "Option::is_none")]
170    pub e2e_latency_ms: Option<u128>,
171}
172
173#[derive(Default, Debug, Clone, Copy, Serialize, Deserialize, PartialEq)]
174pub enum EditPredictionRejectReason {
175    /// New requests were triggered before this one completed
176    Canceled,
177    /// No edits returned
178    Empty,
179    /// Edits returned, but none remained after interpolation
180    InterpolatedEmpty,
181    /// The new prediction was preferred over the current one
182    Replaced,
183    /// The current prediction was preferred over the new one
184    CurrentPreferred,
185    /// The current prediction was discarded
186    #[default]
187    Discarded,
188    /// The current prediction was explicitly rejected by the user
189    Rejected,
190}
191
192#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Clone, Copy, Serialize, Deserialize)]
193#[serde(rename_all = "snake_case")]
194pub enum CompletionIntent {
195    UserPrompt,
196    ToolResults,
197    ThreadSummarization,
198    ThreadContextSummarization,
199    CreateFile,
200    EditFile,
201    InlineAssist,
202    TerminalInlineAssist,
203    GenerateGitCommitMessage,
204}
205
206#[derive(Debug, Serialize, Deserialize)]
207pub struct CompletionBody {
208    #[serde(skip_serializing_if = "Option::is_none", default)]
209    pub thread_id: Option<String>,
210    #[serde(skip_serializing_if = "Option::is_none", default)]
211    pub prompt_id: Option<String>,
212    #[serde(skip_serializing_if = "Option::is_none", default)]
213    pub intent: Option<CompletionIntent>,
214    pub provider: LanguageModelProvider,
215    pub model: String,
216    pub provider_request: serde_json::Value,
217}
218
219#[derive(Debug, PartialEq, Clone, Serialize, Deserialize)]
220#[serde(rename_all = "snake_case")]
221pub enum CompletionRequestStatus {
222    Queued {
223        position: usize,
224    },
225    Started,
226    Failed {
227        code: String,
228        message: String,
229        request_id: Uuid,
230        /// Retry duration in seconds.
231        retry_after: Option<f64>,
232    },
233    /// The cloud sends a StreamEnded message when the stream from the LLM provider finishes.
234    StreamEnded,
235    #[serde(other)]
236    Unknown,
237}
238
239#[derive(Serialize, Deserialize)]
240#[serde(rename_all = "snake_case")]
241pub enum CompletionEvent<T> {
242    Status(CompletionRequestStatus),
243    Event(T),
244}
245
246impl<T> CompletionEvent<T> {
247    pub fn into_status(self) -> Option<CompletionRequestStatus> {
248        match self {
249            Self::Status(status) => Some(status),
250            Self::Event(_) => None,
251        }
252    }
253
254    pub fn into_event(self) -> Option<T> {
255        match self {
256            Self::Event(event) => Some(event),
257            Self::Status(_) => None,
258        }
259    }
260}
261
262#[derive(Serialize, Deserialize)]
263pub struct WebSearchBody {
264    pub query: String,
265}
266
267#[derive(Debug, Serialize, Deserialize, Clone)]
268pub struct WebSearchResponse {
269    pub results: Vec<WebSearchResult>,
270}
271
272#[derive(Debug, Serialize, Deserialize, Clone)]
273pub struct WebSearchResult {
274    pub title: String,
275    pub url: String,
276    pub text: String,
277}
278
279#[derive(Serialize, Deserialize)]
280pub struct CountTokensBody {
281    pub provider: LanguageModelProvider,
282    pub model: String,
283    pub provider_request: serde_json::Value,
284}
285
286#[derive(Serialize, Deserialize)]
287pub struct CountTokensResponse {
288    pub tokens: usize,
289}
290
291#[derive(Debug, PartialEq, Eq, Hash, Clone, Serialize, Deserialize)]
292pub struct LanguageModelId(pub Arc<str>);
293
294impl std::fmt::Display for LanguageModelId {
295    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
296        write!(f, "{}", self.0)
297    }
298}
299
300#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
301pub struct LanguageModel {
302    pub provider: LanguageModelProvider,
303    pub id: LanguageModelId,
304    pub display_name: String,
305    #[serde(default)]
306    pub is_latest: bool,
307    pub max_token_count: usize,
308    pub max_token_count_in_max_mode: Option<usize>,
309    pub max_output_tokens: usize,
310    pub supports_tools: bool,
311    pub supports_images: bool,
312    pub supports_thinking: bool,
313    #[serde(default)]
314    pub supports_fast_mode: bool,
315    pub supported_effort_levels: Vec<SupportedEffortLevel>,
316    #[serde(default)]
317    pub supports_streaming_tools: bool,
318    /// Only used by OpenAI and xAI.
319    #[serde(default)]
320    pub supports_parallel_tool_calls: bool,
321}
322
323#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
324pub struct SupportedEffortLevel {
325    pub name: Arc<str>,
326    pub value: Arc<str>,
327    #[serde(default, skip_serializing_if = "Option::is_none")]
328    pub is_default: Option<bool>,
329}
330
331#[derive(Debug, Serialize, Deserialize)]
332pub struct ListModelsResponse {
333    pub models: Vec<LanguageModel>,
334    pub default_model: Option<LanguageModelId>,
335    pub default_fast_model: Option<LanguageModelId>,
336    pub recommended_models: Vec<LanguageModelId>,
337}
338
339#[derive(Debug, PartialEq, Serialize, Deserialize)]
340pub struct CurrentUsage {
341    pub edit_predictions: UsageData,
342}
343
344#[derive(Debug, PartialEq, Serialize, Deserialize)]
345pub struct UsageData {
346    pub used: u32,
347    pub limit: UsageLimit,
348}
349
350#[cfg(test)]
351mod tests {
352    use super::*;
353
354    #[test]
355    fn test_usage_limit_from_str() {
356        let limit = UsageLimit::from_str("unlimited").unwrap();
357        assert!(matches!(limit, UsageLimit::Unlimited));
358
359        let limit = UsageLimit::from_str(&0.to_string()).unwrap();
360        assert!(matches!(limit, UsageLimit::Limited(0)));
361
362        let limit = UsageLimit::from_str(&50.to_string()).unwrap();
363        assert!(matches!(limit, UsageLimit::Limited(50)));
364
365        for value in ["not_a_number", "50xyz"] {
366            let limit = UsageLimit::from_str(value);
367            assert!(limit.is_err());
368        }
369    }
370}