1use crate::{
2 ContextServerRegistry, CopyPathTool, CreateDirectoryTool, DbLanguageModel, DbThread,
3 DeletePathTool, DiagnosticsTool, EditFileTool, FetchTool, FindPathTool, GrepTool,
4 ListDirectoryTool, MovePathTool, NowTool, OpenTool, ReadFileTool, SystemPromptTemplate,
5 Template, Templates, TerminalTool, ThinkingTool, WebSearchTool,
6};
7use acp_thread::{MentionUri, UserMessageId};
8use action_log::ActionLog;
9use agent::thread::{GitState, ProjectSnapshot, WorktreeSnapshot};
10use agent_client_protocol as acp;
11use agent_settings::{
12 AgentProfileId, AgentSettings, CompletionMode, SUMMARIZE_THREAD_DETAILED_PROMPT,
13 SUMMARIZE_THREAD_PROMPT,
14};
15use anyhow::{Context as _, Result, anyhow};
16use assistant_tool::adapt_schema_to_format;
17use chrono::{DateTime, Utc};
18use client::{ModelRequestUsage, RequestUsage};
19use cloud_llm_client::{CompletionIntent, CompletionRequestStatus, UsageLimit};
20use collections::{HashMap, IndexMap};
21use fs::Fs;
22use futures::{
23 FutureExt,
24 channel::{mpsc, oneshot},
25 future::Shared,
26 stream::FuturesUnordered,
27};
28use git::repository::DiffType;
29use gpui::{
30 App, AppContext, AsyncApp, Context, Entity, EventEmitter, SharedString, Task, WeakEntity,
31};
32use language_model::{
33 LanguageModel, LanguageModelCompletionError, LanguageModelCompletionEvent, LanguageModelExt,
34 LanguageModelImage, LanguageModelProviderId, LanguageModelRegistry, LanguageModelRequest,
35 LanguageModelRequestMessage, LanguageModelRequestTool, LanguageModelToolResult,
36 LanguageModelToolResultContent, LanguageModelToolSchemaFormat, LanguageModelToolUse,
37 LanguageModelToolUseId, Role, SelectedModel, StopReason, TokenUsage,
38};
39use project::{
40 Project,
41 git_store::{GitStore, RepositoryState},
42};
43use prompt_store::ProjectContext;
44use schemars::{JsonSchema, Schema};
45use serde::{Deserialize, Serialize};
46use settings::{Settings, update_settings_file};
47use smol::stream::StreamExt;
48use std::{
49 collections::BTreeMap,
50 path::Path,
51 sync::Arc,
52 time::{Duration, Instant},
53};
54use std::{fmt::Write, ops::Range};
55use util::{ResultExt, markdown::MarkdownCodeBlock};
56use uuid::Uuid;
57
58const TOOL_CANCELED_MESSAGE: &str = "Tool canceled by user";
59
60/// The ID of the user prompt that initiated a request.
61///
62/// This equates to the user physically submitting a message to the model (e.g., by pressing the Enter key).
63#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Clone, Serialize, Deserialize)]
64pub struct PromptId(Arc<str>);
65
66impl PromptId {
67 pub fn new() -> Self {
68 Self(Uuid::new_v4().to_string().into())
69 }
70}
71
72impl std::fmt::Display for PromptId {
73 fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
74 write!(f, "{}", self.0)
75 }
76}
77
78pub(crate) const MAX_RETRY_ATTEMPTS: u8 = 4;
79pub(crate) const BASE_RETRY_DELAY: Duration = Duration::from_secs(5);
80
81#[derive(Debug, Clone)]
82enum RetryStrategy {
83 ExponentialBackoff {
84 initial_delay: Duration,
85 max_attempts: u8,
86 },
87 Fixed {
88 delay: Duration,
89 max_attempts: u8,
90 },
91}
92
93#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
94pub enum Message {
95 User(UserMessage),
96 Agent(AgentMessage),
97 Resume,
98}
99
100impl Message {
101 pub fn as_agent_message(&self) -> Option<&AgentMessage> {
102 match self {
103 Message::Agent(agent_message) => Some(agent_message),
104 _ => None,
105 }
106 }
107
108 pub fn to_request(&self) -> Vec<LanguageModelRequestMessage> {
109 match self {
110 Message::User(message) => vec![message.to_request()],
111 Message::Agent(message) => message.to_request(),
112 Message::Resume => vec![LanguageModelRequestMessage {
113 role: Role::User,
114 content: vec!["Continue where you left off".into()],
115 cache: false,
116 }],
117 }
118 }
119
120 pub fn to_markdown(&self) -> String {
121 match self {
122 Message::User(message) => message.to_markdown(),
123 Message::Agent(message) => message.to_markdown(),
124 Message::Resume => "[resumed after tool use limit was reached]".into(),
125 }
126 }
127
128 pub fn role(&self) -> Role {
129 match self {
130 Message::User(_) | Message::Resume => Role::User,
131 Message::Agent(_) => Role::Assistant,
132 }
133 }
134}
135
136#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
137pub struct UserMessage {
138 pub id: UserMessageId,
139 pub content: Vec<UserMessageContent>,
140}
141
142#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
143pub enum UserMessageContent {
144 Text(String),
145 Mention { uri: MentionUri, content: String },
146 Image(LanguageModelImage),
147}
148
149impl UserMessage {
150 pub fn to_markdown(&self) -> String {
151 let mut markdown = String::from("## User\n\n");
152
153 for content in &self.content {
154 match content {
155 UserMessageContent::Text(text) => {
156 markdown.push_str(text);
157 markdown.push('\n');
158 }
159 UserMessageContent::Image(_) => {
160 markdown.push_str("<image />\n");
161 }
162 UserMessageContent::Mention { uri, content } => {
163 if !content.is_empty() {
164 let _ = writeln!(&mut markdown, "{}\n\n{}", uri.as_link(), content);
165 } else {
166 let _ = writeln!(&mut markdown, "{}", uri.as_link());
167 }
168 }
169 }
170 }
171
172 markdown
173 }
174
175 fn to_request(&self) -> LanguageModelRequestMessage {
176 let mut message = LanguageModelRequestMessage {
177 role: Role::User,
178 content: Vec::with_capacity(self.content.len()),
179 cache: false,
180 };
181
182 const OPEN_CONTEXT: &str = "<context>\n\
183 The following items were attached by the user. \
184 They are up-to-date and don't need to be re-read.\n\n";
185
186 const OPEN_FILES_TAG: &str = "<files>";
187 const OPEN_DIRECTORIES_TAG: &str = "<directories>";
188 const OPEN_SYMBOLS_TAG: &str = "<symbols>";
189 const OPEN_THREADS_TAG: &str = "<threads>";
190 const OPEN_FETCH_TAG: &str = "<fetched_urls>";
191 const OPEN_RULES_TAG: &str =
192 "<rules>\nThe user has specified the following rules that should be applied:\n";
193
194 let mut file_context = OPEN_FILES_TAG.to_string();
195 let mut directory_context = OPEN_DIRECTORIES_TAG.to_string();
196 let mut symbol_context = OPEN_SYMBOLS_TAG.to_string();
197 let mut thread_context = OPEN_THREADS_TAG.to_string();
198 let mut fetch_context = OPEN_FETCH_TAG.to_string();
199 let mut rules_context = OPEN_RULES_TAG.to_string();
200
201 for chunk in &self.content {
202 let chunk = match chunk {
203 UserMessageContent::Text(text) => {
204 language_model::MessageContent::Text(text.clone())
205 }
206 UserMessageContent::Image(value) => {
207 language_model::MessageContent::Image(value.clone())
208 }
209 UserMessageContent::Mention { uri, content } => {
210 match uri {
211 MentionUri::File { abs_path } => {
212 write!(
213 &mut symbol_context,
214 "\n{}",
215 MarkdownCodeBlock {
216 tag: &codeblock_tag(abs_path, None),
217 text: &content.to_string(),
218 }
219 )
220 .ok();
221 }
222 MentionUri::Directory { .. } => {
223 write!(&mut directory_context, "\n{}\n", content).ok();
224 }
225 MentionUri::Symbol {
226 path, line_range, ..
227 }
228 | MentionUri::Selection {
229 path, line_range, ..
230 } => {
231 write!(
232 &mut rules_context,
233 "\n{}",
234 MarkdownCodeBlock {
235 tag: &codeblock_tag(path, Some(line_range)),
236 text: content
237 }
238 )
239 .ok();
240 }
241 MentionUri::Thread { .. } => {
242 write!(&mut thread_context, "\n{}\n", content).ok();
243 }
244 MentionUri::TextThread { .. } => {
245 write!(&mut thread_context, "\n{}\n", content).ok();
246 }
247 MentionUri::Rule { .. } => {
248 write!(
249 &mut rules_context,
250 "\n{}",
251 MarkdownCodeBlock {
252 tag: "",
253 text: content
254 }
255 )
256 .ok();
257 }
258 MentionUri::Fetch { url } => {
259 write!(&mut fetch_context, "\nFetch: {}\n\n{}", url, content).ok();
260 }
261 }
262
263 language_model::MessageContent::Text(uri.as_link().to_string())
264 }
265 };
266
267 message.content.push(chunk);
268 }
269
270 let len_before_context = message.content.len();
271
272 if file_context.len() > OPEN_FILES_TAG.len() {
273 file_context.push_str("</files>\n");
274 message
275 .content
276 .push(language_model::MessageContent::Text(file_context));
277 }
278
279 if directory_context.len() > OPEN_DIRECTORIES_TAG.len() {
280 directory_context.push_str("</directories>\n");
281 message
282 .content
283 .push(language_model::MessageContent::Text(directory_context));
284 }
285
286 if symbol_context.len() > OPEN_SYMBOLS_TAG.len() {
287 symbol_context.push_str("</symbols>\n");
288 message
289 .content
290 .push(language_model::MessageContent::Text(symbol_context));
291 }
292
293 if thread_context.len() > OPEN_THREADS_TAG.len() {
294 thread_context.push_str("</threads>\n");
295 message
296 .content
297 .push(language_model::MessageContent::Text(thread_context));
298 }
299
300 if fetch_context.len() > OPEN_FETCH_TAG.len() {
301 fetch_context.push_str("</fetched_urls>\n");
302 message
303 .content
304 .push(language_model::MessageContent::Text(fetch_context));
305 }
306
307 if rules_context.len() > OPEN_RULES_TAG.len() {
308 rules_context.push_str("</user_rules>\n");
309 message
310 .content
311 .push(language_model::MessageContent::Text(rules_context));
312 }
313
314 if message.content.len() > len_before_context {
315 message.content.insert(
316 len_before_context,
317 language_model::MessageContent::Text(OPEN_CONTEXT.into()),
318 );
319 message
320 .content
321 .push(language_model::MessageContent::Text("</context>".into()));
322 }
323
324 message
325 }
326}
327
328fn codeblock_tag(full_path: &Path, line_range: Option<&Range<u32>>) -> String {
329 let mut result = String::new();
330
331 if let Some(extension) = full_path.extension().and_then(|ext| ext.to_str()) {
332 let _ = write!(result, "{} ", extension);
333 }
334
335 let _ = write!(result, "{}", full_path.display());
336
337 if let Some(range) = line_range {
338 if range.start == range.end {
339 let _ = write!(result, ":{}", range.start + 1);
340 } else {
341 let _ = write!(result, ":{}-{}", range.start + 1, range.end + 1);
342 }
343 }
344
345 result
346}
347
348impl AgentMessage {
349 pub fn to_markdown(&self) -> String {
350 let mut markdown = String::from("## Assistant\n\n");
351
352 for content in &self.content {
353 match content {
354 AgentMessageContent::Text(text) => {
355 markdown.push_str(text);
356 markdown.push('\n');
357 }
358 AgentMessageContent::Thinking { text, .. } => {
359 markdown.push_str("<think>");
360 markdown.push_str(text);
361 markdown.push_str("</think>\n");
362 }
363 AgentMessageContent::RedactedThinking(_) => {
364 markdown.push_str("<redacted_thinking />\n")
365 }
366 AgentMessageContent::ToolUse(tool_use) => {
367 markdown.push_str(&format!(
368 "**Tool Use**: {} (ID: {})\n",
369 tool_use.name, tool_use.id
370 ));
371 markdown.push_str(&format!(
372 "{}\n",
373 MarkdownCodeBlock {
374 tag: "json",
375 text: &format!("{:#}", tool_use.input)
376 }
377 ));
378 }
379 }
380 }
381
382 for tool_result in self.tool_results.values() {
383 markdown.push_str(&format!(
384 "**Tool Result**: {} (ID: {})\n\n",
385 tool_result.tool_name, tool_result.tool_use_id
386 ));
387 if tool_result.is_error {
388 markdown.push_str("**ERROR:**\n");
389 }
390
391 match &tool_result.content {
392 LanguageModelToolResultContent::Text(text) => {
393 writeln!(markdown, "{text}\n").ok();
394 }
395 LanguageModelToolResultContent::Image(_) => {
396 writeln!(markdown, "<image />\n").ok();
397 }
398 }
399
400 if let Some(output) = tool_result.output.as_ref() {
401 writeln!(
402 markdown,
403 "**Debug Output**:\n\n```json\n{}\n```\n",
404 serde_json::to_string_pretty(output).unwrap()
405 )
406 .unwrap();
407 }
408 }
409
410 markdown
411 }
412
413 pub fn to_request(&self) -> Vec<LanguageModelRequestMessage> {
414 let mut assistant_message = LanguageModelRequestMessage {
415 role: Role::Assistant,
416 content: Vec::with_capacity(self.content.len()),
417 cache: false,
418 };
419 for chunk in &self.content {
420 let chunk = match chunk {
421 AgentMessageContent::Text(text) => {
422 language_model::MessageContent::Text(text.clone())
423 }
424 AgentMessageContent::Thinking { text, signature } => {
425 language_model::MessageContent::Thinking {
426 text: text.clone(),
427 signature: signature.clone(),
428 }
429 }
430 AgentMessageContent::RedactedThinking(value) => {
431 language_model::MessageContent::RedactedThinking(value.clone())
432 }
433 AgentMessageContent::ToolUse(value) => {
434 language_model::MessageContent::ToolUse(value.clone())
435 }
436 };
437 assistant_message.content.push(chunk);
438 }
439
440 let mut user_message = LanguageModelRequestMessage {
441 role: Role::User,
442 content: Vec::new(),
443 cache: false,
444 };
445
446 for tool_result in self.tool_results.values() {
447 user_message
448 .content
449 .push(language_model::MessageContent::ToolResult(
450 tool_result.clone(),
451 ));
452 }
453
454 let mut messages = Vec::new();
455 if !assistant_message.content.is_empty() {
456 messages.push(assistant_message);
457 }
458 if !user_message.content.is_empty() {
459 messages.push(user_message);
460 }
461 messages
462 }
463}
464
465#[derive(Default, Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
466pub struct AgentMessage {
467 pub content: Vec<AgentMessageContent>,
468 pub tool_results: IndexMap<LanguageModelToolUseId, LanguageModelToolResult>,
469}
470
471#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
472pub enum AgentMessageContent {
473 Text(String),
474 Thinking {
475 text: String,
476 signature: Option<String>,
477 },
478 RedactedThinking(String),
479 ToolUse(LanguageModelToolUse),
480}
481
482#[derive(Debug)]
483pub enum ThreadEvent {
484 UserMessage(UserMessage),
485 AgentText(String),
486 AgentThinking(String),
487 ToolCall(acp::ToolCall),
488 ToolCallUpdate(acp_thread::ToolCallUpdate),
489 ToolCallAuthorization(ToolCallAuthorization),
490 Retry(acp_thread::RetryStatus),
491 Stop(acp::StopReason),
492}
493
494#[derive(Debug)]
495pub struct ToolCallAuthorization {
496 pub tool_call: acp::ToolCallUpdate,
497 pub options: Vec<acp::PermissionOption>,
498 pub response: oneshot::Sender<acp::PermissionOptionId>,
499}
500
501#[derive(Debug, thiserror::Error)]
502enum CompletionError {
503 #[error("max tokens")]
504 MaxTokens,
505 #[error("refusal")]
506 Refusal,
507 #[error(transparent)]
508 Other(#[from] anyhow::Error),
509}
510
511pub struct Thread {
512 id: acp::SessionId,
513 prompt_id: PromptId,
514 updated_at: DateTime<Utc>,
515 title: Option<SharedString>,
516 pending_title_generation: Option<Task<()>>,
517 summary: Option<SharedString>,
518 messages: Vec<Message>,
519 completion_mode: CompletionMode,
520 /// Holds the task that handles agent interaction until the end of the turn.
521 /// Survives across multiple requests as the model performs tool calls and
522 /// we run tools, report their results.
523 running_turn: Option<RunningTurn>,
524 pending_message: Option<AgentMessage>,
525 tools: BTreeMap<SharedString, Arc<dyn AnyAgentTool>>,
526 tool_use_limit_reached: bool,
527 request_token_usage: HashMap<UserMessageId, language_model::TokenUsage>,
528 #[allow(unused)]
529 cumulative_token_usage: TokenUsage,
530 #[allow(unused)]
531 initial_project_snapshot: Shared<Task<Option<Arc<ProjectSnapshot>>>>,
532 context_server_registry: Entity<ContextServerRegistry>,
533 profile_id: AgentProfileId,
534 project_context: Entity<ProjectContext>,
535 templates: Arc<Templates>,
536 model: Option<Arc<dyn LanguageModel>>,
537 summarization_model: Option<Arc<dyn LanguageModel>>,
538 pub(crate) project: Entity<Project>,
539 pub(crate) action_log: Entity<ActionLog>,
540}
541
542impl Thread {
543 pub fn new(
544 project: Entity<Project>,
545 project_context: Entity<ProjectContext>,
546 context_server_registry: Entity<ContextServerRegistry>,
547 action_log: Entity<ActionLog>,
548 templates: Arc<Templates>,
549 model: Option<Arc<dyn LanguageModel>>,
550 cx: &mut Context<Self>,
551 ) -> Self {
552 let profile_id = AgentSettings::get_global(cx).default_profile.clone();
553 Self {
554 id: acp::SessionId(uuid::Uuid::new_v4().to_string().into()),
555 prompt_id: PromptId::new(),
556 updated_at: Utc::now(),
557 title: None,
558 pending_title_generation: None,
559 summary: None,
560 messages: Vec::new(),
561 completion_mode: AgentSettings::get_global(cx).preferred_completion_mode,
562 running_turn: None,
563 pending_message: None,
564 tools: BTreeMap::default(),
565 tool_use_limit_reached: false,
566 request_token_usage: HashMap::default(),
567 cumulative_token_usage: TokenUsage::default(),
568 initial_project_snapshot: {
569 let project_snapshot = Self::project_snapshot(project.clone(), cx);
570 cx.foreground_executor()
571 .spawn(async move { Some(project_snapshot.await) })
572 .shared()
573 },
574 context_server_registry,
575 profile_id,
576 project_context,
577 templates,
578 model,
579 summarization_model: None,
580 project,
581 action_log,
582 }
583 }
584
585 pub fn id(&self) -> &acp::SessionId {
586 &self.id
587 }
588
589 pub fn replay(
590 &mut self,
591 cx: &mut Context<Self>,
592 ) -> mpsc::UnboundedReceiver<Result<ThreadEvent>> {
593 let (tx, rx) = mpsc::unbounded();
594 let stream = ThreadEventStream(tx);
595 for message in &self.messages {
596 match message {
597 Message::User(user_message) => stream.send_user_message(user_message),
598 Message::Agent(assistant_message) => {
599 for content in &assistant_message.content {
600 match content {
601 AgentMessageContent::Text(text) => stream.send_text(text),
602 AgentMessageContent::Thinking { text, .. } => {
603 stream.send_thinking(text)
604 }
605 AgentMessageContent::RedactedThinking(_) => {}
606 AgentMessageContent::ToolUse(tool_use) => {
607 self.replay_tool_call(
608 tool_use,
609 assistant_message.tool_results.get(&tool_use.id),
610 &stream,
611 cx,
612 );
613 }
614 }
615 }
616 }
617 Message::Resume => {}
618 }
619 }
620 rx
621 }
622
623 fn replay_tool_call(
624 &self,
625 tool_use: &LanguageModelToolUse,
626 tool_result: Option<&LanguageModelToolResult>,
627 stream: &ThreadEventStream,
628 cx: &mut Context<Self>,
629 ) {
630 let Some(tool) = self.tools.get(tool_use.name.as_ref()) else {
631 stream
632 .0
633 .unbounded_send(Ok(ThreadEvent::ToolCall(acp::ToolCall {
634 id: acp::ToolCallId(tool_use.id.to_string().into()),
635 title: tool_use.name.to_string(),
636 kind: acp::ToolKind::Other,
637 status: acp::ToolCallStatus::Failed,
638 content: Vec::new(),
639 locations: Vec::new(),
640 raw_input: Some(tool_use.input.clone()),
641 raw_output: None,
642 })))
643 .ok();
644 return;
645 };
646
647 let title = tool.initial_title(tool_use.input.clone());
648 let kind = tool.kind();
649 stream.send_tool_call(&tool_use.id, title, kind, tool_use.input.clone());
650
651 let output = tool_result
652 .as_ref()
653 .and_then(|result| result.output.clone());
654 if let Some(output) = output.clone() {
655 let tool_event_stream = ToolCallEventStream::new(
656 tool_use.id.clone(),
657 stream.clone(),
658 Some(self.project.read(cx).fs().clone()),
659 );
660 tool.replay(tool_use.input.clone(), output, tool_event_stream, cx)
661 .log_err();
662 }
663
664 stream.update_tool_call_fields(
665 &tool_use.id,
666 acp::ToolCallUpdateFields {
667 status: Some(acp::ToolCallStatus::Completed),
668 raw_output: output,
669 ..Default::default()
670 },
671 );
672 }
673
674 pub fn from_db(
675 id: acp::SessionId,
676 db_thread: DbThread,
677 project: Entity<Project>,
678 project_context: Entity<ProjectContext>,
679 context_server_registry: Entity<ContextServerRegistry>,
680 action_log: Entity<ActionLog>,
681 templates: Arc<Templates>,
682 cx: &mut Context<Self>,
683 ) -> Self {
684 let profile_id = db_thread
685 .profile
686 .unwrap_or_else(|| AgentSettings::get_global(cx).default_profile.clone());
687 let model = LanguageModelRegistry::global(cx).update(cx, |registry, cx| {
688 db_thread
689 .model
690 .and_then(|model| {
691 let model = SelectedModel {
692 provider: model.provider.clone().into(),
693 model: model.model.into(),
694 };
695 registry.select_model(&model, cx)
696 })
697 .or_else(|| registry.default_model())
698 .map(|model| model.model)
699 });
700
701 Self {
702 id,
703 prompt_id: PromptId::new(),
704 title: if db_thread.title.is_empty() {
705 None
706 } else {
707 Some(db_thread.title.clone())
708 },
709 pending_title_generation: None,
710 summary: db_thread.detailed_summary,
711 messages: db_thread.messages,
712 completion_mode: db_thread.completion_mode.unwrap_or_default(),
713 running_turn: None,
714 pending_message: None,
715 tools: BTreeMap::default(),
716 tool_use_limit_reached: false,
717 request_token_usage: db_thread.request_token_usage.clone(),
718 cumulative_token_usage: db_thread.cumulative_token_usage,
719 initial_project_snapshot: Task::ready(db_thread.initial_project_snapshot).shared(),
720 context_server_registry,
721 profile_id,
722 project_context,
723 templates,
724 model,
725 summarization_model: None,
726 project,
727 action_log,
728 updated_at: db_thread.updated_at,
729 }
730 }
731
732 pub fn to_db(&self, cx: &App) -> Task<DbThread> {
733 let initial_project_snapshot = self.initial_project_snapshot.clone();
734 let mut thread = DbThread {
735 title: self.title(),
736 messages: self.messages.clone(),
737 updated_at: self.updated_at,
738 detailed_summary: self.summary.clone(),
739 initial_project_snapshot: None,
740 cumulative_token_usage: self.cumulative_token_usage,
741 request_token_usage: self.request_token_usage.clone(),
742 model: self.model.as_ref().map(|model| DbLanguageModel {
743 provider: model.provider_id().to_string(),
744 model: model.name().0.to_string(),
745 }),
746 completion_mode: Some(self.completion_mode),
747 profile: Some(self.profile_id.clone()),
748 };
749
750 cx.background_spawn(async move {
751 let initial_project_snapshot = initial_project_snapshot.await;
752 thread.initial_project_snapshot = initial_project_snapshot;
753 thread
754 })
755 }
756
757 /// Create a snapshot of the current project state including git information and unsaved buffers.
758 fn project_snapshot(
759 project: Entity<Project>,
760 cx: &mut Context<Self>,
761 ) -> Task<Arc<agent::thread::ProjectSnapshot>> {
762 let git_store = project.read(cx).git_store().clone();
763 let worktree_snapshots: Vec<_> = project
764 .read(cx)
765 .visible_worktrees(cx)
766 .map(|worktree| Self::worktree_snapshot(worktree, git_store.clone(), cx))
767 .collect();
768
769 cx.spawn(async move |_, cx| {
770 let worktree_snapshots = futures::future::join_all(worktree_snapshots).await;
771
772 let mut unsaved_buffers = Vec::new();
773 cx.update(|app_cx| {
774 let buffer_store = project.read(app_cx).buffer_store();
775 for buffer_handle in buffer_store.read(app_cx).buffers() {
776 let buffer = buffer_handle.read(app_cx);
777 if buffer.is_dirty()
778 && let Some(file) = buffer.file()
779 {
780 let path = file.path().to_string_lossy().to_string();
781 unsaved_buffers.push(path);
782 }
783 }
784 })
785 .ok();
786
787 Arc::new(ProjectSnapshot {
788 worktree_snapshots,
789 unsaved_buffer_paths: unsaved_buffers,
790 timestamp: Utc::now(),
791 })
792 })
793 }
794
795 fn worktree_snapshot(
796 worktree: Entity<project::Worktree>,
797 git_store: Entity<GitStore>,
798 cx: &App,
799 ) -> Task<agent::thread::WorktreeSnapshot> {
800 cx.spawn(async move |cx| {
801 // Get worktree path and snapshot
802 let worktree_info = cx.update(|app_cx| {
803 let worktree = worktree.read(app_cx);
804 let path = worktree.abs_path().to_string_lossy().to_string();
805 let snapshot = worktree.snapshot();
806 (path, snapshot)
807 });
808
809 let Ok((worktree_path, _snapshot)) = worktree_info else {
810 return WorktreeSnapshot {
811 worktree_path: String::new(),
812 git_state: None,
813 };
814 };
815
816 let git_state = git_store
817 .update(cx, |git_store, cx| {
818 git_store
819 .repositories()
820 .values()
821 .find(|repo| {
822 repo.read(cx)
823 .abs_path_to_repo_path(&worktree.read(cx).abs_path())
824 .is_some()
825 })
826 .cloned()
827 })
828 .ok()
829 .flatten()
830 .map(|repo| {
831 repo.update(cx, |repo, _| {
832 let current_branch =
833 repo.branch.as_ref().map(|branch| branch.name().to_owned());
834 repo.send_job(None, |state, _| async move {
835 let RepositoryState::Local { backend, .. } = state else {
836 return GitState {
837 remote_url: None,
838 head_sha: None,
839 current_branch,
840 diff: None,
841 };
842 };
843
844 let remote_url = backend.remote_url("origin");
845 let head_sha = backend.head_sha().await;
846 let diff = backend.diff(DiffType::HeadToWorktree).await.ok();
847
848 GitState {
849 remote_url,
850 head_sha,
851 current_branch,
852 diff,
853 }
854 })
855 })
856 });
857
858 let git_state = match git_state {
859 Some(git_state) => match git_state.ok() {
860 Some(git_state) => git_state.await.ok(),
861 None => None,
862 },
863 None => None,
864 };
865
866 WorktreeSnapshot {
867 worktree_path,
868 git_state,
869 }
870 })
871 }
872
873 pub fn project_context(&self) -> &Entity<ProjectContext> {
874 &self.project_context
875 }
876
877 pub fn project(&self) -> &Entity<Project> {
878 &self.project
879 }
880
881 pub fn action_log(&self) -> &Entity<ActionLog> {
882 &self.action_log
883 }
884
885 pub fn is_empty(&self) -> bool {
886 self.messages.is_empty() && self.title.is_none()
887 }
888
889 pub fn model(&self) -> Option<&Arc<dyn LanguageModel>> {
890 self.model.as_ref()
891 }
892
893 pub fn set_model(&mut self, model: Arc<dyn LanguageModel>, cx: &mut Context<Self>) {
894 let old_usage = self.latest_token_usage();
895 self.model = Some(model);
896 let new_usage = self.latest_token_usage();
897 if old_usage != new_usage {
898 cx.emit(TokenUsageUpdated(new_usage));
899 }
900 cx.notify()
901 }
902
903 pub fn summarization_model(&self) -> Option<&Arc<dyn LanguageModel>> {
904 self.summarization_model.as_ref()
905 }
906
907 pub fn set_summarization_model(
908 &mut self,
909 model: Option<Arc<dyn LanguageModel>>,
910 cx: &mut Context<Self>,
911 ) {
912 self.summarization_model = model;
913 cx.notify()
914 }
915
916 pub fn completion_mode(&self) -> CompletionMode {
917 self.completion_mode
918 }
919
920 pub fn set_completion_mode(&mut self, mode: CompletionMode, cx: &mut Context<Self>) {
921 let old_usage = self.latest_token_usage();
922 self.completion_mode = mode;
923 let new_usage = self.latest_token_usage();
924 if old_usage != new_usage {
925 cx.emit(TokenUsageUpdated(new_usage));
926 }
927 cx.notify()
928 }
929
930 #[cfg(any(test, feature = "test-support"))]
931 pub fn last_message(&self) -> Option<Message> {
932 if let Some(message) = self.pending_message.clone() {
933 Some(Message::Agent(message))
934 } else {
935 self.messages.last().cloned()
936 }
937 }
938
939 pub fn add_default_tools(&mut self, cx: &mut Context<Self>) {
940 let language_registry = self.project.read(cx).languages().clone();
941 self.add_tool(CopyPathTool::new(self.project.clone()));
942 self.add_tool(CreateDirectoryTool::new(self.project.clone()));
943 self.add_tool(DeletePathTool::new(
944 self.project.clone(),
945 self.action_log.clone(),
946 ));
947 self.add_tool(DiagnosticsTool::new(self.project.clone()));
948 self.add_tool(EditFileTool::new(cx.weak_entity(), language_registry));
949 self.add_tool(FetchTool::new(self.project.read(cx).client().http_client()));
950 self.add_tool(FindPathTool::new(self.project.clone()));
951 self.add_tool(GrepTool::new(self.project.clone()));
952 self.add_tool(ListDirectoryTool::new(self.project.clone()));
953 self.add_tool(MovePathTool::new(self.project.clone()));
954 self.add_tool(NowTool);
955 self.add_tool(OpenTool::new(self.project.clone()));
956 self.add_tool(ReadFileTool::new(
957 self.project.clone(),
958 self.action_log.clone(),
959 ));
960 self.add_tool(TerminalTool::new(self.project.clone(), cx));
961 self.add_tool(ThinkingTool);
962 self.add_tool(WebSearchTool); // TODO: Enable this only if it's a zed model.
963 }
964
965 pub fn add_tool(&mut self, tool: impl AgentTool) {
966 self.tools.insert(tool.name(), tool.erase());
967 }
968
969 pub fn remove_tool(&mut self, name: &str) -> bool {
970 self.tools.remove(name).is_some()
971 }
972
973 pub fn profile(&self) -> &AgentProfileId {
974 &self.profile_id
975 }
976
977 pub fn set_profile(&mut self, profile_id: AgentProfileId) {
978 self.profile_id = profile_id;
979 }
980
981 pub fn cancel(&mut self, cx: &mut Context<Self>) {
982 if let Some(running_turn) = self.running_turn.take() {
983 running_turn.cancel();
984 }
985 self.flush_pending_message(cx);
986 }
987
988 fn update_token_usage(&mut self, update: language_model::TokenUsage, cx: &mut Context<Self>) {
989 let Some(last_user_message) = self.last_user_message() else {
990 return;
991 };
992
993 self.request_token_usage
994 .insert(last_user_message.id.clone(), update);
995 cx.emit(TokenUsageUpdated(self.latest_token_usage()));
996 cx.notify();
997 }
998
999 pub fn truncate(&mut self, message_id: UserMessageId, cx: &mut Context<Self>) -> Result<()> {
1000 self.cancel(cx);
1001 let Some(position) = self.messages.iter().position(
1002 |msg| matches!(msg, Message::User(UserMessage { id, .. }) if id == &message_id),
1003 ) else {
1004 return Err(anyhow!("Message not found"));
1005 };
1006
1007 for message in self.messages.drain(position..) {
1008 match message {
1009 Message::User(message) => {
1010 self.request_token_usage.remove(&message.id);
1011 }
1012 Message::Agent(_) | Message::Resume => {}
1013 }
1014 }
1015 self.summary = None;
1016 cx.notify();
1017 Ok(())
1018 }
1019
1020 pub fn latest_token_usage(&self) -> Option<acp_thread::TokenUsage> {
1021 let last_user_message = self.last_user_message()?;
1022 let tokens = self.request_token_usage.get(&last_user_message.id)?;
1023 let model = self.model.clone()?;
1024
1025 Some(acp_thread::TokenUsage {
1026 max_tokens: model.max_token_count_for_mode(self.completion_mode.into()),
1027 used_tokens: tokens.total_tokens(),
1028 })
1029 }
1030
1031 pub fn resume(
1032 &mut self,
1033 cx: &mut Context<Self>,
1034 ) -> Result<mpsc::UnboundedReceiver<Result<ThreadEvent>>> {
1035 anyhow::ensure!(
1036 self.tool_use_limit_reached,
1037 "can only resume after tool use limit is reached"
1038 );
1039
1040 self.messages.push(Message::Resume);
1041 cx.notify();
1042
1043 log::info!("Total messages in thread: {}", self.messages.len());
1044 self.run_turn(cx)
1045 }
1046
1047 /// Sending a message results in the model streaming a response, which could include tool calls.
1048 /// After calling tools, the model will stops and waits for any outstanding tool calls to be completed and their results sent.
1049 /// The returned channel will report all the occurrences in which the model stops before erroring or ending its turn.
1050 pub fn send<T>(
1051 &mut self,
1052 id: UserMessageId,
1053 content: impl IntoIterator<Item = T>,
1054 cx: &mut Context<Self>,
1055 ) -> Result<mpsc::UnboundedReceiver<Result<ThreadEvent>>>
1056 where
1057 T: Into<UserMessageContent>,
1058 {
1059 let model = self.model().context("No language model configured")?;
1060
1061 log::info!("Thread::send called with model: {:?}", model.name());
1062 self.advance_prompt_id();
1063
1064 let content = content.into_iter().map(Into::into).collect::<Vec<_>>();
1065 log::debug!("Thread::send content: {:?}", content);
1066
1067 self.messages
1068 .push(Message::User(UserMessage { id, content }));
1069 cx.notify();
1070
1071 log::info!("Total messages in thread: {}", self.messages.len());
1072 self.run_turn(cx)
1073 }
1074
1075 fn run_turn(
1076 &mut self,
1077 cx: &mut Context<Self>,
1078 ) -> Result<mpsc::UnboundedReceiver<Result<ThreadEvent>>> {
1079 self.cancel(cx);
1080
1081 let model = self.model.clone().context("No language model configured")?;
1082 let (events_tx, events_rx) = mpsc::unbounded::<Result<ThreadEvent>>();
1083 let event_stream = ThreadEventStream(events_tx);
1084 let message_ix = self.messages.len().saturating_sub(1);
1085 self.tool_use_limit_reached = false;
1086 self.summary = None;
1087 self.running_turn = Some(RunningTurn {
1088 event_stream: event_stream.clone(),
1089 _task: cx.spawn(async move |this, cx| {
1090 log::info!("Starting agent turn execution");
1091
1092 let turn_result: Result<()> = async {
1093 let mut intent = CompletionIntent::UserPrompt;
1094 loop {
1095 Self::stream_completion(&this, &model, intent, &event_stream, cx).await?;
1096
1097 let mut end_turn = true;
1098 this.update(cx, |this, cx| {
1099 // Generate title if needed.
1100 if this.title.is_none() && this.pending_title_generation.is_none() {
1101 this.generate_title(cx);
1102 }
1103
1104 // End the turn if the model didn't use tools.
1105 let message = this.pending_message.as_ref();
1106 end_turn =
1107 message.map_or(true, |message| message.tool_results.is_empty());
1108 this.flush_pending_message(cx);
1109 })?;
1110
1111 if this.read_with(cx, |this, _| this.tool_use_limit_reached)? {
1112 log::info!("Tool use limit reached, completing turn");
1113 return Err(language_model::ToolUseLimitReachedError.into());
1114 } else if end_turn {
1115 log::info!("No tool uses found, completing turn");
1116 return Ok(());
1117 } else {
1118 intent = CompletionIntent::ToolResults;
1119 }
1120 }
1121 }
1122 .await;
1123 _ = this.update(cx, |this, cx| this.flush_pending_message(cx));
1124
1125 match turn_result {
1126 Ok(()) => {
1127 log::info!("Turn execution completed");
1128 event_stream.send_stop(acp::StopReason::EndTurn);
1129 }
1130 Err(error) => {
1131 log::error!("Turn execution failed: {:?}", error);
1132 match error.downcast::<CompletionError>() {
1133 Ok(CompletionError::Refusal) => {
1134 event_stream.send_stop(acp::StopReason::Refusal);
1135 _ = this.update(cx, |this, _| this.messages.truncate(message_ix));
1136 }
1137 Ok(CompletionError::MaxTokens) => {
1138 event_stream.send_stop(acp::StopReason::MaxTokens);
1139 }
1140 Ok(CompletionError::Other(error)) | Err(error) => {
1141 event_stream.send_error(error);
1142 }
1143 }
1144 }
1145 }
1146
1147 _ = this.update(cx, |this, _| this.running_turn.take());
1148 }),
1149 });
1150 Ok(events_rx)
1151 }
1152
1153 async fn stream_completion(
1154 this: &WeakEntity<Self>,
1155 model: &Arc<dyn LanguageModel>,
1156 completion_intent: CompletionIntent,
1157 event_stream: &ThreadEventStream,
1158 cx: &mut AsyncApp,
1159 ) -> Result<()> {
1160 log::debug!("Stream completion started successfully");
1161 let request = this.update(cx, |this, cx| {
1162 this.build_completion_request(completion_intent, cx)
1163 })??;
1164
1165 let mut attempt = None;
1166 'retry: loop {
1167 telemetry::event!(
1168 "Agent Thread Completion",
1169 thread_id = this.read_with(cx, |this, _| this.id.to_string())?,
1170 prompt_id = this.read_with(cx, |this, _| this.prompt_id.to_string())?,
1171 model = model.telemetry_id(),
1172 model_provider = model.provider_id().to_string(),
1173 attempt
1174 );
1175
1176 log::info!(
1177 "Calling model.stream_completion, attempt {}",
1178 attempt.unwrap_or(0)
1179 );
1180 let mut events = model
1181 .stream_completion(request.clone(), cx)
1182 .await
1183 .map_err(|error| anyhow!(error))?;
1184 let mut tool_results = FuturesUnordered::new();
1185
1186 while let Some(event) = events.next().await {
1187 match event {
1188 Ok(event) => {
1189 log::trace!("Received completion event: {:?}", event);
1190 tool_results.extend(this.update(cx, |this, cx| {
1191 this.handle_streamed_completion_event(event, event_stream, cx)
1192 })??);
1193 }
1194 Err(error) => {
1195 let completion_mode =
1196 this.read_with(cx, |thread, _cx| thread.completion_mode())?;
1197 if completion_mode == CompletionMode::Normal {
1198 return Err(anyhow!(error))?;
1199 }
1200
1201 let Some(strategy) = Self::retry_strategy_for(&error) else {
1202 return Err(anyhow!(error))?;
1203 };
1204
1205 let max_attempts = match &strategy {
1206 RetryStrategy::ExponentialBackoff { max_attempts, .. } => *max_attempts,
1207 RetryStrategy::Fixed { max_attempts, .. } => *max_attempts,
1208 };
1209
1210 let attempt = attempt.get_or_insert(0u8);
1211
1212 *attempt += 1;
1213
1214 let attempt = *attempt;
1215 if attempt > max_attempts {
1216 return Err(anyhow!(error))?;
1217 }
1218
1219 let delay = match &strategy {
1220 RetryStrategy::ExponentialBackoff { initial_delay, .. } => {
1221 let delay_secs =
1222 initial_delay.as_secs() * 2u64.pow((attempt - 1) as u32);
1223 Duration::from_secs(delay_secs)
1224 }
1225 RetryStrategy::Fixed { delay, .. } => *delay,
1226 };
1227 log::debug!("Retry attempt {attempt} with delay {delay:?}");
1228
1229 event_stream.send_retry(acp_thread::RetryStatus {
1230 last_error: error.to_string().into(),
1231 attempt: attempt as usize,
1232 max_attempts: max_attempts as usize,
1233 started_at: Instant::now(),
1234 duration: delay,
1235 });
1236
1237 cx.background_executor().timer(delay).await;
1238 continue 'retry;
1239 }
1240 }
1241 }
1242
1243 while let Some(tool_result) = tool_results.next().await {
1244 log::info!("Tool finished {:?}", tool_result);
1245
1246 event_stream.update_tool_call_fields(
1247 &tool_result.tool_use_id,
1248 acp::ToolCallUpdateFields {
1249 status: Some(if tool_result.is_error {
1250 acp::ToolCallStatus::Failed
1251 } else {
1252 acp::ToolCallStatus::Completed
1253 }),
1254 raw_output: tool_result.output.clone(),
1255 ..Default::default()
1256 },
1257 );
1258 this.update(cx, |this, _cx| {
1259 this.pending_message()
1260 .tool_results
1261 .insert(tool_result.tool_use_id.clone(), tool_result);
1262 })?;
1263 }
1264
1265 return Ok(());
1266 }
1267 }
1268
1269 pub fn build_system_message(&self, cx: &App) -> LanguageModelRequestMessage {
1270 log::debug!("Building system message");
1271 let prompt = SystemPromptTemplate {
1272 project: self.project_context.read(cx),
1273 available_tools: self.tools.keys().cloned().collect(),
1274 }
1275 .render(&self.templates)
1276 .context("failed to build system prompt")
1277 .expect("Invalid template");
1278 log::debug!("System message built");
1279 LanguageModelRequestMessage {
1280 role: Role::System,
1281 content: vec![prompt.into()],
1282 cache: true,
1283 }
1284 }
1285
1286 /// A helper method that's called on every streamed completion event.
1287 /// Returns an optional tool result task, which the main agentic loop will
1288 /// send back to the model when it resolves.
1289 fn handle_streamed_completion_event(
1290 &mut self,
1291 event: LanguageModelCompletionEvent,
1292 event_stream: &ThreadEventStream,
1293 cx: &mut Context<Self>,
1294 ) -> Result<Option<Task<LanguageModelToolResult>>> {
1295 log::trace!("Handling streamed completion event: {:?}", event);
1296 use LanguageModelCompletionEvent::*;
1297
1298 match event {
1299 StartMessage { .. } => {
1300 self.flush_pending_message(cx);
1301 self.pending_message = Some(AgentMessage::default());
1302 }
1303 Text(new_text) => self.handle_text_event(new_text, event_stream, cx),
1304 Thinking { text, signature } => {
1305 self.handle_thinking_event(text, signature, event_stream, cx)
1306 }
1307 RedactedThinking { data } => self.handle_redacted_thinking_event(data, cx),
1308 ToolUse(tool_use) => {
1309 return Ok(self.handle_tool_use_event(tool_use, event_stream, cx));
1310 }
1311 ToolUseJsonParseError {
1312 id,
1313 tool_name,
1314 raw_input,
1315 json_parse_error,
1316 } => {
1317 return Ok(Some(Task::ready(
1318 self.handle_tool_use_json_parse_error_event(
1319 id,
1320 tool_name,
1321 raw_input,
1322 json_parse_error,
1323 ),
1324 )));
1325 }
1326 UsageUpdate(usage) => {
1327 telemetry::event!(
1328 "Agent Thread Completion Usage Updated",
1329 thread_id = self.id.to_string(),
1330 prompt_id = self.prompt_id.to_string(),
1331 model = self.model.as_ref().map(|m| m.telemetry_id()),
1332 model_provider = self.model.as_ref().map(|m| m.provider_id().to_string()),
1333 input_tokens = usage.input_tokens,
1334 output_tokens = usage.output_tokens,
1335 cache_creation_input_tokens = usage.cache_creation_input_tokens,
1336 cache_read_input_tokens = usage.cache_read_input_tokens,
1337 );
1338 self.update_token_usage(usage, cx);
1339 }
1340 StatusUpdate(CompletionRequestStatus::UsageUpdated { amount, limit }) => {
1341 self.update_model_request_usage(amount, limit, cx);
1342 }
1343 StatusUpdate(
1344 CompletionRequestStatus::Started
1345 | CompletionRequestStatus::Queued { .. }
1346 | CompletionRequestStatus::Failed { .. },
1347 ) => {}
1348 StatusUpdate(CompletionRequestStatus::ToolUseLimitReached) => {
1349 self.tool_use_limit_reached = true;
1350 }
1351 Stop(StopReason::Refusal) => return Err(CompletionError::Refusal.into()),
1352 Stop(StopReason::MaxTokens) => return Err(CompletionError::MaxTokens.into()),
1353 Stop(StopReason::ToolUse | StopReason::EndTurn) => {}
1354 }
1355
1356 Ok(None)
1357 }
1358
1359 fn handle_text_event(
1360 &mut self,
1361 new_text: String,
1362 event_stream: &ThreadEventStream,
1363 cx: &mut Context<Self>,
1364 ) {
1365 event_stream.send_text(&new_text);
1366
1367 let last_message = self.pending_message();
1368 if let Some(AgentMessageContent::Text(text)) = last_message.content.last_mut() {
1369 text.push_str(&new_text);
1370 } else {
1371 last_message
1372 .content
1373 .push(AgentMessageContent::Text(new_text));
1374 }
1375
1376 cx.notify();
1377 }
1378
1379 fn handle_thinking_event(
1380 &mut self,
1381 new_text: String,
1382 new_signature: Option<String>,
1383 event_stream: &ThreadEventStream,
1384 cx: &mut Context<Self>,
1385 ) {
1386 event_stream.send_thinking(&new_text);
1387
1388 let last_message = self.pending_message();
1389 if let Some(AgentMessageContent::Thinking { text, signature }) =
1390 last_message.content.last_mut()
1391 {
1392 text.push_str(&new_text);
1393 *signature = new_signature.or(signature.take());
1394 } else {
1395 last_message.content.push(AgentMessageContent::Thinking {
1396 text: new_text,
1397 signature: new_signature,
1398 });
1399 }
1400
1401 cx.notify();
1402 }
1403
1404 fn handle_redacted_thinking_event(&mut self, data: String, cx: &mut Context<Self>) {
1405 let last_message = self.pending_message();
1406 last_message
1407 .content
1408 .push(AgentMessageContent::RedactedThinking(data));
1409 cx.notify();
1410 }
1411
1412 fn handle_tool_use_event(
1413 &mut self,
1414 tool_use: LanguageModelToolUse,
1415 event_stream: &ThreadEventStream,
1416 cx: &mut Context<Self>,
1417 ) -> Option<Task<LanguageModelToolResult>> {
1418 cx.notify();
1419
1420 let tool = self.tools.get(tool_use.name.as_ref()).cloned();
1421 let mut title = SharedString::from(&tool_use.name);
1422 let mut kind = acp::ToolKind::Other;
1423 if let Some(tool) = tool.as_ref() {
1424 title = tool.initial_title(tool_use.input.clone());
1425 kind = tool.kind();
1426 }
1427
1428 // Ensure the last message ends in the current tool use
1429 let last_message = self.pending_message();
1430 let push_new_tool_use = last_message.content.last_mut().is_none_or(|content| {
1431 if let AgentMessageContent::ToolUse(last_tool_use) = content {
1432 if last_tool_use.id == tool_use.id {
1433 *last_tool_use = tool_use.clone();
1434 false
1435 } else {
1436 true
1437 }
1438 } else {
1439 true
1440 }
1441 });
1442
1443 if push_new_tool_use {
1444 event_stream.send_tool_call(&tool_use.id, title, kind, tool_use.input.clone());
1445 last_message
1446 .content
1447 .push(AgentMessageContent::ToolUse(tool_use.clone()));
1448 } else {
1449 event_stream.update_tool_call_fields(
1450 &tool_use.id,
1451 acp::ToolCallUpdateFields {
1452 title: Some(title.into()),
1453 kind: Some(kind),
1454 raw_input: Some(tool_use.input.clone()),
1455 ..Default::default()
1456 },
1457 );
1458 }
1459
1460 if !tool_use.is_input_complete {
1461 return None;
1462 }
1463
1464 let Some(tool) = tool else {
1465 let content = format!("No tool named {} exists", tool_use.name);
1466 return Some(Task::ready(LanguageModelToolResult {
1467 content: LanguageModelToolResultContent::Text(Arc::from(content)),
1468 tool_use_id: tool_use.id,
1469 tool_name: tool_use.name,
1470 is_error: true,
1471 output: None,
1472 }));
1473 };
1474
1475 let fs = self.project.read(cx).fs().clone();
1476 let tool_event_stream =
1477 ToolCallEventStream::new(tool_use.id.clone(), event_stream.clone(), Some(fs));
1478 tool_event_stream.update_fields(acp::ToolCallUpdateFields {
1479 status: Some(acp::ToolCallStatus::InProgress),
1480 ..Default::default()
1481 });
1482 let supports_images = self.model().is_some_and(|model| model.supports_images());
1483 let tool_result = tool.run(tool_use.input, tool_event_stream, cx);
1484 log::info!("Running tool {}", tool_use.name);
1485 Some(cx.foreground_executor().spawn(async move {
1486 let tool_result = tool_result.await.and_then(|output| {
1487 if let LanguageModelToolResultContent::Image(_) = &output.llm_output
1488 && !supports_images
1489 {
1490 return Err(anyhow!(
1491 "Attempted to read an image, but this model doesn't support it.",
1492 ));
1493 }
1494 Ok(output)
1495 });
1496
1497 match tool_result {
1498 Ok(output) => LanguageModelToolResult {
1499 tool_use_id: tool_use.id,
1500 tool_name: tool_use.name,
1501 is_error: false,
1502 content: output.llm_output,
1503 output: Some(output.raw_output),
1504 },
1505 Err(error) => LanguageModelToolResult {
1506 tool_use_id: tool_use.id,
1507 tool_name: tool_use.name,
1508 is_error: true,
1509 content: LanguageModelToolResultContent::Text(Arc::from(error.to_string())),
1510 output: None,
1511 },
1512 }
1513 }))
1514 }
1515
1516 fn handle_tool_use_json_parse_error_event(
1517 &mut self,
1518 tool_use_id: LanguageModelToolUseId,
1519 tool_name: Arc<str>,
1520 raw_input: Arc<str>,
1521 json_parse_error: String,
1522 ) -> LanguageModelToolResult {
1523 let tool_output = format!("Error parsing input JSON: {json_parse_error}");
1524 LanguageModelToolResult {
1525 tool_use_id,
1526 tool_name,
1527 is_error: true,
1528 content: LanguageModelToolResultContent::Text(tool_output.into()),
1529 output: Some(serde_json::Value::String(raw_input.to_string())),
1530 }
1531 }
1532
1533 fn update_model_request_usage(&self, amount: usize, limit: UsageLimit, cx: &mut Context<Self>) {
1534 self.project
1535 .read(cx)
1536 .user_store()
1537 .update(cx, |user_store, cx| {
1538 user_store.update_model_request_usage(
1539 ModelRequestUsage(RequestUsage {
1540 amount: amount as i32,
1541 limit,
1542 }),
1543 cx,
1544 )
1545 });
1546 }
1547
1548 pub fn title(&self) -> SharedString {
1549 self.title.clone().unwrap_or("New Thread".into())
1550 }
1551
1552 pub fn summary(&mut self, cx: &mut Context<Self>) -> Task<Result<SharedString>> {
1553 if let Some(summary) = self.summary.as_ref() {
1554 return Task::ready(Ok(summary.clone()));
1555 }
1556 let Some(model) = self.summarization_model.clone() else {
1557 return Task::ready(Err(anyhow!("No summarization model available")));
1558 };
1559 let mut request = LanguageModelRequest {
1560 intent: Some(CompletionIntent::ThreadContextSummarization),
1561 temperature: AgentSettings::temperature_for_model(&model, cx),
1562 ..Default::default()
1563 };
1564
1565 for message in &self.messages {
1566 request.messages.extend(message.to_request());
1567 }
1568
1569 request.messages.push(LanguageModelRequestMessage {
1570 role: Role::User,
1571 content: vec![SUMMARIZE_THREAD_DETAILED_PROMPT.into()],
1572 cache: false,
1573 });
1574 cx.spawn(async move |this, cx| {
1575 let mut summary = String::new();
1576 let mut messages = model.stream_completion(request, cx).await?;
1577 while let Some(event) = messages.next().await {
1578 let event = event?;
1579 let text = match event {
1580 LanguageModelCompletionEvent::Text(text) => text,
1581 LanguageModelCompletionEvent::StatusUpdate(
1582 CompletionRequestStatus::UsageUpdated { amount, limit },
1583 ) => {
1584 this.update(cx, |thread, cx| {
1585 thread.update_model_request_usage(amount, limit, cx);
1586 })?;
1587 continue;
1588 }
1589 _ => continue,
1590 };
1591
1592 let mut lines = text.lines();
1593 summary.extend(lines.next());
1594 }
1595
1596 log::info!("Setting summary: {}", summary);
1597 let summary = SharedString::from(summary);
1598
1599 this.update(cx, |this, cx| {
1600 this.summary = Some(summary.clone());
1601 cx.notify()
1602 })?;
1603
1604 Ok(summary)
1605 })
1606 }
1607
1608 fn generate_title(&mut self, cx: &mut Context<Self>) {
1609 let Some(model) = self.summarization_model.clone() else {
1610 return;
1611 };
1612
1613 log::info!(
1614 "Generating title with model: {:?}",
1615 self.summarization_model.as_ref().map(|model| model.name())
1616 );
1617 let mut request = LanguageModelRequest {
1618 intent: Some(CompletionIntent::ThreadSummarization),
1619 temperature: AgentSettings::temperature_for_model(&model, cx),
1620 ..Default::default()
1621 };
1622
1623 for message in &self.messages {
1624 request.messages.extend(message.to_request());
1625 }
1626
1627 request.messages.push(LanguageModelRequestMessage {
1628 role: Role::User,
1629 content: vec![SUMMARIZE_THREAD_PROMPT.into()],
1630 cache: false,
1631 });
1632 self.pending_title_generation = Some(cx.spawn(async move |this, cx| {
1633 let mut title = String::new();
1634
1635 let generate = async {
1636 let mut messages = model.stream_completion(request, cx).await?;
1637 while let Some(event) = messages.next().await {
1638 let event = event?;
1639 let text = match event {
1640 LanguageModelCompletionEvent::Text(text) => text,
1641 LanguageModelCompletionEvent::StatusUpdate(
1642 CompletionRequestStatus::UsageUpdated { amount, limit },
1643 ) => {
1644 this.update(cx, |thread, cx| {
1645 thread.update_model_request_usage(amount, limit, cx);
1646 })?;
1647 continue;
1648 }
1649 _ => continue,
1650 };
1651
1652 let mut lines = text.lines();
1653 title.extend(lines.next());
1654
1655 // Stop if the LLM generated multiple lines.
1656 if lines.next().is_some() {
1657 break;
1658 }
1659 }
1660 anyhow::Ok(())
1661 };
1662
1663 if generate.await.context("failed to generate title").is_ok() {
1664 _ = this.update(cx, |this, cx| this.set_title(title.into(), cx));
1665 }
1666 _ = this.update(cx, |this, _| this.pending_title_generation = None);
1667 }));
1668 }
1669
1670 pub fn set_title(&mut self, title: SharedString, cx: &mut Context<Self>) {
1671 self.pending_title_generation = None;
1672 if Some(&title) != self.title.as_ref() {
1673 self.title = Some(title);
1674 cx.emit(TitleUpdated);
1675 cx.notify();
1676 }
1677 }
1678
1679 fn last_user_message(&self) -> Option<&UserMessage> {
1680 self.messages
1681 .iter()
1682 .rev()
1683 .find_map(|message| match message {
1684 Message::User(user_message) => Some(user_message),
1685 Message::Agent(_) => None,
1686 Message::Resume => None,
1687 })
1688 }
1689
1690 fn pending_message(&mut self) -> &mut AgentMessage {
1691 self.pending_message.get_or_insert_default()
1692 }
1693
1694 fn flush_pending_message(&mut self, cx: &mut Context<Self>) {
1695 let Some(mut message) = self.pending_message.take() else {
1696 return;
1697 };
1698
1699 for content in &message.content {
1700 let AgentMessageContent::ToolUse(tool_use) = content else {
1701 continue;
1702 };
1703
1704 if !message.tool_results.contains_key(&tool_use.id) {
1705 message.tool_results.insert(
1706 tool_use.id.clone(),
1707 LanguageModelToolResult {
1708 tool_use_id: tool_use.id.clone(),
1709 tool_name: tool_use.name.clone(),
1710 is_error: true,
1711 content: LanguageModelToolResultContent::Text(TOOL_CANCELED_MESSAGE.into()),
1712 output: None,
1713 },
1714 );
1715 }
1716 }
1717
1718 self.messages.push(Message::Agent(message));
1719 self.updated_at = Utc::now();
1720 self.summary = None;
1721 cx.notify()
1722 }
1723
1724 pub(crate) fn build_completion_request(
1725 &self,
1726 completion_intent: CompletionIntent,
1727 cx: &mut App,
1728 ) -> Result<LanguageModelRequest> {
1729 let model = self.model().context("No language model configured")?;
1730
1731 log::debug!("Building completion request");
1732 log::debug!("Completion intent: {:?}", completion_intent);
1733 log::debug!("Completion mode: {:?}", self.completion_mode);
1734
1735 let messages = self.build_request_messages(cx);
1736 log::info!("Request will include {} messages", messages.len());
1737
1738 let tools = if let Some(tools) = self.tools(cx).log_err() {
1739 tools
1740 .filter_map(|tool| {
1741 let tool_name = tool.name().to_string();
1742 log::trace!("Including tool: {}", tool_name);
1743 Some(LanguageModelRequestTool {
1744 name: tool_name,
1745 description: tool.description().to_string(),
1746 input_schema: tool.input_schema(model.tool_input_format()).log_err()?,
1747 })
1748 })
1749 .collect()
1750 } else {
1751 Vec::new()
1752 };
1753
1754 log::info!("Request includes {} tools", tools.len());
1755
1756 let request = LanguageModelRequest {
1757 thread_id: Some(self.id.to_string()),
1758 prompt_id: Some(self.prompt_id.to_string()),
1759 intent: Some(completion_intent),
1760 mode: Some(self.completion_mode.into()),
1761 messages,
1762 tools,
1763 tool_choice: None,
1764 stop: Vec::new(),
1765 temperature: AgentSettings::temperature_for_model(model, cx),
1766 thinking_allowed: true,
1767 };
1768
1769 log::debug!("Completion request built successfully");
1770 Ok(request)
1771 }
1772
1773 fn tools<'a>(&'a self, cx: &'a App) -> Result<impl Iterator<Item = &'a Arc<dyn AnyAgentTool>>> {
1774 let model = self.model().context("No language model configured")?;
1775
1776 let profile = AgentSettings::get_global(cx)
1777 .profiles
1778 .get(&self.profile_id)
1779 .context("profile not found")?;
1780 let provider_id = model.provider_id();
1781
1782 Ok(self
1783 .tools
1784 .iter()
1785 .filter(move |(_, tool)| tool.supported_provider(&provider_id))
1786 .filter_map(|(tool_name, tool)| {
1787 if profile.is_tool_enabled(tool_name) {
1788 Some(tool)
1789 } else {
1790 None
1791 }
1792 })
1793 .chain(self.context_server_registry.read(cx).servers().flat_map(
1794 |(server_id, tools)| {
1795 tools.iter().filter_map(|(tool_name, tool)| {
1796 if profile.is_context_server_tool_enabled(&server_id.0, tool_name) {
1797 Some(tool)
1798 } else {
1799 None
1800 }
1801 })
1802 },
1803 )))
1804 }
1805
1806 fn build_request_messages(&self, cx: &App) -> Vec<LanguageModelRequestMessage> {
1807 log::trace!(
1808 "Building request messages from {} thread messages",
1809 self.messages.len()
1810 );
1811 let mut messages = vec![self.build_system_message(cx)];
1812 for message in &self.messages {
1813 messages.extend(message.to_request());
1814 }
1815
1816 if let Some(message) = self.pending_message.as_ref() {
1817 messages.extend(message.to_request());
1818 }
1819
1820 if let Some(last_user_message) = messages
1821 .iter_mut()
1822 .rev()
1823 .find(|message| message.role == Role::User)
1824 {
1825 last_user_message.cache = true;
1826 }
1827
1828 messages
1829 }
1830
1831 pub fn to_markdown(&self) -> String {
1832 let mut markdown = String::new();
1833 for (ix, message) in self.messages.iter().enumerate() {
1834 if ix > 0 {
1835 markdown.push('\n');
1836 }
1837 markdown.push_str(&message.to_markdown());
1838 }
1839
1840 if let Some(message) = self.pending_message.as_ref() {
1841 markdown.push('\n');
1842 markdown.push_str(&message.to_markdown());
1843 }
1844
1845 markdown
1846 }
1847
1848 fn advance_prompt_id(&mut self) {
1849 self.prompt_id = PromptId::new();
1850 }
1851
1852 fn retry_strategy_for(error: &LanguageModelCompletionError) -> Option<RetryStrategy> {
1853 use LanguageModelCompletionError::*;
1854 use http_client::StatusCode;
1855
1856 // General strategy here:
1857 // - If retrying won't help (e.g. invalid API key or payload too large), return None so we don't retry at all.
1858 // - If it's a time-based issue (e.g. server overloaded, rate limit exceeded), retry up to 4 times with exponential backoff.
1859 // - If it's an issue that *might* be fixed by retrying (e.g. internal server error), retry up to 3 times.
1860 match error {
1861 HttpResponseError {
1862 status_code: StatusCode::TOO_MANY_REQUESTS,
1863 ..
1864 } => Some(RetryStrategy::ExponentialBackoff {
1865 initial_delay: BASE_RETRY_DELAY,
1866 max_attempts: MAX_RETRY_ATTEMPTS,
1867 }),
1868 ServerOverloaded { retry_after, .. } | RateLimitExceeded { retry_after, .. } => {
1869 Some(RetryStrategy::Fixed {
1870 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
1871 max_attempts: MAX_RETRY_ATTEMPTS,
1872 })
1873 }
1874 UpstreamProviderError {
1875 status,
1876 retry_after,
1877 ..
1878 } => match *status {
1879 StatusCode::TOO_MANY_REQUESTS | StatusCode::SERVICE_UNAVAILABLE => {
1880 Some(RetryStrategy::Fixed {
1881 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
1882 max_attempts: MAX_RETRY_ATTEMPTS,
1883 })
1884 }
1885 StatusCode::INTERNAL_SERVER_ERROR => Some(RetryStrategy::Fixed {
1886 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
1887 // Internal Server Error could be anything, retry up to 3 times.
1888 max_attempts: 3,
1889 }),
1890 status => {
1891 // There is no StatusCode variant for the unofficial HTTP 529 ("The service is overloaded"),
1892 // but we frequently get them in practice. See https://http.dev/529
1893 if status.as_u16() == 529 {
1894 Some(RetryStrategy::Fixed {
1895 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
1896 max_attempts: MAX_RETRY_ATTEMPTS,
1897 })
1898 } else {
1899 Some(RetryStrategy::Fixed {
1900 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
1901 max_attempts: 2,
1902 })
1903 }
1904 }
1905 },
1906 ApiInternalServerError { .. } => Some(RetryStrategy::Fixed {
1907 delay: BASE_RETRY_DELAY,
1908 max_attempts: 3,
1909 }),
1910 ApiReadResponseError { .. }
1911 | HttpSend { .. }
1912 | DeserializeResponse { .. }
1913 | BadRequestFormat { .. } => Some(RetryStrategy::Fixed {
1914 delay: BASE_RETRY_DELAY,
1915 max_attempts: 3,
1916 }),
1917 // Retrying these errors definitely shouldn't help.
1918 HttpResponseError {
1919 status_code:
1920 StatusCode::PAYLOAD_TOO_LARGE | StatusCode::FORBIDDEN | StatusCode::UNAUTHORIZED,
1921 ..
1922 }
1923 | AuthenticationError { .. }
1924 | PermissionError { .. }
1925 | NoApiKey { .. }
1926 | ApiEndpointNotFound { .. }
1927 | PromptTooLarge { .. } => None,
1928 // These errors might be transient, so retry them
1929 SerializeRequest { .. } | BuildRequestBody { .. } => Some(RetryStrategy::Fixed {
1930 delay: BASE_RETRY_DELAY,
1931 max_attempts: 1,
1932 }),
1933 // Retry all other 4xx and 5xx errors once.
1934 HttpResponseError { status_code, .. }
1935 if status_code.is_client_error() || status_code.is_server_error() =>
1936 {
1937 Some(RetryStrategy::Fixed {
1938 delay: BASE_RETRY_DELAY,
1939 max_attempts: 3,
1940 })
1941 }
1942 Other(err)
1943 if err.is::<language_model::PaymentRequiredError>()
1944 || err.is::<language_model::ModelRequestLimitReachedError>() =>
1945 {
1946 // Retrying won't help for Payment Required or Model Request Limit errors (where
1947 // the user must upgrade to usage-based billing to get more requests, or else wait
1948 // for a significant amount of time for the request limit to reset).
1949 None
1950 }
1951 // Conservatively assume that any other errors are non-retryable
1952 HttpResponseError { .. } | Other(..) => Some(RetryStrategy::Fixed {
1953 delay: BASE_RETRY_DELAY,
1954 max_attempts: 2,
1955 }),
1956 }
1957 }
1958}
1959
1960struct RunningTurn {
1961 /// Holds the task that handles agent interaction until the end of the turn.
1962 /// Survives across multiple requests as the model performs tool calls and
1963 /// we run tools, report their results.
1964 _task: Task<()>,
1965 /// The current event stream for the running turn. Used to report a final
1966 /// cancellation event if we cancel the turn.
1967 event_stream: ThreadEventStream,
1968}
1969
1970impl RunningTurn {
1971 fn cancel(self) {
1972 log::debug!("Cancelling in progress turn");
1973 self.event_stream.send_canceled();
1974 }
1975}
1976
1977pub struct TokenUsageUpdated(pub Option<acp_thread::TokenUsage>);
1978
1979impl EventEmitter<TokenUsageUpdated> for Thread {}
1980
1981pub struct TitleUpdated;
1982
1983impl EventEmitter<TitleUpdated> for Thread {}
1984
1985pub trait AgentTool
1986where
1987 Self: 'static + Sized,
1988{
1989 type Input: for<'de> Deserialize<'de> + Serialize + JsonSchema;
1990 type Output: for<'de> Deserialize<'de> + Serialize + Into<LanguageModelToolResultContent>;
1991
1992 fn name(&self) -> SharedString;
1993
1994 fn description(&self) -> SharedString {
1995 let schema = schemars::schema_for!(Self::Input);
1996 SharedString::new(
1997 schema
1998 .get("description")
1999 .and_then(|description| description.as_str())
2000 .unwrap_or_default(),
2001 )
2002 }
2003
2004 fn kind(&self) -> acp::ToolKind;
2005
2006 /// The initial tool title to display. Can be updated during the tool run.
2007 fn initial_title(&self, input: Result<Self::Input, serde_json::Value>) -> SharedString;
2008
2009 /// Returns the JSON schema that describes the tool's input.
2010 fn input_schema(&self, format: LanguageModelToolSchemaFormat) -> Schema {
2011 crate::tool_schema::root_schema_for::<Self::Input>(format)
2012 }
2013
2014 /// Some tools rely on a provider for the underlying billing or other reasons.
2015 /// Allow the tool to check if they are compatible, or should be filtered out.
2016 fn supported_provider(&self, _provider: &LanguageModelProviderId) -> bool {
2017 true
2018 }
2019
2020 /// Runs the tool with the provided input.
2021 fn run(
2022 self: Arc<Self>,
2023 input: Self::Input,
2024 event_stream: ToolCallEventStream,
2025 cx: &mut App,
2026 ) -> Task<Result<Self::Output>>;
2027
2028 /// Emits events for a previous execution of the tool.
2029 fn replay(
2030 &self,
2031 _input: Self::Input,
2032 _output: Self::Output,
2033 _event_stream: ToolCallEventStream,
2034 _cx: &mut App,
2035 ) -> Result<()> {
2036 Ok(())
2037 }
2038
2039 fn erase(self) -> Arc<dyn AnyAgentTool> {
2040 Arc::new(Erased(Arc::new(self)))
2041 }
2042}
2043
2044pub struct Erased<T>(T);
2045
2046pub struct AgentToolOutput {
2047 pub llm_output: LanguageModelToolResultContent,
2048 pub raw_output: serde_json::Value,
2049}
2050
2051pub trait AnyAgentTool {
2052 fn name(&self) -> SharedString;
2053 fn description(&self) -> SharedString;
2054 fn kind(&self) -> acp::ToolKind;
2055 fn initial_title(&self, input: serde_json::Value) -> SharedString;
2056 fn input_schema(&self, format: LanguageModelToolSchemaFormat) -> Result<serde_json::Value>;
2057 fn supported_provider(&self, _provider: &LanguageModelProviderId) -> bool {
2058 true
2059 }
2060 fn run(
2061 self: Arc<Self>,
2062 input: serde_json::Value,
2063 event_stream: ToolCallEventStream,
2064 cx: &mut App,
2065 ) -> Task<Result<AgentToolOutput>>;
2066 fn replay(
2067 &self,
2068 input: serde_json::Value,
2069 output: serde_json::Value,
2070 event_stream: ToolCallEventStream,
2071 cx: &mut App,
2072 ) -> Result<()>;
2073}
2074
2075impl<T> AnyAgentTool for Erased<Arc<T>>
2076where
2077 T: AgentTool,
2078{
2079 fn name(&self) -> SharedString {
2080 self.0.name()
2081 }
2082
2083 fn description(&self) -> SharedString {
2084 self.0.description()
2085 }
2086
2087 fn kind(&self) -> agent_client_protocol::ToolKind {
2088 self.0.kind()
2089 }
2090
2091 fn initial_title(&self, input: serde_json::Value) -> SharedString {
2092 let parsed_input = serde_json::from_value(input.clone()).map_err(|_| input);
2093 self.0.initial_title(parsed_input)
2094 }
2095
2096 fn input_schema(&self, format: LanguageModelToolSchemaFormat) -> Result<serde_json::Value> {
2097 let mut json = serde_json::to_value(self.0.input_schema(format))?;
2098 adapt_schema_to_format(&mut json, format)?;
2099 Ok(json)
2100 }
2101
2102 fn supported_provider(&self, provider: &LanguageModelProviderId) -> bool {
2103 self.0.supported_provider(provider)
2104 }
2105
2106 fn run(
2107 self: Arc<Self>,
2108 input: serde_json::Value,
2109 event_stream: ToolCallEventStream,
2110 cx: &mut App,
2111 ) -> Task<Result<AgentToolOutput>> {
2112 cx.spawn(async move |cx| {
2113 let input = serde_json::from_value(input)?;
2114 let output = cx
2115 .update(|cx| self.0.clone().run(input, event_stream, cx))?
2116 .await?;
2117 let raw_output = serde_json::to_value(&output)?;
2118 Ok(AgentToolOutput {
2119 llm_output: output.into(),
2120 raw_output,
2121 })
2122 })
2123 }
2124
2125 fn replay(
2126 &self,
2127 input: serde_json::Value,
2128 output: serde_json::Value,
2129 event_stream: ToolCallEventStream,
2130 cx: &mut App,
2131 ) -> Result<()> {
2132 let input = serde_json::from_value(input)?;
2133 let output = serde_json::from_value(output)?;
2134 self.0.replay(input, output, event_stream, cx)
2135 }
2136}
2137
2138#[derive(Clone)]
2139struct ThreadEventStream(mpsc::UnboundedSender<Result<ThreadEvent>>);
2140
2141impl ThreadEventStream {
2142 fn send_user_message(&self, message: &UserMessage) {
2143 self.0
2144 .unbounded_send(Ok(ThreadEvent::UserMessage(message.clone())))
2145 .ok();
2146 }
2147
2148 fn send_text(&self, text: &str) {
2149 self.0
2150 .unbounded_send(Ok(ThreadEvent::AgentText(text.to_string())))
2151 .ok();
2152 }
2153
2154 fn send_thinking(&self, text: &str) {
2155 self.0
2156 .unbounded_send(Ok(ThreadEvent::AgentThinking(text.to_string())))
2157 .ok();
2158 }
2159
2160 fn send_tool_call(
2161 &self,
2162 id: &LanguageModelToolUseId,
2163 title: SharedString,
2164 kind: acp::ToolKind,
2165 input: serde_json::Value,
2166 ) {
2167 self.0
2168 .unbounded_send(Ok(ThreadEvent::ToolCall(Self::initial_tool_call(
2169 id,
2170 title.to_string(),
2171 kind,
2172 input,
2173 ))))
2174 .ok();
2175 }
2176
2177 fn initial_tool_call(
2178 id: &LanguageModelToolUseId,
2179 title: String,
2180 kind: acp::ToolKind,
2181 input: serde_json::Value,
2182 ) -> acp::ToolCall {
2183 acp::ToolCall {
2184 id: acp::ToolCallId(id.to_string().into()),
2185 title,
2186 kind,
2187 status: acp::ToolCallStatus::Pending,
2188 content: vec![],
2189 locations: vec![],
2190 raw_input: Some(input),
2191 raw_output: None,
2192 }
2193 }
2194
2195 fn update_tool_call_fields(
2196 &self,
2197 tool_use_id: &LanguageModelToolUseId,
2198 fields: acp::ToolCallUpdateFields,
2199 ) {
2200 self.0
2201 .unbounded_send(Ok(ThreadEvent::ToolCallUpdate(
2202 acp::ToolCallUpdate {
2203 id: acp::ToolCallId(tool_use_id.to_string().into()),
2204 fields,
2205 }
2206 .into(),
2207 )))
2208 .ok();
2209 }
2210
2211 fn send_retry(&self, status: acp_thread::RetryStatus) {
2212 self.0.unbounded_send(Ok(ThreadEvent::Retry(status))).ok();
2213 }
2214
2215 fn send_stop(&self, reason: acp::StopReason) {
2216 self.0.unbounded_send(Ok(ThreadEvent::Stop(reason))).ok();
2217 }
2218
2219 fn send_canceled(&self) {
2220 self.0
2221 .unbounded_send(Ok(ThreadEvent::Stop(acp::StopReason::Cancelled)))
2222 .ok();
2223 }
2224
2225 fn send_error(&self, error: impl Into<anyhow::Error>) {
2226 self.0.unbounded_send(Err(error.into())).ok();
2227 }
2228}
2229
2230#[derive(Clone)]
2231pub struct ToolCallEventStream {
2232 tool_use_id: LanguageModelToolUseId,
2233 stream: ThreadEventStream,
2234 fs: Option<Arc<dyn Fs>>,
2235}
2236
2237impl ToolCallEventStream {
2238 #[cfg(test)]
2239 pub fn test() -> (Self, ToolCallEventStreamReceiver) {
2240 let (events_tx, events_rx) = mpsc::unbounded::<Result<ThreadEvent>>();
2241
2242 let stream = ToolCallEventStream::new("test_id".into(), ThreadEventStream(events_tx), None);
2243
2244 (stream, ToolCallEventStreamReceiver(events_rx))
2245 }
2246
2247 fn new(
2248 tool_use_id: LanguageModelToolUseId,
2249 stream: ThreadEventStream,
2250 fs: Option<Arc<dyn Fs>>,
2251 ) -> Self {
2252 Self {
2253 tool_use_id,
2254 stream,
2255 fs,
2256 }
2257 }
2258
2259 pub fn update_fields(&self, fields: acp::ToolCallUpdateFields) {
2260 self.stream
2261 .update_tool_call_fields(&self.tool_use_id, fields);
2262 }
2263
2264 pub fn update_diff(&self, diff: Entity<acp_thread::Diff>) {
2265 self.stream
2266 .0
2267 .unbounded_send(Ok(ThreadEvent::ToolCallUpdate(
2268 acp_thread::ToolCallUpdateDiff {
2269 id: acp::ToolCallId(self.tool_use_id.to_string().into()),
2270 diff,
2271 }
2272 .into(),
2273 )))
2274 .ok();
2275 }
2276
2277 pub fn update_terminal(&self, terminal: Entity<acp_thread::Terminal>) {
2278 self.stream
2279 .0
2280 .unbounded_send(Ok(ThreadEvent::ToolCallUpdate(
2281 acp_thread::ToolCallUpdateTerminal {
2282 id: acp::ToolCallId(self.tool_use_id.to_string().into()),
2283 terminal,
2284 }
2285 .into(),
2286 )))
2287 .ok();
2288 }
2289
2290 pub fn authorize(&self, title: impl Into<String>, cx: &mut App) -> Task<Result<()>> {
2291 if agent_settings::AgentSettings::get_global(cx).always_allow_tool_actions {
2292 return Task::ready(Ok(()));
2293 }
2294
2295 let (response_tx, response_rx) = oneshot::channel();
2296 self.stream
2297 .0
2298 .unbounded_send(Ok(ThreadEvent::ToolCallAuthorization(
2299 ToolCallAuthorization {
2300 tool_call: acp::ToolCallUpdate {
2301 id: acp::ToolCallId(self.tool_use_id.to_string().into()),
2302 fields: acp::ToolCallUpdateFields {
2303 title: Some(title.into()),
2304 ..Default::default()
2305 },
2306 },
2307 options: vec![
2308 acp::PermissionOption {
2309 id: acp::PermissionOptionId("always_allow".into()),
2310 name: "Always Allow".into(),
2311 kind: acp::PermissionOptionKind::AllowAlways,
2312 },
2313 acp::PermissionOption {
2314 id: acp::PermissionOptionId("allow".into()),
2315 name: "Allow".into(),
2316 kind: acp::PermissionOptionKind::AllowOnce,
2317 },
2318 acp::PermissionOption {
2319 id: acp::PermissionOptionId("deny".into()),
2320 name: "Deny".into(),
2321 kind: acp::PermissionOptionKind::RejectOnce,
2322 },
2323 ],
2324 response: response_tx,
2325 },
2326 )))
2327 .ok();
2328 let fs = self.fs.clone();
2329 cx.spawn(async move |cx| match response_rx.await?.0.as_ref() {
2330 "always_allow" => {
2331 if let Some(fs) = fs.clone() {
2332 cx.update(|cx| {
2333 update_settings_file::<AgentSettings>(fs, cx, |settings, _| {
2334 settings.set_always_allow_tool_actions(true);
2335 });
2336 })?;
2337 }
2338
2339 Ok(())
2340 }
2341 "allow" => Ok(()),
2342 _ => Err(anyhow!("Permission to run tool denied by user")),
2343 })
2344 }
2345}
2346
2347#[cfg(test)]
2348pub struct ToolCallEventStreamReceiver(mpsc::UnboundedReceiver<Result<ThreadEvent>>);
2349
2350#[cfg(test)]
2351impl ToolCallEventStreamReceiver {
2352 pub async fn expect_authorization(&mut self) -> ToolCallAuthorization {
2353 let event = self.0.next().await;
2354 if let Some(Ok(ThreadEvent::ToolCallAuthorization(auth))) = event {
2355 auth
2356 } else {
2357 panic!("Expected ToolCallAuthorization but got: {:?}", event);
2358 }
2359 }
2360
2361 pub async fn expect_terminal(&mut self) -> Entity<acp_thread::Terminal> {
2362 let event = self.0.next().await;
2363 if let Some(Ok(ThreadEvent::ToolCallUpdate(acp_thread::ToolCallUpdate::UpdateTerminal(
2364 update,
2365 )))) = event
2366 {
2367 update.terminal
2368 } else {
2369 panic!("Expected terminal but got: {:?}", event);
2370 }
2371 }
2372}
2373
2374#[cfg(test)]
2375impl std::ops::Deref for ToolCallEventStreamReceiver {
2376 type Target = mpsc::UnboundedReceiver<Result<ThreadEvent>>;
2377
2378 fn deref(&self) -> &Self::Target {
2379 &self.0
2380 }
2381}
2382
2383#[cfg(test)]
2384impl std::ops::DerefMut for ToolCallEventStreamReceiver {
2385 fn deref_mut(&mut self) -> &mut Self::Target {
2386 &mut self.0
2387 }
2388}
2389
2390impl From<&str> for UserMessageContent {
2391 fn from(text: &str) -> Self {
2392 Self::Text(text.into())
2393 }
2394}
2395
2396impl From<acp::ContentBlock> for UserMessageContent {
2397 fn from(value: acp::ContentBlock) -> Self {
2398 match value {
2399 acp::ContentBlock::Text(text_content) => Self::Text(text_content.text),
2400 acp::ContentBlock::Image(image_content) => Self::Image(convert_image(image_content)),
2401 acp::ContentBlock::Audio(_) => {
2402 // TODO
2403 Self::Text("[audio]".to_string())
2404 }
2405 acp::ContentBlock::ResourceLink(resource_link) => {
2406 match MentionUri::parse(&resource_link.uri) {
2407 Ok(uri) => Self::Mention {
2408 uri,
2409 content: String::new(),
2410 },
2411 Err(err) => {
2412 log::error!("Failed to parse mention link: {}", err);
2413 Self::Text(format!("[{}]({})", resource_link.name, resource_link.uri))
2414 }
2415 }
2416 }
2417 acp::ContentBlock::Resource(resource) => match resource.resource {
2418 acp::EmbeddedResourceResource::TextResourceContents(resource) => {
2419 match MentionUri::parse(&resource.uri) {
2420 Ok(uri) => Self::Mention {
2421 uri,
2422 content: resource.text,
2423 },
2424 Err(err) => {
2425 log::error!("Failed to parse mention link: {}", err);
2426 Self::Text(
2427 MarkdownCodeBlock {
2428 tag: &resource.uri,
2429 text: &resource.text,
2430 }
2431 .to_string(),
2432 )
2433 }
2434 }
2435 }
2436 acp::EmbeddedResourceResource::BlobResourceContents(_) => {
2437 // TODO
2438 Self::Text("[blob]".to_string())
2439 }
2440 },
2441 }
2442 }
2443}
2444
2445impl From<UserMessageContent> for acp::ContentBlock {
2446 fn from(content: UserMessageContent) -> Self {
2447 match content {
2448 UserMessageContent::Text(text) => acp::ContentBlock::Text(acp::TextContent {
2449 text,
2450 annotations: None,
2451 }),
2452 UserMessageContent::Image(image) => acp::ContentBlock::Image(acp::ImageContent {
2453 data: image.source.to_string(),
2454 mime_type: "image/png".to_string(),
2455 annotations: None,
2456 uri: None,
2457 }),
2458 UserMessageContent::Mention { uri, content } => {
2459 acp::ContentBlock::Resource(acp::EmbeddedResource {
2460 resource: acp::EmbeddedResourceResource::TextResourceContents(
2461 acp::TextResourceContents {
2462 mime_type: None,
2463 text: content,
2464 uri: uri.to_uri().to_string(),
2465 },
2466 ),
2467 annotations: None,
2468 })
2469 }
2470 }
2471 }
2472}
2473
2474fn convert_image(image_content: acp::ImageContent) -> LanguageModelImage {
2475 LanguageModelImage {
2476 source: image_content.data.into(),
2477 // TODO: make this optional?
2478 size: gpui::Size::new(0.into(), 0.into()),
2479 }
2480}