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, AgentProfileSettings, AgentSettings, CompletionMode,
13 SUMMARIZE_THREAD_DETAILED_PROMPT, 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, HashSet, 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 ops::RangeInclusive,
51 path::Path,
52 rc::Rc,
53 sync::Arc,
54 time::{Duration, Instant},
55};
56use std::{fmt::Write, path::PathBuf};
57use util::{ResultExt, debug_panic, markdown::MarkdownCodeBlock};
58use uuid::Uuid;
59
60const TOOL_CANCELED_MESSAGE: &str = "Tool canceled by user";
61pub const MAX_TOOL_NAME_LENGTH: usize = 64;
62
63/// The ID of the user prompt that initiated a request.
64///
65/// This equates to the user physically submitting a message to the model (e.g., by pressing the Enter key).
66#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Clone, Serialize, Deserialize)]
67pub struct PromptId(Arc<str>);
68
69impl PromptId {
70 pub fn new() -> Self {
71 Self(Uuid::new_v4().to_string().into())
72 }
73}
74
75impl std::fmt::Display for PromptId {
76 fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
77 write!(f, "{}", self.0)
78 }
79}
80
81pub(crate) const MAX_RETRY_ATTEMPTS: u8 = 4;
82pub(crate) const BASE_RETRY_DELAY: Duration = Duration::from_secs(5);
83
84#[derive(Debug, Clone)]
85enum RetryStrategy {
86 ExponentialBackoff {
87 initial_delay: Duration,
88 max_attempts: u8,
89 },
90 Fixed {
91 delay: Duration,
92 max_attempts: u8,
93 },
94}
95
96#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
97pub enum Message {
98 User(UserMessage),
99 Agent(AgentMessage),
100 Resume,
101}
102
103impl Message {
104 pub fn as_agent_message(&self) -> Option<&AgentMessage> {
105 match self {
106 Message::Agent(agent_message) => Some(agent_message),
107 _ => None,
108 }
109 }
110
111 pub fn to_request(&self) -> Vec<LanguageModelRequestMessage> {
112 match self {
113 Message::User(message) => vec![message.to_request()],
114 Message::Agent(message) => message.to_request(),
115 Message::Resume => vec![LanguageModelRequestMessage {
116 role: Role::User,
117 content: vec!["Continue where you left off".into()],
118 cache: false,
119 }],
120 }
121 }
122
123 pub fn to_markdown(&self) -> String {
124 match self {
125 Message::User(message) => message.to_markdown(),
126 Message::Agent(message) => message.to_markdown(),
127 Message::Resume => "[resume]\n".into(),
128 }
129 }
130
131 pub fn role(&self) -> Role {
132 match self {
133 Message::User(_) | Message::Resume => Role::User,
134 Message::Agent(_) => Role::Assistant,
135 }
136 }
137}
138
139#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
140pub struct UserMessage {
141 pub id: UserMessageId,
142 pub content: Vec<UserMessageContent>,
143}
144
145#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
146pub enum UserMessageContent {
147 Text(String),
148 Mention { uri: MentionUri, content: String },
149 Image(LanguageModelImage),
150}
151
152impl UserMessage {
153 pub fn to_markdown(&self) -> String {
154 let mut markdown = String::from("## User\n\n");
155
156 for content in &self.content {
157 match content {
158 UserMessageContent::Text(text) => {
159 markdown.push_str(text);
160 markdown.push('\n');
161 }
162 UserMessageContent::Image(_) => {
163 markdown.push_str("<image />\n");
164 }
165 UserMessageContent::Mention { uri, content } => {
166 if !content.is_empty() {
167 let _ = writeln!(&mut markdown, "{}\n\n{}", uri.as_link(), content);
168 } else {
169 let _ = writeln!(&mut markdown, "{}", uri.as_link());
170 }
171 }
172 }
173 }
174
175 markdown
176 }
177
178 fn to_request(&self) -> LanguageModelRequestMessage {
179 let mut message = LanguageModelRequestMessage {
180 role: Role::User,
181 content: Vec::with_capacity(self.content.len()),
182 cache: false,
183 };
184
185 const OPEN_CONTEXT: &str = "<context>\n\
186 The following items were attached by the user. \
187 They are up-to-date and don't need to be re-read.\n\n";
188
189 const OPEN_FILES_TAG: &str = "<files>";
190 const OPEN_DIRECTORIES_TAG: &str = "<directories>";
191 const OPEN_SYMBOLS_TAG: &str = "<symbols>";
192 const OPEN_SELECTIONS_TAG: &str = "<selections>";
193 const OPEN_THREADS_TAG: &str = "<threads>";
194 const OPEN_FETCH_TAG: &str = "<fetched_urls>";
195 const OPEN_RULES_TAG: &str =
196 "<rules>\nThe user has specified the following rules that should be applied:\n";
197
198 let mut file_context = OPEN_FILES_TAG.to_string();
199 let mut directory_context = OPEN_DIRECTORIES_TAG.to_string();
200 let mut symbol_context = OPEN_SYMBOLS_TAG.to_string();
201 let mut selection_context = OPEN_SELECTIONS_TAG.to_string();
202 let mut thread_context = OPEN_THREADS_TAG.to_string();
203 let mut fetch_context = OPEN_FETCH_TAG.to_string();
204 let mut rules_context = OPEN_RULES_TAG.to_string();
205
206 for chunk in &self.content {
207 let chunk = match chunk {
208 UserMessageContent::Text(text) => {
209 language_model::MessageContent::Text(text.clone())
210 }
211 UserMessageContent::Image(value) => {
212 language_model::MessageContent::Image(value.clone())
213 }
214 UserMessageContent::Mention { uri, content } => {
215 match uri {
216 MentionUri::File { abs_path } => {
217 write!(
218 &mut file_context,
219 "\n{}",
220 MarkdownCodeBlock {
221 tag: &codeblock_tag(abs_path, None),
222 text: &content.to_string(),
223 }
224 )
225 .ok();
226 }
227 MentionUri::PastedImage => {
228 debug_panic!("pasted image URI should not be used in mention content")
229 }
230 MentionUri::Directory { .. } => {
231 write!(&mut directory_context, "\n{}\n", content).ok();
232 }
233 MentionUri::Symbol {
234 abs_path: path,
235 line_range,
236 ..
237 } => {
238 write!(
239 &mut symbol_context,
240 "\n{}",
241 MarkdownCodeBlock {
242 tag: &codeblock_tag(path, Some(line_range)),
243 text: content
244 }
245 )
246 .ok();
247 }
248 MentionUri::Selection {
249 abs_path: path,
250 line_range,
251 ..
252 } => {
253 write!(
254 &mut selection_context,
255 "\n{}",
256 MarkdownCodeBlock {
257 tag: &codeblock_tag(
258 path.as_deref().unwrap_or("Untitled".as_ref()),
259 Some(line_range)
260 ),
261 text: content
262 }
263 )
264 .ok();
265 }
266 MentionUri::Thread { .. } => {
267 write!(&mut thread_context, "\n{}\n", content).ok();
268 }
269 MentionUri::TextThread { .. } => {
270 write!(&mut thread_context, "\n{}\n", content).ok();
271 }
272 MentionUri::Rule { .. } => {
273 write!(
274 &mut rules_context,
275 "\n{}",
276 MarkdownCodeBlock {
277 tag: "",
278 text: content
279 }
280 )
281 .ok();
282 }
283 MentionUri::Fetch { url } => {
284 write!(&mut fetch_context, "\nFetch: {}\n\n{}", url, content).ok();
285 }
286 }
287
288 language_model::MessageContent::Text(uri.as_link().to_string())
289 }
290 };
291
292 message.content.push(chunk);
293 }
294
295 let len_before_context = message.content.len();
296
297 if file_context.len() > OPEN_FILES_TAG.len() {
298 file_context.push_str("</files>\n");
299 message
300 .content
301 .push(language_model::MessageContent::Text(file_context));
302 }
303
304 if directory_context.len() > OPEN_DIRECTORIES_TAG.len() {
305 directory_context.push_str("</directories>\n");
306 message
307 .content
308 .push(language_model::MessageContent::Text(directory_context));
309 }
310
311 if symbol_context.len() > OPEN_SYMBOLS_TAG.len() {
312 symbol_context.push_str("</symbols>\n");
313 message
314 .content
315 .push(language_model::MessageContent::Text(symbol_context));
316 }
317
318 if selection_context.len() > OPEN_SELECTIONS_TAG.len() {
319 selection_context.push_str("</selections>\n");
320 message
321 .content
322 .push(language_model::MessageContent::Text(selection_context));
323 }
324
325 if thread_context.len() > OPEN_THREADS_TAG.len() {
326 thread_context.push_str("</threads>\n");
327 message
328 .content
329 .push(language_model::MessageContent::Text(thread_context));
330 }
331
332 if fetch_context.len() > OPEN_FETCH_TAG.len() {
333 fetch_context.push_str("</fetched_urls>\n");
334 message
335 .content
336 .push(language_model::MessageContent::Text(fetch_context));
337 }
338
339 if rules_context.len() > OPEN_RULES_TAG.len() {
340 rules_context.push_str("</user_rules>\n");
341 message
342 .content
343 .push(language_model::MessageContent::Text(rules_context));
344 }
345
346 if message.content.len() > len_before_context {
347 message.content.insert(
348 len_before_context,
349 language_model::MessageContent::Text(OPEN_CONTEXT.into()),
350 );
351 message
352 .content
353 .push(language_model::MessageContent::Text("</context>".into()));
354 }
355
356 message
357 }
358}
359
360fn codeblock_tag(full_path: &Path, line_range: Option<&RangeInclusive<u32>>) -> String {
361 let mut result = String::new();
362
363 if let Some(extension) = full_path.extension().and_then(|ext| ext.to_str()) {
364 let _ = write!(result, "{} ", extension);
365 }
366
367 let _ = write!(result, "{}", full_path.display());
368
369 if let Some(range) = line_range {
370 if range.start() == range.end() {
371 let _ = write!(result, ":{}", range.start() + 1);
372 } else {
373 let _ = write!(result, ":{}-{}", range.start() + 1, range.end() + 1);
374 }
375 }
376
377 result
378}
379
380impl AgentMessage {
381 pub fn to_markdown(&self) -> String {
382 let mut markdown = String::from("## Assistant\n\n");
383
384 for content in &self.content {
385 match content {
386 AgentMessageContent::Text(text) => {
387 markdown.push_str(text);
388 markdown.push('\n');
389 }
390 AgentMessageContent::Thinking { text, .. } => {
391 markdown.push_str("<think>");
392 markdown.push_str(text);
393 markdown.push_str("</think>\n");
394 }
395 AgentMessageContent::RedactedThinking(_) => {
396 markdown.push_str("<redacted_thinking />\n")
397 }
398 AgentMessageContent::ToolUse(tool_use) => {
399 markdown.push_str(&format!(
400 "**Tool Use**: {} (ID: {})\n",
401 tool_use.name, tool_use.id
402 ));
403 markdown.push_str(&format!(
404 "{}\n",
405 MarkdownCodeBlock {
406 tag: "json",
407 text: &format!("{:#}", tool_use.input)
408 }
409 ));
410 }
411 }
412 }
413
414 for tool_result in self.tool_results.values() {
415 markdown.push_str(&format!(
416 "**Tool Result**: {} (ID: {})\n\n",
417 tool_result.tool_name, tool_result.tool_use_id
418 ));
419 if tool_result.is_error {
420 markdown.push_str("**ERROR:**\n");
421 }
422
423 match &tool_result.content {
424 LanguageModelToolResultContent::Text(text) => {
425 writeln!(markdown, "{text}\n").ok();
426 }
427 LanguageModelToolResultContent::Image(_) => {
428 writeln!(markdown, "<image />\n").ok();
429 }
430 }
431
432 if let Some(output) = tool_result.output.as_ref() {
433 writeln!(
434 markdown,
435 "**Debug Output**:\n\n```json\n{}\n```\n",
436 serde_json::to_string_pretty(output).unwrap()
437 )
438 .unwrap();
439 }
440 }
441
442 markdown
443 }
444
445 pub fn to_request(&self) -> Vec<LanguageModelRequestMessage> {
446 let mut assistant_message = LanguageModelRequestMessage {
447 role: Role::Assistant,
448 content: Vec::with_capacity(self.content.len()),
449 cache: false,
450 };
451 for chunk in &self.content {
452 match chunk {
453 AgentMessageContent::Text(text) => {
454 assistant_message
455 .content
456 .push(language_model::MessageContent::Text(text.clone()));
457 }
458 AgentMessageContent::Thinking { text, signature } => {
459 assistant_message
460 .content
461 .push(language_model::MessageContent::Thinking {
462 text: text.clone(),
463 signature: signature.clone(),
464 });
465 }
466 AgentMessageContent::RedactedThinking(value) => {
467 assistant_message.content.push(
468 language_model::MessageContent::RedactedThinking(value.clone()),
469 );
470 }
471 AgentMessageContent::ToolUse(tool_use) => {
472 if self.tool_results.contains_key(&tool_use.id) {
473 assistant_message
474 .content
475 .push(language_model::MessageContent::ToolUse(tool_use.clone()));
476 }
477 }
478 };
479 }
480
481 let mut user_message = LanguageModelRequestMessage {
482 role: Role::User,
483 content: Vec::new(),
484 cache: false,
485 };
486
487 for tool_result in self.tool_results.values() {
488 let mut tool_result = tool_result.clone();
489 // Surprisingly, the API fails if we return an empty string here.
490 // It thinks we are sending a tool use without a tool result.
491 if tool_result.content.is_empty() {
492 tool_result.content = "<Tool returned an empty string>".into();
493 }
494 user_message
495 .content
496 .push(language_model::MessageContent::ToolResult(tool_result));
497 }
498
499 let mut messages = Vec::new();
500 if !assistant_message.content.is_empty() {
501 messages.push(assistant_message);
502 }
503 if !user_message.content.is_empty() {
504 messages.push(user_message);
505 }
506 messages
507 }
508}
509
510#[derive(Default, Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
511pub struct AgentMessage {
512 pub content: Vec<AgentMessageContent>,
513 pub tool_results: IndexMap<LanguageModelToolUseId, LanguageModelToolResult>,
514}
515
516#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
517pub enum AgentMessageContent {
518 Text(String),
519 Thinking {
520 text: String,
521 signature: Option<String>,
522 },
523 RedactedThinking(String),
524 ToolUse(LanguageModelToolUse),
525}
526
527pub trait TerminalHandle {
528 fn id(&self, cx: &AsyncApp) -> Result<acp::TerminalId>;
529 fn current_output(&self, cx: &AsyncApp) -> Result<acp::TerminalOutputResponse>;
530 fn wait_for_exit(&self, cx: &AsyncApp) -> Result<Shared<Task<acp::TerminalExitStatus>>>;
531}
532
533pub trait ThreadEnvironment {
534 fn create_terminal(
535 &self,
536 command: String,
537 cwd: Option<PathBuf>,
538 output_byte_limit: Option<u64>,
539 cx: &mut AsyncApp,
540 ) -> Task<Result<Rc<dyn TerminalHandle>>>;
541}
542
543#[derive(Debug)]
544pub enum ThreadEvent {
545 UserMessage(UserMessage),
546 AgentText(String),
547 AgentThinking(String),
548 ToolCall(acp::ToolCall),
549 ToolCallUpdate(acp_thread::ToolCallUpdate),
550 ToolCallAuthorization(ToolCallAuthorization),
551 Retry(acp_thread::RetryStatus),
552 Stop(acp::StopReason),
553}
554
555#[derive(Debug)]
556pub struct NewTerminal {
557 pub command: String,
558 pub output_byte_limit: Option<u64>,
559 pub cwd: Option<PathBuf>,
560 pub response: oneshot::Sender<Result<Entity<acp_thread::Terminal>>>,
561}
562
563#[derive(Debug)]
564pub struct ToolCallAuthorization {
565 pub tool_call: acp::ToolCallUpdate,
566 pub options: Vec<acp::PermissionOption>,
567 pub response: oneshot::Sender<acp::PermissionOptionId>,
568}
569
570#[derive(Debug, thiserror::Error)]
571enum CompletionError {
572 #[error("max tokens")]
573 MaxTokens,
574 #[error("refusal")]
575 Refusal,
576 #[error(transparent)]
577 Other(#[from] anyhow::Error),
578}
579
580pub struct Thread {
581 id: acp::SessionId,
582 prompt_id: PromptId,
583 updated_at: DateTime<Utc>,
584 title: Option<SharedString>,
585 pending_title_generation: Option<Task<()>>,
586 summary: Option<SharedString>,
587 messages: Vec<Message>,
588 completion_mode: CompletionMode,
589 /// Holds the task that handles agent interaction until the end of the turn.
590 /// Survives across multiple requests as the model performs tool calls and
591 /// we run tools, report their results.
592 running_turn: Option<RunningTurn>,
593 pending_message: Option<AgentMessage>,
594 tools: BTreeMap<SharedString, Arc<dyn AnyAgentTool>>,
595 tool_use_limit_reached: bool,
596 request_token_usage: HashMap<UserMessageId, language_model::TokenUsage>,
597 #[allow(unused)]
598 cumulative_token_usage: TokenUsage,
599 #[allow(unused)]
600 initial_project_snapshot: Shared<Task<Option<Arc<ProjectSnapshot>>>>,
601 context_server_registry: Entity<ContextServerRegistry>,
602 profile_id: AgentProfileId,
603 project_context: Entity<ProjectContext>,
604 templates: Arc<Templates>,
605 model: Option<Arc<dyn LanguageModel>>,
606 summarization_model: Option<Arc<dyn LanguageModel>>,
607 prompt_capabilities_tx: watch::Sender<acp::PromptCapabilities>,
608 pub(crate) prompt_capabilities_rx: watch::Receiver<acp::PromptCapabilities>,
609 pub(crate) project: Entity<Project>,
610 pub(crate) action_log: Entity<ActionLog>,
611}
612
613impl Thread {
614 fn prompt_capabilities(model: Option<&dyn LanguageModel>) -> acp::PromptCapabilities {
615 let image = model.map_or(true, |model| model.supports_images());
616 acp::PromptCapabilities {
617 meta: None,
618 image,
619 audio: false,
620 embedded_context: true,
621 }
622 }
623
624 pub fn new(
625 project: Entity<Project>,
626 project_context: Entity<ProjectContext>,
627 context_server_registry: Entity<ContextServerRegistry>,
628 templates: Arc<Templates>,
629 model: Option<Arc<dyn LanguageModel>>,
630 cx: &mut Context<Self>,
631 ) -> Self {
632 let profile_id = AgentSettings::get_global(cx).default_profile.clone();
633 let action_log = cx.new(|_cx| ActionLog::new(project.clone()));
634 let (prompt_capabilities_tx, prompt_capabilities_rx) =
635 watch::channel(Self::prompt_capabilities(model.as_deref()));
636 Self {
637 id: acp::SessionId(uuid::Uuid::new_v4().to_string().into()),
638 prompt_id: PromptId::new(),
639 updated_at: Utc::now(),
640 title: None,
641 pending_title_generation: None,
642 summary: None,
643 messages: Vec::new(),
644 completion_mode: AgentSettings::get_global(cx).preferred_completion_mode,
645 running_turn: None,
646 pending_message: None,
647 tools: BTreeMap::default(),
648 tool_use_limit_reached: false,
649 request_token_usage: HashMap::default(),
650 cumulative_token_usage: TokenUsage::default(),
651 initial_project_snapshot: {
652 let project_snapshot = Self::project_snapshot(project.clone(), cx);
653 cx.foreground_executor()
654 .spawn(async move { Some(project_snapshot.await) })
655 .shared()
656 },
657 context_server_registry,
658 profile_id,
659 project_context,
660 templates,
661 model,
662 summarization_model: None,
663 prompt_capabilities_tx,
664 prompt_capabilities_rx,
665 project,
666 action_log,
667 }
668 }
669
670 pub fn id(&self) -> &acp::SessionId {
671 &self.id
672 }
673
674 pub fn replay(
675 &mut self,
676 cx: &mut Context<Self>,
677 ) -> mpsc::UnboundedReceiver<Result<ThreadEvent>> {
678 let (tx, rx) = mpsc::unbounded();
679 let stream = ThreadEventStream(tx);
680 for message in &self.messages {
681 match message {
682 Message::User(user_message) => stream.send_user_message(user_message),
683 Message::Agent(assistant_message) => {
684 for content in &assistant_message.content {
685 match content {
686 AgentMessageContent::Text(text) => stream.send_text(text),
687 AgentMessageContent::Thinking { text, .. } => {
688 stream.send_thinking(text)
689 }
690 AgentMessageContent::RedactedThinking(_) => {}
691 AgentMessageContent::ToolUse(tool_use) => {
692 self.replay_tool_call(
693 tool_use,
694 assistant_message.tool_results.get(&tool_use.id),
695 &stream,
696 cx,
697 );
698 }
699 }
700 }
701 }
702 Message::Resume => {}
703 }
704 }
705 rx
706 }
707
708 fn replay_tool_call(
709 &self,
710 tool_use: &LanguageModelToolUse,
711 tool_result: Option<&LanguageModelToolResult>,
712 stream: &ThreadEventStream,
713 cx: &mut Context<Self>,
714 ) {
715 let tool = self.tools.get(tool_use.name.as_ref()).cloned().or_else(|| {
716 self.context_server_registry
717 .read(cx)
718 .servers()
719 .find_map(|(_, tools)| {
720 if let Some(tool) = tools.get(tool_use.name.as_ref()) {
721 Some(tool.clone())
722 } else {
723 None
724 }
725 })
726 });
727
728 let Some(tool) = tool else {
729 stream
730 .0
731 .unbounded_send(Ok(ThreadEvent::ToolCall(acp::ToolCall {
732 meta: None,
733 id: acp::ToolCallId(tool_use.id.to_string().into()),
734 title: tool_use.name.to_string(),
735 kind: acp::ToolKind::Other,
736 status: acp::ToolCallStatus::Failed,
737 content: Vec::new(),
738 locations: Vec::new(),
739 raw_input: Some(tool_use.input.clone()),
740 raw_output: None,
741 })))
742 .ok();
743 return;
744 };
745
746 let title = tool.initial_title(tool_use.input.clone(), cx);
747 let kind = tool.kind();
748 stream.send_tool_call(&tool_use.id, title, kind, tool_use.input.clone());
749
750 let output = tool_result
751 .as_ref()
752 .and_then(|result| result.output.clone());
753 if let Some(output) = output.clone() {
754 let tool_event_stream = ToolCallEventStream::new(
755 tool_use.id.clone(),
756 stream.clone(),
757 Some(self.project.read(cx).fs().clone()),
758 );
759 tool.replay(tool_use.input.clone(), output, tool_event_stream, cx)
760 .log_err();
761 }
762
763 stream.update_tool_call_fields(
764 &tool_use.id,
765 acp::ToolCallUpdateFields {
766 status: Some(
767 tool_result
768 .as_ref()
769 .map_or(acp::ToolCallStatus::Failed, |result| {
770 if result.is_error {
771 acp::ToolCallStatus::Failed
772 } else {
773 acp::ToolCallStatus::Completed
774 }
775 }),
776 ),
777 raw_output: output,
778 ..Default::default()
779 },
780 );
781 }
782
783 pub fn from_db(
784 id: acp::SessionId,
785 db_thread: DbThread,
786 project: Entity<Project>,
787 project_context: Entity<ProjectContext>,
788 context_server_registry: Entity<ContextServerRegistry>,
789 action_log: Entity<ActionLog>,
790 templates: Arc<Templates>,
791 cx: &mut Context<Self>,
792 ) -> Self {
793 let profile_id = db_thread
794 .profile
795 .unwrap_or_else(|| AgentSettings::get_global(cx).default_profile.clone());
796 let model = LanguageModelRegistry::global(cx).update(cx, |registry, cx| {
797 db_thread
798 .model
799 .and_then(|model| {
800 let model = SelectedModel {
801 provider: model.provider.clone().into(),
802 model: model.model.into(),
803 };
804 registry.select_model(&model, cx)
805 })
806 .or_else(|| registry.default_model())
807 .map(|model| model.model)
808 });
809 let (prompt_capabilities_tx, prompt_capabilities_rx) =
810 watch::channel(Self::prompt_capabilities(model.as_deref()));
811
812 Self {
813 id,
814 prompt_id: PromptId::new(),
815 title: if db_thread.title.is_empty() {
816 None
817 } else {
818 Some(db_thread.title.clone())
819 },
820 pending_title_generation: None,
821 summary: db_thread.detailed_summary,
822 messages: db_thread.messages,
823 completion_mode: db_thread.completion_mode.unwrap_or_default(),
824 running_turn: None,
825 pending_message: None,
826 tools: BTreeMap::default(),
827 tool_use_limit_reached: false,
828 request_token_usage: db_thread.request_token_usage.clone(),
829 cumulative_token_usage: db_thread.cumulative_token_usage,
830 initial_project_snapshot: Task::ready(db_thread.initial_project_snapshot).shared(),
831 context_server_registry,
832 profile_id,
833 project_context,
834 templates,
835 model,
836 summarization_model: None,
837 project,
838 action_log,
839 updated_at: db_thread.updated_at,
840 prompt_capabilities_tx,
841 prompt_capabilities_rx,
842 }
843 }
844
845 pub fn to_db(&self, cx: &App) -> Task<DbThread> {
846 let initial_project_snapshot = self.initial_project_snapshot.clone();
847 let mut thread = DbThread {
848 title: self.title(),
849 messages: self.messages.clone(),
850 updated_at: self.updated_at,
851 detailed_summary: self.summary.clone(),
852 initial_project_snapshot: None,
853 cumulative_token_usage: self.cumulative_token_usage,
854 request_token_usage: self.request_token_usage.clone(),
855 model: self.model.as_ref().map(|model| DbLanguageModel {
856 provider: model.provider_id().to_string(),
857 model: model.name().0.to_string(),
858 }),
859 completion_mode: Some(self.completion_mode),
860 profile: Some(self.profile_id.clone()),
861 };
862
863 cx.background_spawn(async move {
864 let initial_project_snapshot = initial_project_snapshot.await;
865 thread.initial_project_snapshot = initial_project_snapshot;
866 thread
867 })
868 }
869
870 /// Create a snapshot of the current project state including git information and unsaved buffers.
871 fn project_snapshot(
872 project: Entity<Project>,
873 cx: &mut Context<Self>,
874 ) -> Task<Arc<agent::thread::ProjectSnapshot>> {
875 let git_store = project.read(cx).git_store().clone();
876 let worktree_snapshots: Vec<_> = project
877 .read(cx)
878 .visible_worktrees(cx)
879 .map(|worktree| Self::worktree_snapshot(worktree, git_store.clone(), cx))
880 .collect();
881
882 cx.spawn(async move |_, _| {
883 let worktree_snapshots = futures::future::join_all(worktree_snapshots).await;
884
885 Arc::new(ProjectSnapshot {
886 worktree_snapshots,
887 timestamp: Utc::now(),
888 })
889 })
890 }
891
892 fn worktree_snapshot(
893 worktree: Entity<project::Worktree>,
894 git_store: Entity<GitStore>,
895 cx: &App,
896 ) -> Task<agent::thread::WorktreeSnapshot> {
897 cx.spawn(async move |cx| {
898 // Get worktree path and snapshot
899 let worktree_info = cx.update(|app_cx| {
900 let worktree = worktree.read(app_cx);
901 let path = worktree.abs_path().to_string_lossy().into_owned();
902 let snapshot = worktree.snapshot();
903 (path, snapshot)
904 });
905
906 let Ok((worktree_path, _snapshot)) = worktree_info else {
907 return WorktreeSnapshot {
908 worktree_path: String::new(),
909 git_state: None,
910 };
911 };
912
913 let git_state = git_store
914 .update(cx, |git_store, cx| {
915 git_store
916 .repositories()
917 .values()
918 .find(|repo| {
919 repo.read(cx)
920 .abs_path_to_repo_path(&worktree.read(cx).abs_path())
921 .is_some()
922 })
923 .cloned()
924 })
925 .ok()
926 .flatten()
927 .map(|repo| {
928 repo.update(cx, |repo, _| {
929 let current_branch =
930 repo.branch.as_ref().map(|branch| branch.name().to_owned());
931 repo.send_job(None, |state, _| async move {
932 let RepositoryState::Local { backend, .. } = state else {
933 return GitState {
934 remote_url: None,
935 head_sha: None,
936 current_branch,
937 diff: None,
938 };
939 };
940
941 let remote_url = backend.remote_url("origin");
942 let head_sha = backend.head_sha().await;
943 let diff = backend.diff(DiffType::HeadToWorktree).await.ok();
944
945 GitState {
946 remote_url,
947 head_sha,
948 current_branch,
949 diff,
950 }
951 })
952 })
953 });
954
955 let git_state = match git_state {
956 Some(git_state) => match git_state.ok() {
957 Some(git_state) => git_state.await.ok(),
958 None => None,
959 },
960 None => None,
961 };
962
963 WorktreeSnapshot {
964 worktree_path,
965 git_state,
966 }
967 })
968 }
969
970 pub fn project_context(&self) -> &Entity<ProjectContext> {
971 &self.project_context
972 }
973
974 pub fn project(&self) -> &Entity<Project> {
975 &self.project
976 }
977
978 pub fn action_log(&self) -> &Entity<ActionLog> {
979 &self.action_log
980 }
981
982 pub fn is_empty(&self) -> bool {
983 self.messages.is_empty() && self.title.is_none()
984 }
985
986 pub fn model(&self) -> Option<&Arc<dyn LanguageModel>> {
987 self.model.as_ref()
988 }
989
990 pub fn set_model(&mut self, model: Arc<dyn LanguageModel>, cx: &mut Context<Self>) {
991 let old_usage = self.latest_token_usage();
992 self.model = Some(model);
993 let new_caps = Self::prompt_capabilities(self.model.as_deref());
994 let new_usage = self.latest_token_usage();
995 if old_usage != new_usage {
996 cx.emit(TokenUsageUpdated(new_usage));
997 }
998 self.prompt_capabilities_tx.send(new_caps).log_err();
999 cx.notify()
1000 }
1001
1002 pub fn summarization_model(&self) -> Option<&Arc<dyn LanguageModel>> {
1003 self.summarization_model.as_ref()
1004 }
1005
1006 pub fn set_summarization_model(
1007 &mut self,
1008 model: Option<Arc<dyn LanguageModel>>,
1009 cx: &mut Context<Self>,
1010 ) {
1011 self.summarization_model = model;
1012 cx.notify()
1013 }
1014
1015 pub fn completion_mode(&self) -> CompletionMode {
1016 self.completion_mode
1017 }
1018
1019 pub fn set_completion_mode(&mut self, mode: CompletionMode, cx: &mut Context<Self>) {
1020 let old_usage = self.latest_token_usage();
1021 self.completion_mode = mode;
1022 let new_usage = self.latest_token_usage();
1023 if old_usage != new_usage {
1024 cx.emit(TokenUsageUpdated(new_usage));
1025 }
1026 cx.notify()
1027 }
1028
1029 #[cfg(any(test, feature = "test-support"))]
1030 pub fn last_message(&self) -> Option<Message> {
1031 if let Some(message) = self.pending_message.clone() {
1032 Some(Message::Agent(message))
1033 } else {
1034 self.messages.last().cloned()
1035 }
1036 }
1037
1038 pub fn add_default_tools(
1039 &mut self,
1040 environment: Rc<dyn ThreadEnvironment>,
1041 cx: &mut Context<Self>,
1042 ) {
1043 let language_registry = self.project.read(cx).languages().clone();
1044 self.add_tool(CopyPathTool::new(self.project.clone()));
1045 self.add_tool(CreateDirectoryTool::new(self.project.clone()));
1046 self.add_tool(DeletePathTool::new(
1047 self.project.clone(),
1048 self.action_log.clone(),
1049 ));
1050 self.add_tool(DiagnosticsTool::new(self.project.clone()));
1051 self.add_tool(EditFileTool::new(
1052 self.project.clone(),
1053 cx.weak_entity(),
1054 language_registry,
1055 ));
1056 self.add_tool(FetchTool::new(self.project.read(cx).client().http_client()));
1057 self.add_tool(FindPathTool::new(self.project.clone()));
1058 self.add_tool(GrepTool::new(self.project.clone()));
1059 self.add_tool(ListDirectoryTool::new(self.project.clone()));
1060 self.add_tool(MovePathTool::new(self.project.clone()));
1061 self.add_tool(NowTool);
1062 self.add_tool(OpenTool::new(self.project.clone()));
1063 self.add_tool(ReadFileTool::new(
1064 self.project.clone(),
1065 self.action_log.clone(),
1066 ));
1067 self.add_tool(TerminalTool::new(self.project.clone(), environment));
1068 self.add_tool(ThinkingTool);
1069 self.add_tool(WebSearchTool);
1070 }
1071
1072 pub fn add_tool<T: AgentTool>(&mut self, tool: T) {
1073 self.tools.insert(T::name().into(), tool.erase());
1074 }
1075
1076 pub fn remove_tool(&mut self, name: &str) -> bool {
1077 self.tools.remove(name).is_some()
1078 }
1079
1080 pub fn profile(&self) -> &AgentProfileId {
1081 &self.profile_id
1082 }
1083
1084 pub fn set_profile(&mut self, profile_id: AgentProfileId) {
1085 self.profile_id = profile_id;
1086 }
1087
1088 pub fn cancel(&mut self, cx: &mut Context<Self>) {
1089 if let Some(running_turn) = self.running_turn.take() {
1090 running_turn.cancel();
1091 }
1092 self.flush_pending_message(cx);
1093 }
1094
1095 fn update_token_usage(&mut self, update: language_model::TokenUsage, cx: &mut Context<Self>) {
1096 let Some(last_user_message) = self.last_user_message() else {
1097 return;
1098 };
1099
1100 self.request_token_usage
1101 .insert(last_user_message.id.clone(), update);
1102 cx.emit(TokenUsageUpdated(self.latest_token_usage()));
1103 cx.notify();
1104 }
1105
1106 pub fn truncate(&mut self, message_id: UserMessageId, cx: &mut Context<Self>) -> Result<()> {
1107 self.cancel(cx);
1108 let Some(position) = self.messages.iter().position(
1109 |msg| matches!(msg, Message::User(UserMessage { id, .. }) if id == &message_id),
1110 ) else {
1111 return Err(anyhow!("Message not found"));
1112 };
1113
1114 for message in self.messages.drain(position..) {
1115 match message {
1116 Message::User(message) => {
1117 self.request_token_usage.remove(&message.id);
1118 }
1119 Message::Agent(_) | Message::Resume => {}
1120 }
1121 }
1122 self.summary = None;
1123 cx.notify();
1124 Ok(())
1125 }
1126
1127 pub fn latest_token_usage(&self) -> Option<acp_thread::TokenUsage> {
1128 let last_user_message = self.last_user_message()?;
1129 let tokens = self.request_token_usage.get(&last_user_message.id)?;
1130 let model = self.model.clone()?;
1131
1132 Some(acp_thread::TokenUsage {
1133 max_tokens: model.max_token_count_for_mode(self.completion_mode.into()),
1134 used_tokens: tokens.total_tokens(),
1135 })
1136 }
1137
1138 pub fn resume(
1139 &mut self,
1140 cx: &mut Context<Self>,
1141 ) -> Result<mpsc::UnboundedReceiver<Result<ThreadEvent>>> {
1142 self.messages.push(Message::Resume);
1143 cx.notify();
1144
1145 log::debug!("Total messages in thread: {}", self.messages.len());
1146 self.run_turn(cx)
1147 }
1148
1149 /// Sending a message results in the model streaming a response, which could include tool calls.
1150 /// After calling tools, the model will stops and waits for any outstanding tool calls to be completed and their results sent.
1151 /// The returned channel will report all the occurrences in which the model stops before erroring or ending its turn.
1152 pub fn send<T>(
1153 &mut self,
1154 id: UserMessageId,
1155 content: impl IntoIterator<Item = T>,
1156 cx: &mut Context<Self>,
1157 ) -> Result<mpsc::UnboundedReceiver<Result<ThreadEvent>>>
1158 where
1159 T: Into<UserMessageContent>,
1160 {
1161 let model = self.model().context("No language model configured")?;
1162
1163 log::info!("Thread::send called with model: {}", model.name().0);
1164 self.advance_prompt_id();
1165
1166 let content = content.into_iter().map(Into::into).collect::<Vec<_>>();
1167 log::debug!("Thread::send content: {:?}", content);
1168
1169 self.messages
1170 .push(Message::User(UserMessage { id, content }));
1171 cx.notify();
1172
1173 log::debug!("Total messages in thread: {}", self.messages.len());
1174 self.run_turn(cx)
1175 }
1176
1177 fn run_turn(
1178 &mut self,
1179 cx: &mut Context<Self>,
1180 ) -> Result<mpsc::UnboundedReceiver<Result<ThreadEvent>>> {
1181 self.cancel(cx);
1182
1183 let model = self.model.clone().context("No language model configured")?;
1184 let profile = AgentSettings::get_global(cx)
1185 .profiles
1186 .get(&self.profile_id)
1187 .context("Profile not found")?;
1188 let (events_tx, events_rx) = mpsc::unbounded::<Result<ThreadEvent>>();
1189 let event_stream = ThreadEventStream(events_tx);
1190 let message_ix = self.messages.len().saturating_sub(1);
1191 self.tool_use_limit_reached = false;
1192 self.summary = None;
1193 self.running_turn = Some(RunningTurn {
1194 event_stream: event_stream.clone(),
1195 tools: self.enabled_tools(profile, &model, cx),
1196 _task: cx.spawn(async move |this, cx| {
1197 log::debug!("Starting agent turn execution");
1198
1199 let turn_result = Self::run_turn_internal(&this, model, &event_stream, cx).await;
1200 _ = this.update(cx, |this, cx| this.flush_pending_message(cx));
1201
1202 match turn_result {
1203 Ok(()) => {
1204 log::debug!("Turn execution completed");
1205 event_stream.send_stop(acp::StopReason::EndTurn);
1206 }
1207 Err(error) => {
1208 log::error!("Turn execution failed: {:?}", error);
1209 match error.downcast::<CompletionError>() {
1210 Ok(CompletionError::Refusal) => {
1211 event_stream.send_stop(acp::StopReason::Refusal);
1212 _ = this.update(cx, |this, _| this.messages.truncate(message_ix));
1213 }
1214 Ok(CompletionError::MaxTokens) => {
1215 event_stream.send_stop(acp::StopReason::MaxTokens);
1216 }
1217 Ok(CompletionError::Other(error)) | Err(error) => {
1218 event_stream.send_error(error);
1219 }
1220 }
1221 }
1222 }
1223
1224 _ = this.update(cx, |this, _| this.running_turn.take());
1225 }),
1226 });
1227 Ok(events_rx)
1228 }
1229
1230 async fn run_turn_internal(
1231 this: &WeakEntity<Self>,
1232 model: Arc<dyn LanguageModel>,
1233 event_stream: &ThreadEventStream,
1234 cx: &mut AsyncApp,
1235 ) -> Result<()> {
1236 let mut attempt = 0;
1237 let mut intent = CompletionIntent::UserPrompt;
1238 loop {
1239 let request =
1240 this.update(cx, |this, cx| this.build_completion_request(intent, cx))??;
1241
1242 telemetry::event!(
1243 "Agent Thread Completion",
1244 thread_id = this.read_with(cx, |this, _| this.id.to_string())?,
1245 prompt_id = this.read_with(cx, |this, _| this.prompt_id.to_string())?,
1246 model = model.telemetry_id(),
1247 model_provider = model.provider_id().to_string(),
1248 attempt
1249 );
1250
1251 log::debug!("Calling model.stream_completion, attempt {}", attempt);
1252 let mut events = model
1253 .stream_completion(request, cx)
1254 .await
1255 .map_err(|error| anyhow!(error))?;
1256 let mut tool_results = FuturesUnordered::new();
1257 let mut error = None;
1258 while let Some(event) = events.next().await {
1259 log::trace!("Received completion event: {:?}", event);
1260 match event {
1261 Ok(event) => {
1262 tool_results.extend(this.update(cx, |this, cx| {
1263 this.handle_completion_event(event, event_stream, cx)
1264 })??);
1265 }
1266 Err(err) => {
1267 error = Some(err);
1268 break;
1269 }
1270 }
1271 }
1272
1273 let end_turn = tool_results.is_empty();
1274 while let Some(tool_result) = tool_results.next().await {
1275 log::debug!("Tool finished {:?}", tool_result);
1276
1277 event_stream.update_tool_call_fields(
1278 &tool_result.tool_use_id,
1279 acp::ToolCallUpdateFields {
1280 status: Some(if tool_result.is_error {
1281 acp::ToolCallStatus::Failed
1282 } else {
1283 acp::ToolCallStatus::Completed
1284 }),
1285 raw_output: tool_result.output.clone(),
1286 ..Default::default()
1287 },
1288 );
1289 this.update(cx, |this, _cx| {
1290 this.pending_message()
1291 .tool_results
1292 .insert(tool_result.tool_use_id.clone(), tool_result);
1293 })?;
1294 }
1295
1296 this.update(cx, |this, cx| {
1297 this.flush_pending_message(cx);
1298 if this.title.is_none() && this.pending_title_generation.is_none() {
1299 this.generate_title(cx);
1300 }
1301 })?;
1302
1303 if let Some(error) = error {
1304 attempt += 1;
1305 let retry =
1306 this.update(cx, |this, _| this.handle_completion_error(error, attempt))??;
1307 let timer = cx.background_executor().timer(retry.duration);
1308 event_stream.send_retry(retry);
1309 timer.await;
1310 this.update(cx, |this, _cx| {
1311 if let Some(Message::Agent(message)) = this.messages.last() {
1312 if message.tool_results.is_empty() {
1313 intent = CompletionIntent::UserPrompt;
1314 this.messages.push(Message::Resume);
1315 }
1316 }
1317 })?;
1318 } else if this.read_with(cx, |this, _| this.tool_use_limit_reached)? {
1319 return Err(language_model::ToolUseLimitReachedError.into());
1320 } else if end_turn {
1321 return Ok(());
1322 } else {
1323 intent = CompletionIntent::ToolResults;
1324 attempt = 0;
1325 }
1326 }
1327 }
1328
1329 fn handle_completion_error(
1330 &mut self,
1331 error: LanguageModelCompletionError,
1332 attempt: u8,
1333 ) -> Result<acp_thread::RetryStatus> {
1334 if self.completion_mode == CompletionMode::Normal {
1335 return Err(anyhow!(error));
1336 }
1337
1338 let Some(strategy) = Self::retry_strategy_for(&error) else {
1339 return Err(anyhow!(error));
1340 };
1341
1342 let max_attempts = match &strategy {
1343 RetryStrategy::ExponentialBackoff { max_attempts, .. } => *max_attempts,
1344 RetryStrategy::Fixed { max_attempts, .. } => *max_attempts,
1345 };
1346
1347 if attempt > max_attempts {
1348 return Err(anyhow!(error));
1349 }
1350
1351 let delay = match &strategy {
1352 RetryStrategy::ExponentialBackoff { initial_delay, .. } => {
1353 let delay_secs = initial_delay.as_secs() * 2u64.pow((attempt - 1) as u32);
1354 Duration::from_secs(delay_secs)
1355 }
1356 RetryStrategy::Fixed { delay, .. } => *delay,
1357 };
1358 log::debug!("Retry attempt {attempt} with delay {delay:?}");
1359
1360 Ok(acp_thread::RetryStatus {
1361 last_error: error.to_string().into(),
1362 attempt: attempt as usize,
1363 max_attempts: max_attempts as usize,
1364 started_at: Instant::now(),
1365 duration: delay,
1366 })
1367 }
1368
1369 /// A helper method that's called on every streamed completion event.
1370 /// Returns an optional tool result task, which the main agentic loop will
1371 /// send back to the model when it resolves.
1372 fn handle_completion_event(
1373 &mut self,
1374 event: LanguageModelCompletionEvent,
1375 event_stream: &ThreadEventStream,
1376 cx: &mut Context<Self>,
1377 ) -> Result<Option<Task<LanguageModelToolResult>>> {
1378 log::trace!("Handling streamed completion event: {:?}", event);
1379 use LanguageModelCompletionEvent::*;
1380
1381 match event {
1382 StartMessage { .. } => {
1383 self.flush_pending_message(cx);
1384 self.pending_message = Some(AgentMessage::default());
1385 }
1386 Text(new_text) => self.handle_text_event(new_text, event_stream, cx),
1387 Thinking { text, signature } => {
1388 self.handle_thinking_event(text, signature, event_stream, cx)
1389 }
1390 RedactedThinking { data } => self.handle_redacted_thinking_event(data, cx),
1391 ToolUse(tool_use) => {
1392 return Ok(self.handle_tool_use_event(tool_use, event_stream, cx));
1393 }
1394 ToolUseJsonParseError {
1395 id,
1396 tool_name,
1397 raw_input,
1398 json_parse_error,
1399 } => {
1400 return Ok(Some(Task::ready(
1401 self.handle_tool_use_json_parse_error_event(
1402 id,
1403 tool_name,
1404 raw_input,
1405 json_parse_error,
1406 ),
1407 )));
1408 }
1409 UsageUpdate(usage) => {
1410 telemetry::event!(
1411 "Agent Thread Completion Usage Updated",
1412 thread_id = self.id.to_string(),
1413 prompt_id = self.prompt_id.to_string(),
1414 model = self.model.as_ref().map(|m| m.telemetry_id()),
1415 model_provider = self.model.as_ref().map(|m| m.provider_id().to_string()),
1416 input_tokens = usage.input_tokens,
1417 output_tokens = usage.output_tokens,
1418 cache_creation_input_tokens = usage.cache_creation_input_tokens,
1419 cache_read_input_tokens = usage.cache_read_input_tokens,
1420 );
1421 self.update_token_usage(usage, cx);
1422 }
1423 StatusUpdate(CompletionRequestStatus::UsageUpdated { amount, limit }) => {
1424 self.update_model_request_usage(amount, limit, cx);
1425 }
1426 StatusUpdate(
1427 CompletionRequestStatus::Started
1428 | CompletionRequestStatus::Queued { .. }
1429 | CompletionRequestStatus::Failed { .. },
1430 ) => {}
1431 StatusUpdate(CompletionRequestStatus::ToolUseLimitReached) => {
1432 self.tool_use_limit_reached = true;
1433 }
1434 Stop(StopReason::Refusal) => return Err(CompletionError::Refusal.into()),
1435 Stop(StopReason::MaxTokens) => return Err(CompletionError::MaxTokens.into()),
1436 Stop(StopReason::ToolUse | StopReason::EndTurn) => {}
1437 }
1438
1439 Ok(None)
1440 }
1441
1442 fn handle_text_event(
1443 &mut self,
1444 new_text: String,
1445 event_stream: &ThreadEventStream,
1446 cx: &mut Context<Self>,
1447 ) {
1448 event_stream.send_text(&new_text);
1449
1450 let last_message = self.pending_message();
1451 if let Some(AgentMessageContent::Text(text)) = last_message.content.last_mut() {
1452 text.push_str(&new_text);
1453 } else {
1454 last_message
1455 .content
1456 .push(AgentMessageContent::Text(new_text));
1457 }
1458
1459 cx.notify();
1460 }
1461
1462 fn handle_thinking_event(
1463 &mut self,
1464 new_text: String,
1465 new_signature: Option<String>,
1466 event_stream: &ThreadEventStream,
1467 cx: &mut Context<Self>,
1468 ) {
1469 event_stream.send_thinking(&new_text);
1470
1471 let last_message = self.pending_message();
1472 if let Some(AgentMessageContent::Thinking { text, signature }) =
1473 last_message.content.last_mut()
1474 {
1475 text.push_str(&new_text);
1476 *signature = new_signature.or(signature.take());
1477 } else {
1478 last_message.content.push(AgentMessageContent::Thinking {
1479 text: new_text,
1480 signature: new_signature,
1481 });
1482 }
1483
1484 cx.notify();
1485 }
1486
1487 fn handle_redacted_thinking_event(&mut self, data: String, cx: &mut Context<Self>) {
1488 let last_message = self.pending_message();
1489 last_message
1490 .content
1491 .push(AgentMessageContent::RedactedThinking(data));
1492 cx.notify();
1493 }
1494
1495 fn handle_tool_use_event(
1496 &mut self,
1497 tool_use: LanguageModelToolUse,
1498 event_stream: &ThreadEventStream,
1499 cx: &mut Context<Self>,
1500 ) -> Option<Task<LanguageModelToolResult>> {
1501 cx.notify();
1502
1503 let tool = self.tool(tool_use.name.as_ref());
1504 let mut title = SharedString::from(&tool_use.name);
1505 let mut kind = acp::ToolKind::Other;
1506 if let Some(tool) = tool.as_ref() {
1507 title = tool.initial_title(tool_use.input.clone(), cx);
1508 kind = tool.kind();
1509 }
1510
1511 // Ensure the last message ends in the current tool use
1512 let last_message = self.pending_message();
1513 let push_new_tool_use = last_message.content.last_mut().is_none_or(|content| {
1514 if let AgentMessageContent::ToolUse(last_tool_use) = content {
1515 if last_tool_use.id == tool_use.id {
1516 *last_tool_use = tool_use.clone();
1517 false
1518 } else {
1519 true
1520 }
1521 } else {
1522 true
1523 }
1524 });
1525
1526 if push_new_tool_use {
1527 event_stream.send_tool_call(&tool_use.id, title, kind, tool_use.input.clone());
1528 last_message
1529 .content
1530 .push(AgentMessageContent::ToolUse(tool_use.clone()));
1531 } else {
1532 event_stream.update_tool_call_fields(
1533 &tool_use.id,
1534 acp::ToolCallUpdateFields {
1535 title: Some(title.into()),
1536 kind: Some(kind),
1537 raw_input: Some(tool_use.input.clone()),
1538 ..Default::default()
1539 },
1540 );
1541 }
1542
1543 if !tool_use.is_input_complete {
1544 return None;
1545 }
1546
1547 let Some(tool) = tool else {
1548 let content = format!("No tool named {} exists", tool_use.name);
1549 return Some(Task::ready(LanguageModelToolResult {
1550 content: LanguageModelToolResultContent::Text(Arc::from(content)),
1551 tool_use_id: tool_use.id,
1552 tool_name: tool_use.name,
1553 is_error: true,
1554 output: None,
1555 }));
1556 };
1557
1558 let fs = self.project.read(cx).fs().clone();
1559 let tool_event_stream =
1560 ToolCallEventStream::new(tool_use.id.clone(), event_stream.clone(), Some(fs));
1561 tool_event_stream.update_fields(acp::ToolCallUpdateFields {
1562 status: Some(acp::ToolCallStatus::InProgress),
1563 ..Default::default()
1564 });
1565 let supports_images = self.model().is_some_and(|model| model.supports_images());
1566 let tool_result = tool.run(tool_use.input, tool_event_stream, cx);
1567 log::debug!("Running tool {}", tool_use.name);
1568 Some(cx.foreground_executor().spawn(async move {
1569 let tool_result = tool_result.await.and_then(|output| {
1570 if let LanguageModelToolResultContent::Image(_) = &output.llm_output
1571 && !supports_images
1572 {
1573 return Err(anyhow!(
1574 "Attempted to read an image, but this model doesn't support it.",
1575 ));
1576 }
1577 Ok(output)
1578 });
1579
1580 match tool_result {
1581 Ok(output) => LanguageModelToolResult {
1582 tool_use_id: tool_use.id,
1583 tool_name: tool_use.name,
1584 is_error: false,
1585 content: output.llm_output,
1586 output: Some(output.raw_output),
1587 },
1588 Err(error) => LanguageModelToolResult {
1589 tool_use_id: tool_use.id,
1590 tool_name: tool_use.name,
1591 is_error: true,
1592 content: LanguageModelToolResultContent::Text(Arc::from(error.to_string())),
1593 output: Some(error.to_string().into()),
1594 },
1595 }
1596 }))
1597 }
1598
1599 fn handle_tool_use_json_parse_error_event(
1600 &mut self,
1601 tool_use_id: LanguageModelToolUseId,
1602 tool_name: Arc<str>,
1603 raw_input: Arc<str>,
1604 json_parse_error: String,
1605 ) -> LanguageModelToolResult {
1606 let tool_output = format!("Error parsing input JSON: {json_parse_error}");
1607 LanguageModelToolResult {
1608 tool_use_id,
1609 tool_name,
1610 is_error: true,
1611 content: LanguageModelToolResultContent::Text(tool_output.into()),
1612 output: Some(serde_json::Value::String(raw_input.to_string())),
1613 }
1614 }
1615
1616 fn update_model_request_usage(&self, amount: usize, limit: UsageLimit, cx: &mut Context<Self>) {
1617 self.project
1618 .read(cx)
1619 .user_store()
1620 .update(cx, |user_store, cx| {
1621 user_store.update_model_request_usage(
1622 ModelRequestUsage(RequestUsage {
1623 amount: amount as i32,
1624 limit,
1625 }),
1626 cx,
1627 )
1628 });
1629 }
1630
1631 pub fn title(&self) -> SharedString {
1632 self.title.clone().unwrap_or("New Thread".into())
1633 }
1634
1635 pub fn summary(&mut self, cx: &mut Context<Self>) -> Task<Result<SharedString>> {
1636 if let Some(summary) = self.summary.as_ref() {
1637 return Task::ready(Ok(summary.clone()));
1638 }
1639 let Some(model) = self.summarization_model.clone() else {
1640 return Task::ready(Err(anyhow!("No summarization model available")));
1641 };
1642 let mut request = LanguageModelRequest {
1643 intent: Some(CompletionIntent::ThreadContextSummarization),
1644 temperature: AgentSettings::temperature_for_model(&model, cx),
1645 ..Default::default()
1646 };
1647
1648 for message in &self.messages {
1649 request.messages.extend(message.to_request());
1650 }
1651
1652 request.messages.push(LanguageModelRequestMessage {
1653 role: Role::User,
1654 content: vec![SUMMARIZE_THREAD_DETAILED_PROMPT.into()],
1655 cache: false,
1656 });
1657 cx.spawn(async move |this, cx| {
1658 let mut summary = String::new();
1659 let mut messages = model.stream_completion(request, cx).await?;
1660 while let Some(event) = messages.next().await {
1661 let event = event?;
1662 let text = match event {
1663 LanguageModelCompletionEvent::Text(text) => text,
1664 LanguageModelCompletionEvent::StatusUpdate(
1665 CompletionRequestStatus::UsageUpdated { amount, limit },
1666 ) => {
1667 this.update(cx, |thread, cx| {
1668 thread.update_model_request_usage(amount, limit, cx);
1669 })?;
1670 continue;
1671 }
1672 _ => continue,
1673 };
1674
1675 let mut lines = text.lines();
1676 summary.extend(lines.next());
1677 }
1678
1679 log::debug!("Setting summary: {}", summary);
1680 let summary = SharedString::from(summary);
1681
1682 this.update(cx, |this, cx| {
1683 this.summary = Some(summary.clone());
1684 cx.notify()
1685 })?;
1686
1687 Ok(summary)
1688 })
1689 }
1690
1691 fn generate_title(&mut self, cx: &mut Context<Self>) {
1692 let Some(model) = self.summarization_model.clone() else {
1693 return;
1694 };
1695
1696 log::debug!(
1697 "Generating title with model: {:?}",
1698 self.summarization_model.as_ref().map(|model| model.name())
1699 );
1700 let mut request = LanguageModelRequest {
1701 intent: Some(CompletionIntent::ThreadSummarization),
1702 temperature: AgentSettings::temperature_for_model(&model, cx),
1703 ..Default::default()
1704 };
1705
1706 for message in &self.messages {
1707 request.messages.extend(message.to_request());
1708 }
1709
1710 request.messages.push(LanguageModelRequestMessage {
1711 role: Role::User,
1712 content: vec![SUMMARIZE_THREAD_PROMPT.into()],
1713 cache: false,
1714 });
1715 self.pending_title_generation = Some(cx.spawn(async move |this, cx| {
1716 let mut title = String::new();
1717
1718 let generate = async {
1719 let mut messages = model.stream_completion(request, cx).await?;
1720 while let Some(event) = messages.next().await {
1721 let event = event?;
1722 let text = match event {
1723 LanguageModelCompletionEvent::Text(text) => text,
1724 LanguageModelCompletionEvent::StatusUpdate(
1725 CompletionRequestStatus::UsageUpdated { amount, limit },
1726 ) => {
1727 this.update(cx, |thread, cx| {
1728 thread.update_model_request_usage(amount, limit, cx);
1729 })?;
1730 continue;
1731 }
1732 _ => continue,
1733 };
1734
1735 let mut lines = text.lines();
1736 title.extend(lines.next());
1737
1738 // Stop if the LLM generated multiple lines.
1739 if lines.next().is_some() {
1740 break;
1741 }
1742 }
1743 anyhow::Ok(())
1744 };
1745
1746 if generate.await.context("failed to generate title").is_ok() {
1747 _ = this.update(cx, |this, cx| this.set_title(title.into(), cx));
1748 }
1749 _ = this.update(cx, |this, _| this.pending_title_generation = None);
1750 }));
1751 }
1752
1753 pub fn set_title(&mut self, title: SharedString, cx: &mut Context<Self>) {
1754 self.pending_title_generation = None;
1755 if Some(&title) != self.title.as_ref() {
1756 self.title = Some(title);
1757 cx.emit(TitleUpdated);
1758 cx.notify();
1759 }
1760 }
1761
1762 fn last_user_message(&self) -> Option<&UserMessage> {
1763 self.messages
1764 .iter()
1765 .rev()
1766 .find_map(|message| match message {
1767 Message::User(user_message) => Some(user_message),
1768 Message::Agent(_) => None,
1769 Message::Resume => None,
1770 })
1771 }
1772
1773 fn pending_message(&mut self) -> &mut AgentMessage {
1774 self.pending_message.get_or_insert_default()
1775 }
1776
1777 fn flush_pending_message(&mut self, cx: &mut Context<Self>) {
1778 let Some(mut message) = self.pending_message.take() else {
1779 return;
1780 };
1781
1782 if message.content.is_empty() {
1783 return;
1784 }
1785
1786 for content in &message.content {
1787 let AgentMessageContent::ToolUse(tool_use) = content else {
1788 continue;
1789 };
1790
1791 if !message.tool_results.contains_key(&tool_use.id) {
1792 message.tool_results.insert(
1793 tool_use.id.clone(),
1794 LanguageModelToolResult {
1795 tool_use_id: tool_use.id.clone(),
1796 tool_name: tool_use.name.clone(),
1797 is_error: true,
1798 content: LanguageModelToolResultContent::Text(TOOL_CANCELED_MESSAGE.into()),
1799 output: None,
1800 },
1801 );
1802 }
1803 }
1804
1805 self.messages.push(Message::Agent(message));
1806 self.updated_at = Utc::now();
1807 self.summary = None;
1808 cx.notify()
1809 }
1810
1811 pub(crate) fn build_completion_request(
1812 &self,
1813 completion_intent: CompletionIntent,
1814 cx: &App,
1815 ) -> Result<LanguageModelRequest> {
1816 let model = self.model().context("No language model configured")?;
1817 let tools = if let Some(turn) = self.running_turn.as_ref() {
1818 turn.tools
1819 .iter()
1820 .filter_map(|(tool_name, tool)| {
1821 log::trace!("Including tool: {}", tool_name);
1822 Some(LanguageModelRequestTool {
1823 name: tool_name.to_string(),
1824 description: tool.description().to_string(),
1825 input_schema: tool.input_schema(model.tool_input_format()).log_err()?,
1826 })
1827 })
1828 .collect::<Vec<_>>()
1829 } else {
1830 Vec::new()
1831 };
1832
1833 log::debug!("Building completion request");
1834 log::debug!("Completion intent: {:?}", completion_intent);
1835 log::debug!("Completion mode: {:?}", self.completion_mode);
1836
1837 let messages = self.build_request_messages(cx);
1838 log::debug!("Request will include {} messages", messages.len());
1839 log::debug!("Request includes {} tools", tools.len());
1840
1841 let request = LanguageModelRequest {
1842 thread_id: Some(self.id.to_string()),
1843 prompt_id: Some(self.prompt_id.to_string()),
1844 intent: Some(completion_intent),
1845 mode: Some(self.completion_mode.into()),
1846 messages,
1847 tools,
1848 tool_choice: None,
1849 stop: Vec::new(),
1850 temperature: AgentSettings::temperature_for_model(model, cx),
1851 thinking_allowed: true,
1852 };
1853
1854 log::debug!("Completion request built successfully");
1855 Ok(request)
1856 }
1857
1858 fn enabled_tools(
1859 &self,
1860 profile: &AgentProfileSettings,
1861 model: &Arc<dyn LanguageModel>,
1862 cx: &App,
1863 ) -> BTreeMap<SharedString, Arc<dyn AnyAgentTool>> {
1864 fn truncate(tool_name: &SharedString) -> SharedString {
1865 if tool_name.len() > MAX_TOOL_NAME_LENGTH {
1866 let mut truncated = tool_name.to_string();
1867 truncated.truncate(MAX_TOOL_NAME_LENGTH);
1868 truncated.into()
1869 } else {
1870 tool_name.clone()
1871 }
1872 }
1873
1874 let mut tools = self
1875 .tools
1876 .iter()
1877 .filter_map(|(tool_name, tool)| {
1878 if tool.supported_provider(&model.provider_id())
1879 && profile.is_tool_enabled(tool_name)
1880 {
1881 Some((truncate(tool_name), tool.clone()))
1882 } else {
1883 None
1884 }
1885 })
1886 .collect::<BTreeMap<_, _>>();
1887
1888 let mut context_server_tools = Vec::new();
1889 let mut seen_tools = tools.keys().cloned().collect::<HashSet<_>>();
1890 let mut duplicate_tool_names = HashSet::default();
1891 for (server_id, server_tools) in self.context_server_registry.read(cx).servers() {
1892 for (tool_name, tool) in server_tools {
1893 if profile.is_context_server_tool_enabled(&server_id.0, &tool_name) {
1894 let tool_name = truncate(tool_name);
1895 if !seen_tools.insert(tool_name.clone()) {
1896 duplicate_tool_names.insert(tool_name.clone());
1897 }
1898 context_server_tools.push((server_id.clone(), tool_name, tool.clone()));
1899 }
1900 }
1901 }
1902
1903 // When there are duplicate tool names, disambiguate by prefixing them
1904 // with the server ID. In the rare case there isn't enough space for the
1905 // disambiguated tool name, keep only the last tool with this name.
1906 for (server_id, tool_name, tool) in context_server_tools {
1907 if duplicate_tool_names.contains(&tool_name) {
1908 let available = MAX_TOOL_NAME_LENGTH.saturating_sub(tool_name.len());
1909 if available >= 2 {
1910 let mut disambiguated = server_id.0.to_string();
1911 disambiguated.truncate(available - 1);
1912 disambiguated.push('_');
1913 disambiguated.push_str(&tool_name);
1914 tools.insert(disambiguated.into(), tool.clone());
1915 } else {
1916 tools.insert(tool_name, tool.clone());
1917 }
1918 } else {
1919 tools.insert(tool_name, tool.clone());
1920 }
1921 }
1922
1923 tools
1924 }
1925
1926 fn tool(&self, name: &str) -> Option<Arc<dyn AnyAgentTool>> {
1927 self.running_turn.as_ref()?.tools.get(name).cloned()
1928 }
1929
1930 fn build_request_messages(&self, cx: &App) -> Vec<LanguageModelRequestMessage> {
1931 log::trace!(
1932 "Building request messages from {} thread messages",
1933 self.messages.len()
1934 );
1935
1936 let system_prompt = SystemPromptTemplate {
1937 project: self.project_context.read(cx),
1938 available_tools: self.tools.keys().cloned().collect(),
1939 }
1940 .render(&self.templates)
1941 .context("failed to build system prompt")
1942 .expect("Invalid template");
1943 let mut messages = vec![LanguageModelRequestMessage {
1944 role: Role::System,
1945 content: vec![system_prompt.into()],
1946 cache: false,
1947 }];
1948 for message in &self.messages {
1949 messages.extend(message.to_request());
1950 }
1951
1952 if let Some(last_message) = messages.last_mut() {
1953 last_message.cache = true;
1954 }
1955
1956 if let Some(message) = self.pending_message.as_ref() {
1957 messages.extend(message.to_request());
1958 }
1959
1960 messages
1961 }
1962
1963 pub fn to_markdown(&self) -> String {
1964 let mut markdown = String::new();
1965 for (ix, message) in self.messages.iter().enumerate() {
1966 if ix > 0 {
1967 markdown.push('\n');
1968 }
1969 markdown.push_str(&message.to_markdown());
1970 }
1971
1972 if let Some(message) = self.pending_message.as_ref() {
1973 markdown.push('\n');
1974 markdown.push_str(&message.to_markdown());
1975 }
1976
1977 markdown
1978 }
1979
1980 fn advance_prompt_id(&mut self) {
1981 self.prompt_id = PromptId::new();
1982 }
1983
1984 fn retry_strategy_for(error: &LanguageModelCompletionError) -> Option<RetryStrategy> {
1985 use LanguageModelCompletionError::*;
1986 use http_client::StatusCode;
1987
1988 // General strategy here:
1989 // - If retrying won't help (e.g. invalid API key or payload too large), return None so we don't retry at all.
1990 // - If it's a time-based issue (e.g. server overloaded, rate limit exceeded), retry up to 4 times with exponential backoff.
1991 // - If it's an issue that *might* be fixed by retrying (e.g. internal server error), retry up to 3 times.
1992 match error {
1993 HttpResponseError {
1994 status_code: StatusCode::TOO_MANY_REQUESTS,
1995 ..
1996 } => Some(RetryStrategy::ExponentialBackoff {
1997 initial_delay: BASE_RETRY_DELAY,
1998 max_attempts: MAX_RETRY_ATTEMPTS,
1999 }),
2000 ServerOverloaded { retry_after, .. } | RateLimitExceeded { retry_after, .. } => {
2001 Some(RetryStrategy::Fixed {
2002 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
2003 max_attempts: MAX_RETRY_ATTEMPTS,
2004 })
2005 }
2006 UpstreamProviderError {
2007 status,
2008 retry_after,
2009 ..
2010 } => match *status {
2011 StatusCode::TOO_MANY_REQUESTS | StatusCode::SERVICE_UNAVAILABLE => {
2012 Some(RetryStrategy::Fixed {
2013 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
2014 max_attempts: MAX_RETRY_ATTEMPTS,
2015 })
2016 }
2017 StatusCode::INTERNAL_SERVER_ERROR => Some(RetryStrategy::Fixed {
2018 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
2019 // Internal Server Error could be anything, retry up to 3 times.
2020 max_attempts: 3,
2021 }),
2022 status => {
2023 // There is no StatusCode variant for the unofficial HTTP 529 ("The service is overloaded"),
2024 // but we frequently get them in practice. See https://http.dev/529
2025 if status.as_u16() == 529 {
2026 Some(RetryStrategy::Fixed {
2027 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
2028 max_attempts: MAX_RETRY_ATTEMPTS,
2029 })
2030 } else {
2031 Some(RetryStrategy::Fixed {
2032 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
2033 max_attempts: 2,
2034 })
2035 }
2036 }
2037 },
2038 ApiInternalServerError { .. } => Some(RetryStrategy::Fixed {
2039 delay: BASE_RETRY_DELAY,
2040 max_attempts: 3,
2041 }),
2042 ApiReadResponseError { .. }
2043 | HttpSend { .. }
2044 | DeserializeResponse { .. }
2045 | BadRequestFormat { .. } => Some(RetryStrategy::Fixed {
2046 delay: BASE_RETRY_DELAY,
2047 max_attempts: 3,
2048 }),
2049 // Retrying these errors definitely shouldn't help.
2050 HttpResponseError {
2051 status_code:
2052 StatusCode::PAYLOAD_TOO_LARGE | StatusCode::FORBIDDEN | StatusCode::UNAUTHORIZED,
2053 ..
2054 }
2055 | AuthenticationError { .. }
2056 | PermissionError { .. }
2057 | NoApiKey { .. }
2058 | ApiEndpointNotFound { .. }
2059 | PromptTooLarge { .. } => None,
2060 // These errors might be transient, so retry them
2061 SerializeRequest { .. } | BuildRequestBody { .. } => Some(RetryStrategy::Fixed {
2062 delay: BASE_RETRY_DELAY,
2063 max_attempts: 1,
2064 }),
2065 // Retry all other 4xx and 5xx errors once.
2066 HttpResponseError { status_code, .. }
2067 if status_code.is_client_error() || status_code.is_server_error() =>
2068 {
2069 Some(RetryStrategy::Fixed {
2070 delay: BASE_RETRY_DELAY,
2071 max_attempts: 3,
2072 })
2073 }
2074 Other(err)
2075 if err.is::<language_model::PaymentRequiredError>()
2076 || err.is::<language_model::ModelRequestLimitReachedError>() =>
2077 {
2078 // Retrying won't help for Payment Required or Model Request Limit errors (where
2079 // the user must upgrade to usage-based billing to get more requests, or else wait
2080 // for a significant amount of time for the request limit to reset).
2081 None
2082 }
2083 // Conservatively assume that any other errors are non-retryable
2084 HttpResponseError { .. } | Other(..) => Some(RetryStrategy::Fixed {
2085 delay: BASE_RETRY_DELAY,
2086 max_attempts: 2,
2087 }),
2088 }
2089 }
2090}
2091
2092struct RunningTurn {
2093 /// Holds the task that handles agent interaction until the end of the turn.
2094 /// Survives across multiple requests as the model performs tool calls and
2095 /// we run tools, report their results.
2096 _task: Task<()>,
2097 /// The current event stream for the running turn. Used to report a final
2098 /// cancellation event if we cancel the turn.
2099 event_stream: ThreadEventStream,
2100 /// The tools that were enabled for this turn.
2101 tools: BTreeMap<SharedString, Arc<dyn AnyAgentTool>>,
2102}
2103
2104impl RunningTurn {
2105 fn cancel(self) {
2106 log::debug!("Cancelling in progress turn");
2107 self.event_stream.send_canceled();
2108 }
2109}
2110
2111pub struct TokenUsageUpdated(pub Option<acp_thread::TokenUsage>);
2112
2113impl EventEmitter<TokenUsageUpdated> for Thread {}
2114
2115pub struct TitleUpdated;
2116
2117impl EventEmitter<TitleUpdated> for Thread {}
2118
2119pub trait AgentTool
2120where
2121 Self: 'static + Sized,
2122{
2123 type Input: for<'de> Deserialize<'de> + Serialize + JsonSchema;
2124 type Output: for<'de> Deserialize<'de> + Serialize + Into<LanguageModelToolResultContent>;
2125
2126 fn name() -> &'static str;
2127
2128 fn description(&self) -> SharedString {
2129 let schema = schemars::schema_for!(Self::Input);
2130 SharedString::new(
2131 schema
2132 .get("description")
2133 .and_then(|description| description.as_str())
2134 .unwrap_or_default(),
2135 )
2136 }
2137
2138 fn kind() -> acp::ToolKind;
2139
2140 /// The initial tool title to display. Can be updated during the tool run.
2141 fn initial_title(
2142 &self,
2143 input: Result<Self::Input, serde_json::Value>,
2144 cx: &mut App,
2145 ) -> SharedString;
2146
2147 /// Returns the JSON schema that describes the tool's input.
2148 fn input_schema(&self, format: LanguageModelToolSchemaFormat) -> Schema {
2149 crate::tool_schema::root_schema_for::<Self::Input>(format)
2150 }
2151
2152 /// Some tools rely on a provider for the underlying billing or other reasons.
2153 /// Allow the tool to check if they are compatible, or should be filtered out.
2154 fn supported_provider(&self, _provider: &LanguageModelProviderId) -> bool {
2155 true
2156 }
2157
2158 /// Runs the tool with the provided input.
2159 fn run(
2160 self: Arc<Self>,
2161 input: Self::Input,
2162 event_stream: ToolCallEventStream,
2163 cx: &mut App,
2164 ) -> Task<Result<Self::Output>>;
2165
2166 /// Emits events for a previous execution of the tool.
2167 fn replay(
2168 &self,
2169 _input: Self::Input,
2170 _output: Self::Output,
2171 _event_stream: ToolCallEventStream,
2172 _cx: &mut App,
2173 ) -> Result<()> {
2174 Ok(())
2175 }
2176
2177 fn erase(self) -> Arc<dyn AnyAgentTool> {
2178 Arc::new(Erased(Arc::new(self)))
2179 }
2180}
2181
2182pub struct Erased<T>(T);
2183
2184pub struct AgentToolOutput {
2185 pub llm_output: LanguageModelToolResultContent,
2186 pub raw_output: serde_json::Value,
2187}
2188
2189pub trait AnyAgentTool {
2190 fn name(&self) -> SharedString;
2191 fn description(&self) -> SharedString;
2192 fn kind(&self) -> acp::ToolKind;
2193 fn initial_title(&self, input: serde_json::Value, _cx: &mut App) -> SharedString;
2194 fn input_schema(&self, format: LanguageModelToolSchemaFormat) -> Result<serde_json::Value>;
2195 fn supported_provider(&self, _provider: &LanguageModelProviderId) -> bool {
2196 true
2197 }
2198 fn run(
2199 self: Arc<Self>,
2200 input: serde_json::Value,
2201 event_stream: ToolCallEventStream,
2202 cx: &mut App,
2203 ) -> Task<Result<AgentToolOutput>>;
2204 fn replay(
2205 &self,
2206 input: serde_json::Value,
2207 output: serde_json::Value,
2208 event_stream: ToolCallEventStream,
2209 cx: &mut App,
2210 ) -> Result<()>;
2211}
2212
2213impl<T> AnyAgentTool for Erased<Arc<T>>
2214where
2215 T: AgentTool,
2216{
2217 fn name(&self) -> SharedString {
2218 T::name().into()
2219 }
2220
2221 fn description(&self) -> SharedString {
2222 self.0.description()
2223 }
2224
2225 fn kind(&self) -> agent_client_protocol::ToolKind {
2226 T::kind()
2227 }
2228
2229 fn initial_title(&self, input: serde_json::Value, _cx: &mut App) -> SharedString {
2230 let parsed_input = serde_json::from_value(input.clone()).map_err(|_| input);
2231 self.0.initial_title(parsed_input, _cx)
2232 }
2233
2234 fn input_schema(&self, format: LanguageModelToolSchemaFormat) -> Result<serde_json::Value> {
2235 let mut json = serde_json::to_value(self.0.input_schema(format))?;
2236 adapt_schema_to_format(&mut json, format)?;
2237 Ok(json)
2238 }
2239
2240 fn supported_provider(&self, provider: &LanguageModelProviderId) -> bool {
2241 self.0.supported_provider(provider)
2242 }
2243
2244 fn run(
2245 self: Arc<Self>,
2246 input: serde_json::Value,
2247 event_stream: ToolCallEventStream,
2248 cx: &mut App,
2249 ) -> Task<Result<AgentToolOutput>> {
2250 cx.spawn(async move |cx| {
2251 let input = serde_json::from_value(input)?;
2252 let output = cx
2253 .update(|cx| self.0.clone().run(input, event_stream, cx))?
2254 .await?;
2255 let raw_output = serde_json::to_value(&output)?;
2256 Ok(AgentToolOutput {
2257 llm_output: output.into(),
2258 raw_output,
2259 })
2260 })
2261 }
2262
2263 fn replay(
2264 &self,
2265 input: serde_json::Value,
2266 output: serde_json::Value,
2267 event_stream: ToolCallEventStream,
2268 cx: &mut App,
2269 ) -> Result<()> {
2270 let input = serde_json::from_value(input)?;
2271 let output = serde_json::from_value(output)?;
2272 self.0.replay(input, output, event_stream, cx)
2273 }
2274}
2275
2276#[derive(Clone)]
2277struct ThreadEventStream(mpsc::UnboundedSender<Result<ThreadEvent>>);
2278
2279impl ThreadEventStream {
2280 fn send_user_message(&self, message: &UserMessage) {
2281 self.0
2282 .unbounded_send(Ok(ThreadEvent::UserMessage(message.clone())))
2283 .ok();
2284 }
2285
2286 fn send_text(&self, text: &str) {
2287 self.0
2288 .unbounded_send(Ok(ThreadEvent::AgentText(text.to_string())))
2289 .ok();
2290 }
2291
2292 fn send_thinking(&self, text: &str) {
2293 self.0
2294 .unbounded_send(Ok(ThreadEvent::AgentThinking(text.to_string())))
2295 .ok();
2296 }
2297
2298 fn send_tool_call(
2299 &self,
2300 id: &LanguageModelToolUseId,
2301 title: SharedString,
2302 kind: acp::ToolKind,
2303 input: serde_json::Value,
2304 ) {
2305 self.0
2306 .unbounded_send(Ok(ThreadEvent::ToolCall(Self::initial_tool_call(
2307 id,
2308 title.to_string(),
2309 kind,
2310 input,
2311 ))))
2312 .ok();
2313 }
2314
2315 fn initial_tool_call(
2316 id: &LanguageModelToolUseId,
2317 title: String,
2318 kind: acp::ToolKind,
2319 input: serde_json::Value,
2320 ) -> acp::ToolCall {
2321 acp::ToolCall {
2322 meta: None,
2323 id: acp::ToolCallId(id.to_string().into()),
2324 title,
2325 kind,
2326 status: acp::ToolCallStatus::Pending,
2327 content: vec![],
2328 locations: vec![],
2329 raw_input: Some(input),
2330 raw_output: None,
2331 }
2332 }
2333
2334 fn update_tool_call_fields(
2335 &self,
2336 tool_use_id: &LanguageModelToolUseId,
2337 fields: acp::ToolCallUpdateFields,
2338 ) {
2339 self.0
2340 .unbounded_send(Ok(ThreadEvent::ToolCallUpdate(
2341 acp::ToolCallUpdate {
2342 meta: None,
2343 id: acp::ToolCallId(tool_use_id.to_string().into()),
2344 fields,
2345 }
2346 .into(),
2347 )))
2348 .ok();
2349 }
2350
2351 fn send_retry(&self, status: acp_thread::RetryStatus) {
2352 self.0.unbounded_send(Ok(ThreadEvent::Retry(status))).ok();
2353 }
2354
2355 fn send_stop(&self, reason: acp::StopReason) {
2356 self.0.unbounded_send(Ok(ThreadEvent::Stop(reason))).ok();
2357 }
2358
2359 fn send_canceled(&self) {
2360 self.0
2361 .unbounded_send(Ok(ThreadEvent::Stop(acp::StopReason::Cancelled)))
2362 .ok();
2363 }
2364
2365 fn send_error(&self, error: impl Into<anyhow::Error>) {
2366 self.0.unbounded_send(Err(error.into())).ok();
2367 }
2368}
2369
2370#[derive(Clone)]
2371pub struct ToolCallEventStream {
2372 tool_use_id: LanguageModelToolUseId,
2373 stream: ThreadEventStream,
2374 fs: Option<Arc<dyn Fs>>,
2375}
2376
2377impl ToolCallEventStream {
2378 #[cfg(test)]
2379 pub fn test() -> (Self, ToolCallEventStreamReceiver) {
2380 let (events_tx, events_rx) = mpsc::unbounded::<Result<ThreadEvent>>();
2381
2382 let stream = ToolCallEventStream::new("test_id".into(), ThreadEventStream(events_tx), None);
2383
2384 (stream, ToolCallEventStreamReceiver(events_rx))
2385 }
2386
2387 fn new(
2388 tool_use_id: LanguageModelToolUseId,
2389 stream: ThreadEventStream,
2390 fs: Option<Arc<dyn Fs>>,
2391 ) -> Self {
2392 Self {
2393 tool_use_id,
2394 stream,
2395 fs,
2396 }
2397 }
2398
2399 pub fn update_fields(&self, fields: acp::ToolCallUpdateFields) {
2400 self.stream
2401 .update_tool_call_fields(&self.tool_use_id, fields);
2402 }
2403
2404 pub fn update_diff(&self, diff: Entity<acp_thread::Diff>) {
2405 self.stream
2406 .0
2407 .unbounded_send(Ok(ThreadEvent::ToolCallUpdate(
2408 acp_thread::ToolCallUpdateDiff {
2409 id: acp::ToolCallId(self.tool_use_id.to_string().into()),
2410 diff,
2411 }
2412 .into(),
2413 )))
2414 .ok();
2415 }
2416
2417 pub fn authorize(&self, title: impl Into<String>, cx: &mut App) -> Task<Result<()>> {
2418 if agent_settings::AgentSettings::get_global(cx).always_allow_tool_actions {
2419 return Task::ready(Ok(()));
2420 }
2421
2422 let (response_tx, response_rx) = oneshot::channel();
2423 self.stream
2424 .0
2425 .unbounded_send(Ok(ThreadEvent::ToolCallAuthorization(
2426 ToolCallAuthorization {
2427 tool_call: acp::ToolCallUpdate {
2428 meta: None,
2429 id: acp::ToolCallId(self.tool_use_id.to_string().into()),
2430 fields: acp::ToolCallUpdateFields {
2431 title: Some(title.into()),
2432 ..Default::default()
2433 },
2434 },
2435 options: vec![
2436 acp::PermissionOption {
2437 id: acp::PermissionOptionId("always_allow".into()),
2438 name: "Always Allow".into(),
2439 kind: acp::PermissionOptionKind::AllowAlways,
2440 meta: None,
2441 },
2442 acp::PermissionOption {
2443 id: acp::PermissionOptionId("allow".into()),
2444 name: "Allow".into(),
2445 kind: acp::PermissionOptionKind::AllowOnce,
2446 meta: None,
2447 },
2448 acp::PermissionOption {
2449 id: acp::PermissionOptionId("deny".into()),
2450 name: "Deny".into(),
2451 kind: acp::PermissionOptionKind::RejectOnce,
2452 meta: None,
2453 },
2454 ],
2455 response: response_tx,
2456 },
2457 )))
2458 .ok();
2459 let fs = self.fs.clone();
2460 cx.spawn(async move |cx| match response_rx.await?.0.as_ref() {
2461 "always_allow" => {
2462 if let Some(fs) = fs.clone() {
2463 cx.update(|cx| {
2464 update_settings_file(fs, cx, |settings, _| {
2465 settings
2466 .agent
2467 .get_or_insert_default()
2468 .set_always_allow_tool_actions(true);
2469 });
2470 })?;
2471 }
2472
2473 Ok(())
2474 }
2475 "allow" => Ok(()),
2476 _ => Err(anyhow!("Permission to run tool denied by user")),
2477 })
2478 }
2479}
2480
2481#[cfg(test)]
2482pub struct ToolCallEventStreamReceiver(mpsc::UnboundedReceiver<Result<ThreadEvent>>);
2483
2484#[cfg(test)]
2485impl ToolCallEventStreamReceiver {
2486 pub async fn expect_authorization(&mut self) -> ToolCallAuthorization {
2487 let event = self.0.next().await;
2488 if let Some(Ok(ThreadEvent::ToolCallAuthorization(auth))) = event {
2489 auth
2490 } else {
2491 panic!("Expected ToolCallAuthorization but got: {:?}", event);
2492 }
2493 }
2494
2495 pub async fn expect_update_fields(&mut self) -> acp::ToolCallUpdateFields {
2496 let event = self.0.next().await;
2497 if let Some(Ok(ThreadEvent::ToolCallUpdate(acp_thread::ToolCallUpdate::UpdateFields(
2498 update,
2499 )))) = event
2500 {
2501 update.fields
2502 } else {
2503 panic!("Expected update fields but got: {:?}", event);
2504 }
2505 }
2506
2507 pub async fn expect_diff(&mut self) -> Entity<acp_thread::Diff> {
2508 let event = self.0.next().await;
2509 if let Some(Ok(ThreadEvent::ToolCallUpdate(acp_thread::ToolCallUpdate::UpdateDiff(
2510 update,
2511 )))) = event
2512 {
2513 update.diff
2514 } else {
2515 panic!("Expected diff but got: {:?}", event);
2516 }
2517 }
2518
2519 pub async fn expect_terminal(&mut self) -> Entity<acp_thread::Terminal> {
2520 let event = self.0.next().await;
2521 if let Some(Ok(ThreadEvent::ToolCallUpdate(acp_thread::ToolCallUpdate::UpdateTerminal(
2522 update,
2523 )))) = event
2524 {
2525 update.terminal
2526 } else {
2527 panic!("Expected terminal but got: {:?}", event);
2528 }
2529 }
2530}
2531
2532#[cfg(test)]
2533impl std::ops::Deref for ToolCallEventStreamReceiver {
2534 type Target = mpsc::UnboundedReceiver<Result<ThreadEvent>>;
2535
2536 fn deref(&self) -> &Self::Target {
2537 &self.0
2538 }
2539}
2540
2541#[cfg(test)]
2542impl std::ops::DerefMut for ToolCallEventStreamReceiver {
2543 fn deref_mut(&mut self) -> &mut Self::Target {
2544 &mut self.0
2545 }
2546}
2547
2548impl From<&str> for UserMessageContent {
2549 fn from(text: &str) -> Self {
2550 Self::Text(text.into())
2551 }
2552}
2553
2554impl From<acp::ContentBlock> for UserMessageContent {
2555 fn from(value: acp::ContentBlock) -> Self {
2556 match value {
2557 acp::ContentBlock::Text(text_content) => Self::Text(text_content.text),
2558 acp::ContentBlock::Image(image_content) => Self::Image(convert_image(image_content)),
2559 acp::ContentBlock::Audio(_) => {
2560 // TODO
2561 Self::Text("[audio]".to_string())
2562 }
2563 acp::ContentBlock::ResourceLink(resource_link) => {
2564 match MentionUri::parse(&resource_link.uri) {
2565 Ok(uri) => Self::Mention {
2566 uri,
2567 content: String::new(),
2568 },
2569 Err(err) => {
2570 log::error!("Failed to parse mention link: {}", err);
2571 Self::Text(format!("[{}]({})", resource_link.name, resource_link.uri))
2572 }
2573 }
2574 }
2575 acp::ContentBlock::Resource(resource) => match resource.resource {
2576 acp::EmbeddedResourceResource::TextResourceContents(resource) => {
2577 match MentionUri::parse(&resource.uri) {
2578 Ok(uri) => Self::Mention {
2579 uri,
2580 content: resource.text,
2581 },
2582 Err(err) => {
2583 log::error!("Failed to parse mention link: {}", err);
2584 Self::Text(
2585 MarkdownCodeBlock {
2586 tag: &resource.uri,
2587 text: &resource.text,
2588 }
2589 .to_string(),
2590 )
2591 }
2592 }
2593 }
2594 acp::EmbeddedResourceResource::BlobResourceContents(_) => {
2595 // TODO
2596 Self::Text("[blob]".to_string())
2597 }
2598 },
2599 }
2600 }
2601}
2602
2603impl From<UserMessageContent> for acp::ContentBlock {
2604 fn from(content: UserMessageContent) -> Self {
2605 match content {
2606 UserMessageContent::Text(text) => acp::ContentBlock::Text(acp::TextContent {
2607 text,
2608 annotations: None,
2609 meta: None,
2610 }),
2611 UserMessageContent::Image(image) => acp::ContentBlock::Image(acp::ImageContent {
2612 data: image.source.to_string(),
2613 mime_type: "image/png".to_string(),
2614 meta: None,
2615 annotations: None,
2616 uri: None,
2617 }),
2618 UserMessageContent::Mention { uri, content } => {
2619 acp::ContentBlock::Resource(acp::EmbeddedResource {
2620 meta: None,
2621 resource: acp::EmbeddedResourceResource::TextResourceContents(
2622 acp::TextResourceContents {
2623 meta: None,
2624 mime_type: None,
2625 text: content,
2626 uri: uri.to_uri().to_string(),
2627 },
2628 ),
2629 annotations: None,
2630 })
2631 }
2632 }
2633 }
2634}
2635
2636fn convert_image(image_content: acp::ImageContent) -> LanguageModelImage {
2637 LanguageModelImage {
2638 source: image_content.data.into(),
2639 // TODO: make this optional?
2640 size: gpui::Size::new(0.into(), 0.into()),
2641 }
2642}