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 to the file in the repository that contains the input excerpt.
160    #[serde(skip_serializing_if = "Option::is_none", default)]
161    pub input_path: Option<String>,
162    /// Cursor position within the file that contains the input excerpt.
163    #[serde(skip_serializing_if = "Option::is_none", default)]
164    pub cursor_point: Option<Point>,
165    /// SHA of git HEAD commit at time of prediction.
166    #[serde(skip_serializing_if = "Option::is_none", default)]
167    pub head_sha: Option<String>,
168    /// URL of the remote called `origin`.
169    #[serde(skip_serializing_if = "Option::is_none", default)]
170    pub remote_origin_url: Option<String>,
171    /// URL of the remote called `upstream`.
172    #[serde(skip_serializing_if = "Option::is_none", default)]
173    pub remote_upstream_url: Option<String>,
174    /// Recently active files that may be within this repository.
175    #[serde(skip_serializing_if = "Option::is_none", default)]
176    pub recent_files: Option<Vec<PredictEditsRecentFile>>,
177}
178
179/// A zero-indexed point in a text buffer consisting of a row and column.
180#[derive(Debug, Clone, Serialize, Deserialize)]
181pub struct Point {
182    pub row: u32,
183    pub column: u32,
184}
185
186#[derive(Debug, Clone, Serialize, Deserialize)]
187pub struct PredictEditsRecentFile {
188    /// Path to a file within the repository.
189    pub path: String,
190    /// Milliseconds between the editor for this file being active and the request time.
191    pub active_to_now_ms: u32,
192}
193
194#[derive(Debug, Clone, Serialize, Deserialize)]
195pub struct PredictEditsResponse {
196    pub request_id: Uuid,
197    pub output_excerpt: String,
198}
199
200#[derive(Debug, Clone, Serialize, Deserialize)]
201pub struct AcceptEditPredictionBody {
202    pub request_id: Uuid,
203}
204
205#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Clone, Copy, Serialize, Deserialize)]
206#[serde(rename_all = "snake_case")]
207pub enum CompletionMode {
208    Normal,
209    Max,
210}
211
212#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Clone, Copy, Serialize, Deserialize)]
213#[serde(rename_all = "snake_case")]
214pub enum CompletionIntent {
215    UserPrompt,
216    ToolResults,
217    ThreadSummarization,
218    ThreadContextSummarization,
219    CreateFile,
220    EditFile,
221    InlineAssist,
222    TerminalInlineAssist,
223    GenerateGitCommitMessage,
224}
225
226#[derive(Debug, Serialize, Deserialize)]
227pub struct CompletionBody {
228    #[serde(skip_serializing_if = "Option::is_none", default)]
229    pub thread_id: Option<String>,
230    #[serde(skip_serializing_if = "Option::is_none", default)]
231    pub prompt_id: Option<String>,
232    #[serde(skip_serializing_if = "Option::is_none", default)]
233    pub intent: Option<CompletionIntent>,
234    #[serde(skip_serializing_if = "Option::is_none", default)]
235    pub mode: Option<CompletionMode>,
236    pub provider: LanguageModelProvider,
237    pub model: String,
238    pub provider_request: serde_json::Value,
239}
240
241#[derive(Debug, PartialEq, Clone, Serialize, Deserialize)]
242#[serde(rename_all = "snake_case")]
243pub enum CompletionRequestStatus {
244    Queued {
245        position: usize,
246    },
247    Started,
248    Failed {
249        code: String,
250        message: String,
251        request_id: Uuid,
252        /// Retry duration in seconds.
253        retry_after: Option<f64>,
254    },
255    UsageUpdated {
256        amount: usize,
257        limit: UsageLimit,
258    },
259    ToolUseLimitReached,
260}
261
262#[derive(Serialize, Deserialize)]
263#[serde(rename_all = "snake_case")]
264pub enum CompletionEvent<T> {
265    Status(CompletionRequestStatus),
266    Event(T),
267}
268
269impl<T> CompletionEvent<T> {
270    pub fn into_status(self) -> Option<CompletionRequestStatus> {
271        match self {
272            Self::Status(status) => Some(status),
273            Self::Event(_) => None,
274        }
275    }
276
277    pub fn into_event(self) -> Option<T> {
278        match self {
279            Self::Event(event) => Some(event),
280            Self::Status(_) => None,
281        }
282    }
283}
284
285#[derive(Serialize, Deserialize)]
286pub struct WebSearchBody {
287    pub query: String,
288}
289
290#[derive(Debug, Serialize, Deserialize, Clone)]
291pub struct WebSearchResponse {
292    pub results: Vec<WebSearchResult>,
293}
294
295#[derive(Debug, Serialize, Deserialize, Clone)]
296pub struct WebSearchResult {
297    pub title: String,
298    pub url: String,
299    pub text: String,
300}
301
302#[derive(Serialize, Deserialize)]
303pub struct CountTokensBody {
304    pub provider: LanguageModelProvider,
305    pub model: String,
306    pub provider_request: serde_json::Value,
307}
308
309#[derive(Serialize, Deserialize)]
310pub struct CountTokensResponse {
311    pub tokens: usize,
312}
313
314#[derive(Debug, PartialEq, Eq, Hash, Clone, Serialize, Deserialize)]
315pub struct LanguageModelId(pub Arc<str>);
316
317impl std::fmt::Display for LanguageModelId {
318    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
319        write!(f, "{}", self.0)
320    }
321}
322
323#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
324pub struct LanguageModel {
325    pub provider: LanguageModelProvider,
326    pub id: LanguageModelId,
327    pub display_name: String,
328    pub max_token_count: usize,
329    pub max_token_count_in_max_mode: Option<usize>,
330    pub max_output_tokens: usize,
331    pub supports_tools: bool,
332    pub supports_images: bool,
333    pub supports_thinking: bool,
334    pub supports_max_mode: bool,
335}
336
337#[derive(Debug, Serialize, Deserialize)]
338pub struct ListModelsResponse {
339    pub models: Vec<LanguageModel>,
340    pub default_model: LanguageModelId,
341    pub default_fast_model: LanguageModelId,
342    pub recommended_models: Vec<LanguageModelId>,
343}
344
345#[derive(Debug, Serialize, Deserialize)]
346pub struct GetSubscriptionResponse {
347    pub plan: Plan,
348    pub usage: Option<CurrentUsage>,
349}
350
351#[derive(Debug, PartialEq, Serialize, Deserialize)]
352pub struct CurrentUsage {
353    pub model_requests: UsageData,
354    pub edit_predictions: UsageData,
355}
356
357#[derive(Debug, PartialEq, Serialize, Deserialize)]
358pub struct UsageData {
359    pub used: u32,
360    pub limit: UsageLimit,
361}
362
363#[cfg(test)]
364mod tests {
365    use pretty_assertions::assert_eq;
366    use serde_json::json;
367
368    use super::*;
369
370    #[test]
371    fn test_plan_deserialize_snake_case() {
372        let plan = serde_json::from_value::<Plan>(json!("zed_free")).unwrap();
373        assert_eq!(plan, Plan::ZedFree);
374
375        let plan = serde_json::from_value::<Plan>(json!("zed_pro")).unwrap();
376        assert_eq!(plan, Plan::ZedPro);
377
378        let plan = serde_json::from_value::<Plan>(json!("zed_pro_trial")).unwrap();
379        assert_eq!(plan, Plan::ZedProTrial);
380    }
381
382    #[test]
383    fn test_plan_deserialize_aliases() {
384        let plan = serde_json::from_value::<Plan>(json!("Free")).unwrap();
385        assert_eq!(plan, Plan::ZedFree);
386
387        let plan = serde_json::from_value::<Plan>(json!("ZedPro")).unwrap();
388        assert_eq!(plan, Plan::ZedPro);
389
390        let plan = serde_json::from_value::<Plan>(json!("ZedProTrial")).unwrap();
391        assert_eq!(plan, Plan::ZedProTrial);
392    }
393
394    #[test]
395    fn test_usage_limit_from_str() {
396        let limit = UsageLimit::from_str("unlimited").unwrap();
397        assert!(matches!(limit, UsageLimit::Unlimited));
398
399        let limit = UsageLimit::from_str(&0.to_string()).unwrap();
400        assert!(matches!(limit, UsageLimit::Limited(0)));
401
402        let limit = UsageLimit::from_str(&50.to_string()).unwrap();
403        assert!(matches!(limit, UsageLimit::Limited(50)));
404
405        for value in ["not_a_number", "50xyz"] {
406            let limit = UsageLimit::from_str(value);
407            assert!(limit.is_err());
408        }
409    }
410}