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 20/// The name of the header used to indicate when a request failed due to an outdated LLM token.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 22/// A token is considered "outdated" when we can't parse the claims (e.g., after adding a new required claim).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 25/// An outdated token means the token's structure is incompatible with the current server expectations.
 26pub const OUTDATED_LLM_TOKEN_HEADER_NAME: &str = "x-zed-outdated-token";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}
148
149#[derive(Debug, Clone, Deserialize)]
150pub struct RejectEditPredictionsBody {
151    pub rejections: Vec<EditPredictionRejection>,
152}
153
154#[derive(Debug, Clone, Serialize)]
155pub struct RejectEditPredictionsBodyRef<'a> {
156    pub rejections: &'a [EditPredictionRejection],
157}
158
159#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
160pub struct EditPredictionRejection {
161    pub request_id: String,
162    #[serde(default)]
163    pub reason: EditPredictionRejectReason,
164    pub was_shown: bool,
165    #[serde(default, skip_serializing_if = "Option::is_none")]
166    pub model_version: Option<String>,
167}
168
169#[derive(Default, Debug, Clone, Copy, Serialize, Deserialize, PartialEq)]
170pub enum EditPredictionRejectReason {
171    /// New requests were triggered before this one completed
172    Canceled,
173    /// No edits returned
174    Empty,
175    /// Edits returned, but none remained after interpolation
176    InterpolatedEmpty,
177    /// The new prediction was preferred over the current one
178    Replaced,
179    /// The current prediction was preferred over the new one
180    CurrentPreferred,
181    /// The current prediction was discarded
182    #[default]
183    Discarded,
184    /// The current prediction was explicitly rejected by the user
185    Rejected,
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    pub provider: LanguageModelProvider,
211    pub model: String,
212    pub provider_request: serde_json::Value,
213}
214
215#[derive(Debug, PartialEq, Clone, Serialize, Deserialize)]
216#[serde(rename_all = "snake_case")]
217pub enum CompletionRequestStatus {
218    Queued {
219        position: usize,
220    },
221    Started,
222    Failed {
223        code: String,
224        message: String,
225        request_id: Uuid,
226        /// Retry duration in seconds.
227        retry_after: Option<f64>,
228    },
229    /// The cloud sends a StreamEnded message when the stream from the LLM provider finishes.
230    StreamEnded,
231    #[serde(other)]
232    Unknown,
233}
234
235#[derive(Serialize, Deserialize)]
236#[serde(rename_all = "snake_case")]
237pub enum CompletionEvent<T> {
238    Status(CompletionRequestStatus),
239    Event(T),
240}
241
242impl<T> CompletionEvent<T> {
243    pub fn into_status(self) -> Option<CompletionRequestStatus> {
244        match self {
245            Self::Status(status) => Some(status),
246            Self::Event(_) => None,
247        }
248    }
249
250    pub fn into_event(self) -> Option<T> {
251        match self {
252            Self::Event(event) => Some(event),
253            Self::Status(_) => None,
254        }
255    }
256}
257
258#[derive(Serialize, Deserialize)]
259pub struct WebSearchBody {
260    pub query: String,
261}
262
263#[derive(Debug, Serialize, Deserialize, Clone)]
264pub struct WebSearchResponse {
265    pub results: Vec<WebSearchResult>,
266}
267
268#[derive(Debug, Serialize, Deserialize, Clone)]
269pub struct WebSearchResult {
270    pub title: String,
271    pub url: String,
272    pub text: String,
273}
274
275#[derive(Serialize, Deserialize)]
276pub struct CountTokensBody {
277    pub provider: LanguageModelProvider,
278    pub model: String,
279    pub provider_request: serde_json::Value,
280}
281
282#[derive(Serialize, Deserialize)]
283pub struct CountTokensResponse {
284    pub tokens: usize,
285}
286
287#[derive(Debug, PartialEq, Eq, Hash, Clone, Serialize, Deserialize)]
288pub struct LanguageModelId(pub Arc<str>);
289
290impl std::fmt::Display for LanguageModelId {
291    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
292        write!(f, "{}", self.0)
293    }
294}
295
296#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
297pub struct LanguageModel {
298    pub provider: LanguageModelProvider,
299    pub id: LanguageModelId,
300    pub display_name: String,
301    #[serde(default)]
302    pub is_latest: bool,
303    pub max_token_count: usize,
304    pub max_token_count_in_max_mode: Option<usize>,
305    pub max_output_tokens: usize,
306    pub supports_tools: bool,
307    pub supports_images: bool,
308    pub supports_thinking: bool,
309    #[serde(default)]
310    pub supports_fast_mode: bool,
311    pub supported_effort_levels: Vec<SupportedEffortLevel>,
312    #[serde(default)]
313    pub supports_streaming_tools: bool,
314    /// Only used by OpenAI and xAI.
315    #[serde(default)]
316    pub supports_parallel_tool_calls: bool,
317}
318
319#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
320pub struct SupportedEffortLevel {
321    pub name: Arc<str>,
322    pub value: Arc<str>,
323    #[serde(default, skip_serializing_if = "Option::is_none")]
324    pub is_default: Option<bool>,
325}
326
327#[derive(Debug, Serialize, Deserialize)]
328pub struct ListModelsResponse {
329    pub models: Vec<LanguageModel>,
330    pub default_model: Option<LanguageModelId>,
331    pub default_fast_model: Option<LanguageModelId>,
332    pub recommended_models: Vec<LanguageModelId>,
333}
334
335#[derive(Debug, PartialEq, Serialize, Deserialize)]
336pub struct CurrentUsage {
337    pub edit_predictions: UsageData,
338}
339
340#[derive(Debug, PartialEq, Serialize, Deserialize)]
341pub struct UsageData {
342    pub used: u32,
343    pub limit: UsageLimit,
344}
345
346#[cfg(test)]
347mod tests {
348    use super::*;
349
350    #[test]
351    fn test_usage_limit_from_str() {
352        let limit = UsageLimit::from_str("unlimited").unwrap();
353        assert!(matches!(limit, UsageLimit::Unlimited));
354
355        let limit = UsageLimit::from_str(&0.to_string()).unwrap();
356        assert!(matches!(limit, UsageLimit::Limited(0)));
357
358        let limit = UsageLimit::from_str(&50.to_string()).unwrap();
359        assert!(matches!(limit, UsageLimit::Limited(50)));
360
361        for value in ["not_a_number", "50xyz"] {
362            let limit = UsageLimit::from_str(value);
363            assert!(limit.is_err());
364        }
365    }
366}