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 |_, cx| {
883 let worktree_snapshots = futures::future::join_all(worktree_snapshots).await;
884
885 let mut unsaved_buffers = Vec::new();
886 cx.update(|app_cx| {
887 let buffer_store = project.read(app_cx).buffer_store();
888 for buffer_handle in buffer_store.read(app_cx).buffers() {
889 let buffer = buffer_handle.read(app_cx);
890 if buffer.is_dirty()
891 && let Some(file) = buffer.file()
892 {
893 let path = file.path().to_string_lossy().to_string();
894 unsaved_buffers.push(path);
895 }
896 }
897 })
898 .ok();
899
900 Arc::new(ProjectSnapshot {
901 worktree_snapshots,
902 unsaved_buffer_paths: unsaved_buffers,
903 timestamp: Utc::now(),
904 })
905 })
906 }
907
908 fn worktree_snapshot(
909 worktree: Entity<project::Worktree>,
910 git_store: Entity<GitStore>,
911 cx: &App,
912 ) -> Task<agent::thread::WorktreeSnapshot> {
913 cx.spawn(async move |cx| {
914 // Get worktree path and snapshot
915 let worktree_info = cx.update(|app_cx| {
916 let worktree = worktree.read(app_cx);
917 let path = worktree.abs_path().to_string_lossy().to_string();
918 let snapshot = worktree.snapshot();
919 (path, snapshot)
920 });
921
922 let Ok((worktree_path, _snapshot)) = worktree_info else {
923 return WorktreeSnapshot {
924 worktree_path: String::new(),
925 git_state: None,
926 };
927 };
928
929 let git_state = git_store
930 .update(cx, |git_store, cx| {
931 git_store
932 .repositories()
933 .values()
934 .find(|repo| {
935 repo.read(cx)
936 .abs_path_to_repo_path(&worktree.read(cx).abs_path())
937 .is_some()
938 })
939 .cloned()
940 })
941 .ok()
942 .flatten()
943 .map(|repo| {
944 repo.update(cx, |repo, _| {
945 let current_branch =
946 repo.branch.as_ref().map(|branch| branch.name().to_owned());
947 repo.send_job(None, |state, _| async move {
948 let RepositoryState::Local { backend, .. } = state else {
949 return GitState {
950 remote_url: None,
951 head_sha: None,
952 current_branch,
953 diff: None,
954 };
955 };
956
957 let remote_url = backend.remote_url("origin");
958 let head_sha = backend.head_sha().await;
959 let diff = backend.diff(DiffType::HeadToWorktree).await.ok();
960
961 GitState {
962 remote_url,
963 head_sha,
964 current_branch,
965 diff,
966 }
967 })
968 })
969 });
970
971 let git_state = match git_state {
972 Some(git_state) => match git_state.ok() {
973 Some(git_state) => git_state.await.ok(),
974 None => None,
975 },
976 None => None,
977 };
978
979 WorktreeSnapshot {
980 worktree_path,
981 git_state,
982 }
983 })
984 }
985
986 pub fn project_context(&self) -> &Entity<ProjectContext> {
987 &self.project_context
988 }
989
990 pub fn project(&self) -> &Entity<Project> {
991 &self.project
992 }
993
994 pub fn action_log(&self) -> &Entity<ActionLog> {
995 &self.action_log
996 }
997
998 pub fn is_empty(&self) -> bool {
999 self.messages.is_empty() && self.title.is_none()
1000 }
1001
1002 pub fn model(&self) -> Option<&Arc<dyn LanguageModel>> {
1003 self.model.as_ref()
1004 }
1005
1006 pub fn set_model(&mut self, model: Arc<dyn LanguageModel>, cx: &mut Context<Self>) {
1007 let old_usage = self.latest_token_usage();
1008 self.model = Some(model);
1009 let new_caps = Self::prompt_capabilities(self.model.as_deref());
1010 let new_usage = self.latest_token_usage();
1011 if old_usage != new_usage {
1012 cx.emit(TokenUsageUpdated(new_usage));
1013 }
1014 self.prompt_capabilities_tx.send(new_caps).log_err();
1015 cx.notify()
1016 }
1017
1018 pub fn summarization_model(&self) -> Option<&Arc<dyn LanguageModel>> {
1019 self.summarization_model.as_ref()
1020 }
1021
1022 pub fn set_summarization_model(
1023 &mut self,
1024 model: Option<Arc<dyn LanguageModel>>,
1025 cx: &mut Context<Self>,
1026 ) {
1027 self.summarization_model = model;
1028 cx.notify()
1029 }
1030
1031 pub fn completion_mode(&self) -> CompletionMode {
1032 self.completion_mode
1033 }
1034
1035 pub fn set_completion_mode(&mut self, mode: CompletionMode, cx: &mut Context<Self>) {
1036 let old_usage = self.latest_token_usage();
1037 self.completion_mode = mode;
1038 let new_usage = self.latest_token_usage();
1039 if old_usage != new_usage {
1040 cx.emit(TokenUsageUpdated(new_usage));
1041 }
1042 cx.notify()
1043 }
1044
1045 #[cfg(any(test, feature = "test-support"))]
1046 pub fn last_message(&self) -> Option<Message> {
1047 if let Some(message) = self.pending_message.clone() {
1048 Some(Message::Agent(message))
1049 } else {
1050 self.messages.last().cloned()
1051 }
1052 }
1053
1054 pub fn add_default_tools(
1055 &mut self,
1056 environment: Rc<dyn ThreadEnvironment>,
1057 cx: &mut Context<Self>,
1058 ) {
1059 let language_registry = self.project.read(cx).languages().clone();
1060 self.add_tool(CopyPathTool::new(self.project.clone()));
1061 self.add_tool(CreateDirectoryTool::new(self.project.clone()));
1062 self.add_tool(DeletePathTool::new(
1063 self.project.clone(),
1064 self.action_log.clone(),
1065 ));
1066 self.add_tool(DiagnosticsTool::new(self.project.clone()));
1067 self.add_tool(EditFileTool::new(
1068 self.project.clone(),
1069 cx.weak_entity(),
1070 language_registry,
1071 ));
1072 self.add_tool(FetchTool::new(self.project.read(cx).client().http_client()));
1073 self.add_tool(FindPathTool::new(self.project.clone()));
1074 self.add_tool(GrepTool::new(self.project.clone()));
1075 self.add_tool(ListDirectoryTool::new(self.project.clone()));
1076 self.add_tool(MovePathTool::new(self.project.clone()));
1077 self.add_tool(NowTool);
1078 self.add_tool(OpenTool::new(self.project.clone()));
1079 self.add_tool(ReadFileTool::new(
1080 self.project.clone(),
1081 self.action_log.clone(),
1082 ));
1083 self.add_tool(TerminalTool::new(self.project.clone(), environment));
1084 self.add_tool(ThinkingTool);
1085 self.add_tool(WebSearchTool);
1086 }
1087
1088 pub fn add_tool<T: AgentTool>(&mut self, tool: T) {
1089 self.tools.insert(T::name().into(), tool.erase());
1090 }
1091
1092 pub fn remove_tool(&mut self, name: &str) -> bool {
1093 self.tools.remove(name).is_some()
1094 }
1095
1096 pub fn profile(&self) -> &AgentProfileId {
1097 &self.profile_id
1098 }
1099
1100 pub fn set_profile(&mut self, profile_id: AgentProfileId) {
1101 self.profile_id = profile_id;
1102 }
1103
1104 pub fn cancel(&mut self, cx: &mut Context<Self>) {
1105 if let Some(running_turn) = self.running_turn.take() {
1106 running_turn.cancel();
1107 }
1108 self.flush_pending_message(cx);
1109 }
1110
1111 fn update_token_usage(&mut self, update: language_model::TokenUsage, cx: &mut Context<Self>) {
1112 let Some(last_user_message) = self.last_user_message() else {
1113 return;
1114 };
1115
1116 self.request_token_usage
1117 .insert(last_user_message.id.clone(), update);
1118 cx.emit(TokenUsageUpdated(self.latest_token_usage()));
1119 cx.notify();
1120 }
1121
1122 pub fn truncate(&mut self, message_id: UserMessageId, cx: &mut Context<Self>) -> Result<()> {
1123 self.cancel(cx);
1124 let Some(position) = self.messages.iter().position(
1125 |msg| matches!(msg, Message::User(UserMessage { id, .. }) if id == &message_id),
1126 ) else {
1127 return Err(anyhow!("Message not found"));
1128 };
1129
1130 for message in self.messages.drain(position..) {
1131 match message {
1132 Message::User(message) => {
1133 self.request_token_usage.remove(&message.id);
1134 }
1135 Message::Agent(_) | Message::Resume => {}
1136 }
1137 }
1138 self.summary = None;
1139 cx.notify();
1140 Ok(())
1141 }
1142
1143 pub fn latest_token_usage(&self) -> Option<acp_thread::TokenUsage> {
1144 let last_user_message = self.last_user_message()?;
1145 let tokens = self.request_token_usage.get(&last_user_message.id)?;
1146 let model = self.model.clone()?;
1147
1148 Some(acp_thread::TokenUsage {
1149 max_tokens: model.max_token_count_for_mode(self.completion_mode.into()),
1150 used_tokens: tokens.total_tokens(),
1151 })
1152 }
1153
1154 pub fn resume(
1155 &mut self,
1156 cx: &mut Context<Self>,
1157 ) -> Result<mpsc::UnboundedReceiver<Result<ThreadEvent>>> {
1158 self.messages.push(Message::Resume);
1159 cx.notify();
1160
1161 log::debug!("Total messages in thread: {}", self.messages.len());
1162 self.run_turn(cx)
1163 }
1164
1165 /// Sending a message results in the model streaming a response, which could include tool calls.
1166 /// After calling tools, the model will stops and waits for any outstanding tool calls to be completed and their results sent.
1167 /// The returned channel will report all the occurrences in which the model stops before erroring or ending its turn.
1168 pub fn send<T>(
1169 &mut self,
1170 id: UserMessageId,
1171 content: impl IntoIterator<Item = T>,
1172 cx: &mut Context<Self>,
1173 ) -> Result<mpsc::UnboundedReceiver<Result<ThreadEvent>>>
1174 where
1175 T: Into<UserMessageContent>,
1176 {
1177 let model = self.model().context("No language model configured")?;
1178
1179 log::info!("Thread::send called with model: {}", model.name().0);
1180 self.advance_prompt_id();
1181
1182 let content = content.into_iter().map(Into::into).collect::<Vec<_>>();
1183 log::debug!("Thread::send content: {:?}", content);
1184
1185 self.messages
1186 .push(Message::User(UserMessage { id, content }));
1187 cx.notify();
1188
1189 log::debug!("Total messages in thread: {}", self.messages.len());
1190 self.run_turn(cx)
1191 }
1192
1193 fn run_turn(
1194 &mut self,
1195 cx: &mut Context<Self>,
1196 ) -> Result<mpsc::UnboundedReceiver<Result<ThreadEvent>>> {
1197 self.cancel(cx);
1198
1199 let model = self.model.clone().context("No language model configured")?;
1200 let profile = AgentSettings::get_global(cx)
1201 .profiles
1202 .get(&self.profile_id)
1203 .context("Profile not found")?;
1204 let (events_tx, events_rx) = mpsc::unbounded::<Result<ThreadEvent>>();
1205 let event_stream = ThreadEventStream(events_tx);
1206 let message_ix = self.messages.len().saturating_sub(1);
1207 self.tool_use_limit_reached = false;
1208 self.summary = None;
1209 self.running_turn = Some(RunningTurn {
1210 event_stream: event_stream.clone(),
1211 tools: self.enabled_tools(profile, &model, cx),
1212 _task: cx.spawn(async move |this, cx| {
1213 log::debug!("Starting agent turn execution");
1214
1215 let turn_result = Self::run_turn_internal(&this, model, &event_stream, cx).await;
1216 _ = this.update(cx, |this, cx| this.flush_pending_message(cx));
1217
1218 match turn_result {
1219 Ok(()) => {
1220 log::debug!("Turn execution completed");
1221 event_stream.send_stop(acp::StopReason::EndTurn);
1222 }
1223 Err(error) => {
1224 log::error!("Turn execution failed: {:?}", error);
1225 match error.downcast::<CompletionError>() {
1226 Ok(CompletionError::Refusal) => {
1227 event_stream.send_stop(acp::StopReason::Refusal);
1228 _ = this.update(cx, |this, _| this.messages.truncate(message_ix));
1229 }
1230 Ok(CompletionError::MaxTokens) => {
1231 event_stream.send_stop(acp::StopReason::MaxTokens);
1232 }
1233 Ok(CompletionError::Other(error)) | Err(error) => {
1234 event_stream.send_error(error);
1235 }
1236 }
1237 }
1238 }
1239
1240 _ = this.update(cx, |this, _| this.running_turn.take());
1241 }),
1242 });
1243 Ok(events_rx)
1244 }
1245
1246 async fn run_turn_internal(
1247 this: &WeakEntity<Self>,
1248 model: Arc<dyn LanguageModel>,
1249 event_stream: &ThreadEventStream,
1250 cx: &mut AsyncApp,
1251 ) -> Result<()> {
1252 let mut attempt = 0;
1253 let mut intent = CompletionIntent::UserPrompt;
1254 loop {
1255 let request =
1256 this.update(cx, |this, cx| this.build_completion_request(intent, cx))??;
1257
1258 telemetry::event!(
1259 "Agent Thread Completion",
1260 thread_id = this.read_with(cx, |this, _| this.id.to_string())?,
1261 prompt_id = this.read_with(cx, |this, _| this.prompt_id.to_string())?,
1262 model = model.telemetry_id(),
1263 model_provider = model.provider_id().to_string(),
1264 attempt
1265 );
1266
1267 log::debug!("Calling model.stream_completion, attempt {}", attempt);
1268 let mut events = model
1269 .stream_completion(request, cx)
1270 .await
1271 .map_err(|error| anyhow!(error))?;
1272 let mut tool_results = FuturesUnordered::new();
1273 let mut error = None;
1274 while let Some(event) = events.next().await {
1275 log::trace!("Received completion event: {:?}", event);
1276 match event {
1277 Ok(event) => {
1278 tool_results.extend(this.update(cx, |this, cx| {
1279 this.handle_completion_event(event, event_stream, cx)
1280 })??);
1281 }
1282 Err(err) => {
1283 error = Some(err);
1284 break;
1285 }
1286 }
1287 }
1288
1289 let end_turn = tool_results.is_empty();
1290 while let Some(tool_result) = tool_results.next().await {
1291 log::debug!("Tool finished {:?}", tool_result);
1292
1293 event_stream.update_tool_call_fields(
1294 &tool_result.tool_use_id,
1295 acp::ToolCallUpdateFields {
1296 status: Some(if tool_result.is_error {
1297 acp::ToolCallStatus::Failed
1298 } else {
1299 acp::ToolCallStatus::Completed
1300 }),
1301 raw_output: tool_result.output.clone(),
1302 ..Default::default()
1303 },
1304 );
1305 this.update(cx, |this, _cx| {
1306 this.pending_message()
1307 .tool_results
1308 .insert(tool_result.tool_use_id.clone(), tool_result);
1309 })?;
1310 }
1311
1312 this.update(cx, |this, cx| {
1313 this.flush_pending_message(cx);
1314 if this.title.is_none() && this.pending_title_generation.is_none() {
1315 this.generate_title(cx);
1316 }
1317 })?;
1318
1319 if let Some(error) = error {
1320 attempt += 1;
1321 let retry =
1322 this.update(cx, |this, _| this.handle_completion_error(error, attempt))??;
1323 let timer = cx.background_executor().timer(retry.duration);
1324 event_stream.send_retry(retry);
1325 timer.await;
1326 this.update(cx, |this, _cx| {
1327 if let Some(Message::Agent(message)) = this.messages.last() {
1328 if message.tool_results.is_empty() {
1329 intent = CompletionIntent::UserPrompt;
1330 this.messages.push(Message::Resume);
1331 }
1332 }
1333 })?;
1334 } else if this.read_with(cx, |this, _| this.tool_use_limit_reached)? {
1335 return Err(language_model::ToolUseLimitReachedError.into());
1336 } else if end_turn {
1337 return Ok(());
1338 } else {
1339 intent = CompletionIntent::ToolResults;
1340 attempt = 0;
1341 }
1342 }
1343 }
1344
1345 fn handle_completion_error(
1346 &mut self,
1347 error: LanguageModelCompletionError,
1348 attempt: u8,
1349 ) -> Result<acp_thread::RetryStatus> {
1350 if self.completion_mode == CompletionMode::Normal {
1351 return Err(anyhow!(error));
1352 }
1353
1354 let Some(strategy) = Self::retry_strategy_for(&error) else {
1355 return Err(anyhow!(error));
1356 };
1357
1358 let max_attempts = match &strategy {
1359 RetryStrategy::ExponentialBackoff { max_attempts, .. } => *max_attempts,
1360 RetryStrategy::Fixed { max_attempts, .. } => *max_attempts,
1361 };
1362
1363 if attempt > max_attempts {
1364 return Err(anyhow!(error));
1365 }
1366
1367 let delay = match &strategy {
1368 RetryStrategy::ExponentialBackoff { initial_delay, .. } => {
1369 let delay_secs = initial_delay.as_secs() * 2u64.pow((attempt - 1) as u32);
1370 Duration::from_secs(delay_secs)
1371 }
1372 RetryStrategy::Fixed { delay, .. } => *delay,
1373 };
1374 log::debug!("Retry attempt {attempt} with delay {delay:?}");
1375
1376 Ok(acp_thread::RetryStatus {
1377 last_error: error.to_string().into(),
1378 attempt: attempt as usize,
1379 max_attempts: max_attempts as usize,
1380 started_at: Instant::now(),
1381 duration: delay,
1382 })
1383 }
1384
1385 /// A helper method that's called on every streamed completion event.
1386 /// Returns an optional tool result task, which the main agentic loop will
1387 /// send back to the model when it resolves.
1388 fn handle_completion_event(
1389 &mut self,
1390 event: LanguageModelCompletionEvent,
1391 event_stream: &ThreadEventStream,
1392 cx: &mut Context<Self>,
1393 ) -> Result<Option<Task<LanguageModelToolResult>>> {
1394 log::trace!("Handling streamed completion event: {:?}", event);
1395 use LanguageModelCompletionEvent::*;
1396
1397 match event {
1398 StartMessage { .. } => {
1399 self.flush_pending_message(cx);
1400 self.pending_message = Some(AgentMessage::default());
1401 }
1402 Text(new_text) => self.handle_text_event(new_text, event_stream, cx),
1403 Thinking { text, signature } => {
1404 self.handle_thinking_event(text, signature, event_stream, cx)
1405 }
1406 RedactedThinking { data } => self.handle_redacted_thinking_event(data, cx),
1407 ToolUse(tool_use) => {
1408 return Ok(self.handle_tool_use_event(tool_use, event_stream, cx));
1409 }
1410 ToolUseJsonParseError {
1411 id,
1412 tool_name,
1413 raw_input,
1414 json_parse_error,
1415 } => {
1416 return Ok(Some(Task::ready(
1417 self.handle_tool_use_json_parse_error_event(
1418 id,
1419 tool_name,
1420 raw_input,
1421 json_parse_error,
1422 ),
1423 )));
1424 }
1425 UsageUpdate(usage) => {
1426 telemetry::event!(
1427 "Agent Thread Completion Usage Updated",
1428 thread_id = self.id.to_string(),
1429 prompt_id = self.prompt_id.to_string(),
1430 model = self.model.as_ref().map(|m| m.telemetry_id()),
1431 model_provider = self.model.as_ref().map(|m| m.provider_id().to_string()),
1432 input_tokens = usage.input_tokens,
1433 output_tokens = usage.output_tokens,
1434 cache_creation_input_tokens = usage.cache_creation_input_tokens,
1435 cache_read_input_tokens = usage.cache_read_input_tokens,
1436 );
1437 self.update_token_usage(usage, cx);
1438 }
1439 StatusUpdate(CompletionRequestStatus::UsageUpdated { amount, limit }) => {
1440 self.update_model_request_usage(amount, limit, cx);
1441 }
1442 StatusUpdate(
1443 CompletionRequestStatus::Started
1444 | CompletionRequestStatus::Queued { .. }
1445 | CompletionRequestStatus::Failed { .. },
1446 ) => {}
1447 StatusUpdate(CompletionRequestStatus::ToolUseLimitReached) => {
1448 self.tool_use_limit_reached = true;
1449 }
1450 Stop(StopReason::Refusal) => return Err(CompletionError::Refusal.into()),
1451 Stop(StopReason::MaxTokens) => return Err(CompletionError::MaxTokens.into()),
1452 Stop(StopReason::ToolUse | StopReason::EndTurn) => {}
1453 }
1454
1455 Ok(None)
1456 }
1457
1458 fn handle_text_event(
1459 &mut self,
1460 new_text: String,
1461 event_stream: &ThreadEventStream,
1462 cx: &mut Context<Self>,
1463 ) {
1464 event_stream.send_text(&new_text);
1465
1466 let last_message = self.pending_message();
1467 if let Some(AgentMessageContent::Text(text)) = last_message.content.last_mut() {
1468 text.push_str(&new_text);
1469 } else {
1470 last_message
1471 .content
1472 .push(AgentMessageContent::Text(new_text));
1473 }
1474
1475 cx.notify();
1476 }
1477
1478 fn handle_thinking_event(
1479 &mut self,
1480 new_text: String,
1481 new_signature: Option<String>,
1482 event_stream: &ThreadEventStream,
1483 cx: &mut Context<Self>,
1484 ) {
1485 event_stream.send_thinking(&new_text);
1486
1487 let last_message = self.pending_message();
1488 if let Some(AgentMessageContent::Thinking { text, signature }) =
1489 last_message.content.last_mut()
1490 {
1491 text.push_str(&new_text);
1492 *signature = new_signature.or(signature.take());
1493 } else {
1494 last_message.content.push(AgentMessageContent::Thinking {
1495 text: new_text,
1496 signature: new_signature,
1497 });
1498 }
1499
1500 cx.notify();
1501 }
1502
1503 fn handle_redacted_thinking_event(&mut self, data: String, cx: &mut Context<Self>) {
1504 let last_message = self.pending_message();
1505 last_message
1506 .content
1507 .push(AgentMessageContent::RedactedThinking(data));
1508 cx.notify();
1509 }
1510
1511 fn handle_tool_use_event(
1512 &mut self,
1513 tool_use: LanguageModelToolUse,
1514 event_stream: &ThreadEventStream,
1515 cx: &mut Context<Self>,
1516 ) -> Option<Task<LanguageModelToolResult>> {
1517 cx.notify();
1518
1519 let tool = self.tool(tool_use.name.as_ref());
1520 let mut title = SharedString::from(&tool_use.name);
1521 let mut kind = acp::ToolKind::Other;
1522 if let Some(tool) = tool.as_ref() {
1523 title = tool.initial_title(tool_use.input.clone(), cx);
1524 kind = tool.kind();
1525 }
1526
1527 // Ensure the last message ends in the current tool use
1528 let last_message = self.pending_message();
1529 let push_new_tool_use = last_message.content.last_mut().is_none_or(|content| {
1530 if let AgentMessageContent::ToolUse(last_tool_use) = content {
1531 if last_tool_use.id == tool_use.id {
1532 *last_tool_use = tool_use.clone();
1533 false
1534 } else {
1535 true
1536 }
1537 } else {
1538 true
1539 }
1540 });
1541
1542 if push_new_tool_use {
1543 event_stream.send_tool_call(&tool_use.id, title, kind, tool_use.input.clone());
1544 last_message
1545 .content
1546 .push(AgentMessageContent::ToolUse(tool_use.clone()));
1547 } else {
1548 event_stream.update_tool_call_fields(
1549 &tool_use.id,
1550 acp::ToolCallUpdateFields {
1551 title: Some(title.into()),
1552 kind: Some(kind),
1553 raw_input: Some(tool_use.input.clone()),
1554 ..Default::default()
1555 },
1556 );
1557 }
1558
1559 if !tool_use.is_input_complete {
1560 return None;
1561 }
1562
1563 let Some(tool) = tool else {
1564 let content = format!("No tool named {} exists", tool_use.name);
1565 return Some(Task::ready(LanguageModelToolResult {
1566 content: LanguageModelToolResultContent::Text(Arc::from(content)),
1567 tool_use_id: tool_use.id,
1568 tool_name: tool_use.name,
1569 is_error: true,
1570 output: None,
1571 }));
1572 };
1573
1574 let fs = self.project.read(cx).fs().clone();
1575 let tool_event_stream =
1576 ToolCallEventStream::new(tool_use.id.clone(), event_stream.clone(), Some(fs));
1577 tool_event_stream.update_fields(acp::ToolCallUpdateFields {
1578 status: Some(acp::ToolCallStatus::InProgress),
1579 ..Default::default()
1580 });
1581 let supports_images = self.model().is_some_and(|model| model.supports_images());
1582 let tool_result = tool.run(tool_use.input, tool_event_stream, cx);
1583 log::debug!("Running tool {}", tool_use.name);
1584 Some(cx.foreground_executor().spawn(async move {
1585 let tool_result = tool_result.await.and_then(|output| {
1586 if let LanguageModelToolResultContent::Image(_) = &output.llm_output
1587 && !supports_images
1588 {
1589 return Err(anyhow!(
1590 "Attempted to read an image, but this model doesn't support it.",
1591 ));
1592 }
1593 Ok(output)
1594 });
1595
1596 match tool_result {
1597 Ok(output) => LanguageModelToolResult {
1598 tool_use_id: tool_use.id,
1599 tool_name: tool_use.name,
1600 is_error: false,
1601 content: output.llm_output,
1602 output: Some(output.raw_output),
1603 },
1604 Err(error) => LanguageModelToolResult {
1605 tool_use_id: tool_use.id,
1606 tool_name: tool_use.name,
1607 is_error: true,
1608 content: LanguageModelToolResultContent::Text(Arc::from(error.to_string())),
1609 output: Some(error.to_string().into()),
1610 },
1611 }
1612 }))
1613 }
1614
1615 fn handle_tool_use_json_parse_error_event(
1616 &mut self,
1617 tool_use_id: LanguageModelToolUseId,
1618 tool_name: Arc<str>,
1619 raw_input: Arc<str>,
1620 json_parse_error: String,
1621 ) -> LanguageModelToolResult {
1622 let tool_output = format!("Error parsing input JSON: {json_parse_error}");
1623 LanguageModelToolResult {
1624 tool_use_id,
1625 tool_name,
1626 is_error: true,
1627 content: LanguageModelToolResultContent::Text(tool_output.into()),
1628 output: Some(serde_json::Value::String(raw_input.to_string())),
1629 }
1630 }
1631
1632 fn update_model_request_usage(&self, amount: usize, limit: UsageLimit, cx: &mut Context<Self>) {
1633 self.project
1634 .read(cx)
1635 .user_store()
1636 .update(cx, |user_store, cx| {
1637 user_store.update_model_request_usage(
1638 ModelRequestUsage(RequestUsage {
1639 amount: amount as i32,
1640 limit,
1641 }),
1642 cx,
1643 )
1644 });
1645 }
1646
1647 pub fn title(&self) -> SharedString {
1648 self.title.clone().unwrap_or("New Thread".into())
1649 }
1650
1651 pub fn summary(&mut self, cx: &mut Context<Self>) -> Task<Result<SharedString>> {
1652 if let Some(summary) = self.summary.as_ref() {
1653 return Task::ready(Ok(summary.clone()));
1654 }
1655 let Some(model) = self.summarization_model.clone() else {
1656 return Task::ready(Err(anyhow!("No summarization model available")));
1657 };
1658 let mut request = LanguageModelRequest {
1659 intent: Some(CompletionIntent::ThreadContextSummarization),
1660 temperature: AgentSettings::temperature_for_model(&model, cx),
1661 ..Default::default()
1662 };
1663
1664 for message in &self.messages {
1665 request.messages.extend(message.to_request());
1666 }
1667
1668 request.messages.push(LanguageModelRequestMessage {
1669 role: Role::User,
1670 content: vec![SUMMARIZE_THREAD_DETAILED_PROMPT.into()],
1671 cache: false,
1672 });
1673 cx.spawn(async move |this, cx| {
1674 let mut summary = String::new();
1675 let mut messages = model.stream_completion(request, cx).await?;
1676 while let Some(event) = messages.next().await {
1677 let event = event?;
1678 let text = match event {
1679 LanguageModelCompletionEvent::Text(text) => text,
1680 LanguageModelCompletionEvent::StatusUpdate(
1681 CompletionRequestStatus::UsageUpdated { amount, limit },
1682 ) => {
1683 this.update(cx, |thread, cx| {
1684 thread.update_model_request_usage(amount, limit, cx);
1685 })?;
1686 continue;
1687 }
1688 _ => continue,
1689 };
1690
1691 let mut lines = text.lines();
1692 summary.extend(lines.next());
1693 }
1694
1695 log::debug!("Setting summary: {}", summary);
1696 let summary = SharedString::from(summary);
1697
1698 this.update(cx, |this, cx| {
1699 this.summary = Some(summary.clone());
1700 cx.notify()
1701 })?;
1702
1703 Ok(summary)
1704 })
1705 }
1706
1707 fn generate_title(&mut self, cx: &mut Context<Self>) {
1708 let Some(model) = self.summarization_model.clone() else {
1709 return;
1710 };
1711
1712 log::debug!(
1713 "Generating title with model: {:?}",
1714 self.summarization_model.as_ref().map(|model| model.name())
1715 );
1716 let mut request = LanguageModelRequest {
1717 intent: Some(CompletionIntent::ThreadSummarization),
1718 temperature: AgentSettings::temperature_for_model(&model, cx),
1719 ..Default::default()
1720 };
1721
1722 for message in &self.messages {
1723 request.messages.extend(message.to_request());
1724 }
1725
1726 request.messages.push(LanguageModelRequestMessage {
1727 role: Role::User,
1728 content: vec![SUMMARIZE_THREAD_PROMPT.into()],
1729 cache: false,
1730 });
1731 self.pending_title_generation = Some(cx.spawn(async move |this, cx| {
1732 let mut title = String::new();
1733
1734 let generate = async {
1735 let mut messages = model.stream_completion(request, cx).await?;
1736 while let Some(event) = messages.next().await {
1737 let event = event?;
1738 let text = match event {
1739 LanguageModelCompletionEvent::Text(text) => text,
1740 LanguageModelCompletionEvent::StatusUpdate(
1741 CompletionRequestStatus::UsageUpdated { amount, limit },
1742 ) => {
1743 this.update(cx, |thread, cx| {
1744 thread.update_model_request_usage(amount, limit, cx);
1745 })?;
1746 continue;
1747 }
1748 _ => continue,
1749 };
1750
1751 let mut lines = text.lines();
1752 title.extend(lines.next());
1753
1754 // Stop if the LLM generated multiple lines.
1755 if lines.next().is_some() {
1756 break;
1757 }
1758 }
1759 anyhow::Ok(())
1760 };
1761
1762 if generate.await.context("failed to generate title").is_ok() {
1763 _ = this.update(cx, |this, cx| this.set_title(title.into(), cx));
1764 }
1765 _ = this.update(cx, |this, _| this.pending_title_generation = None);
1766 }));
1767 }
1768
1769 pub fn set_title(&mut self, title: SharedString, cx: &mut Context<Self>) {
1770 self.pending_title_generation = None;
1771 if Some(&title) != self.title.as_ref() {
1772 self.title = Some(title);
1773 cx.emit(TitleUpdated);
1774 cx.notify();
1775 }
1776 }
1777
1778 fn last_user_message(&self) -> Option<&UserMessage> {
1779 self.messages
1780 .iter()
1781 .rev()
1782 .find_map(|message| match message {
1783 Message::User(user_message) => Some(user_message),
1784 Message::Agent(_) => None,
1785 Message::Resume => None,
1786 })
1787 }
1788
1789 fn pending_message(&mut self) -> &mut AgentMessage {
1790 self.pending_message.get_or_insert_default()
1791 }
1792
1793 fn flush_pending_message(&mut self, cx: &mut Context<Self>) {
1794 let Some(mut message) = self.pending_message.take() else {
1795 return;
1796 };
1797
1798 if message.content.is_empty() {
1799 return;
1800 }
1801
1802 for content in &message.content {
1803 let AgentMessageContent::ToolUse(tool_use) = content else {
1804 continue;
1805 };
1806
1807 if !message.tool_results.contains_key(&tool_use.id) {
1808 message.tool_results.insert(
1809 tool_use.id.clone(),
1810 LanguageModelToolResult {
1811 tool_use_id: tool_use.id.clone(),
1812 tool_name: tool_use.name.clone(),
1813 is_error: true,
1814 content: LanguageModelToolResultContent::Text(TOOL_CANCELED_MESSAGE.into()),
1815 output: None,
1816 },
1817 );
1818 }
1819 }
1820
1821 self.messages.push(Message::Agent(message));
1822 self.updated_at = Utc::now();
1823 self.summary = None;
1824 cx.notify()
1825 }
1826
1827 pub(crate) fn build_completion_request(
1828 &self,
1829 completion_intent: CompletionIntent,
1830 cx: &App,
1831 ) -> Result<LanguageModelRequest> {
1832 let model = self.model().context("No language model configured")?;
1833 let tools = if let Some(turn) = self.running_turn.as_ref() {
1834 turn.tools
1835 .iter()
1836 .filter_map(|(tool_name, tool)| {
1837 log::trace!("Including tool: {}", tool_name);
1838 Some(LanguageModelRequestTool {
1839 name: tool_name.to_string(),
1840 description: tool.description().to_string(),
1841 input_schema: tool.input_schema(model.tool_input_format()).log_err()?,
1842 })
1843 })
1844 .collect::<Vec<_>>()
1845 } else {
1846 Vec::new()
1847 };
1848
1849 log::debug!("Building completion request");
1850 log::debug!("Completion intent: {:?}", completion_intent);
1851 log::debug!("Completion mode: {:?}", self.completion_mode);
1852
1853 let messages = self.build_request_messages(cx);
1854 log::debug!("Request will include {} messages", messages.len());
1855 log::debug!("Request includes {} tools", tools.len());
1856
1857 let request = LanguageModelRequest {
1858 thread_id: Some(self.id.to_string()),
1859 prompt_id: Some(self.prompt_id.to_string()),
1860 intent: Some(completion_intent),
1861 mode: Some(self.completion_mode.into()),
1862 messages,
1863 tools,
1864 tool_choice: None,
1865 stop: Vec::new(),
1866 temperature: AgentSettings::temperature_for_model(model, cx),
1867 thinking_allowed: true,
1868 };
1869
1870 log::debug!("Completion request built successfully");
1871 Ok(request)
1872 }
1873
1874 fn enabled_tools(
1875 &self,
1876 profile: &AgentProfileSettings,
1877 model: &Arc<dyn LanguageModel>,
1878 cx: &App,
1879 ) -> BTreeMap<SharedString, Arc<dyn AnyAgentTool>> {
1880 fn truncate(tool_name: &SharedString) -> SharedString {
1881 if tool_name.len() > MAX_TOOL_NAME_LENGTH {
1882 let mut truncated = tool_name.to_string();
1883 truncated.truncate(MAX_TOOL_NAME_LENGTH);
1884 truncated.into()
1885 } else {
1886 tool_name.clone()
1887 }
1888 }
1889
1890 let mut tools = self
1891 .tools
1892 .iter()
1893 .filter_map(|(tool_name, tool)| {
1894 if tool.supported_provider(&model.provider_id())
1895 && profile.is_tool_enabled(tool_name)
1896 {
1897 Some((truncate(tool_name), tool.clone()))
1898 } else {
1899 None
1900 }
1901 })
1902 .collect::<BTreeMap<_, _>>();
1903
1904 let mut context_server_tools = Vec::new();
1905 let mut seen_tools = tools.keys().cloned().collect::<HashSet<_>>();
1906 let mut duplicate_tool_names = HashSet::default();
1907 for (server_id, server_tools) in self.context_server_registry.read(cx).servers() {
1908 for (tool_name, tool) in server_tools {
1909 if profile.is_context_server_tool_enabled(&server_id.0, &tool_name) {
1910 let tool_name = truncate(tool_name);
1911 if !seen_tools.insert(tool_name.clone()) {
1912 duplicate_tool_names.insert(tool_name.clone());
1913 }
1914 context_server_tools.push((server_id.clone(), tool_name, tool.clone()));
1915 }
1916 }
1917 }
1918
1919 // When there are duplicate tool names, disambiguate by prefixing them
1920 // with the server ID. In the rare case there isn't enough space for the
1921 // disambiguated tool name, keep only the last tool with this name.
1922 for (server_id, tool_name, tool) in context_server_tools {
1923 if duplicate_tool_names.contains(&tool_name) {
1924 let available = MAX_TOOL_NAME_LENGTH.saturating_sub(tool_name.len());
1925 if available >= 2 {
1926 let mut disambiguated = server_id.0.to_string();
1927 disambiguated.truncate(available - 1);
1928 disambiguated.push('_');
1929 disambiguated.push_str(&tool_name);
1930 tools.insert(disambiguated.into(), tool.clone());
1931 } else {
1932 tools.insert(tool_name, tool.clone());
1933 }
1934 } else {
1935 tools.insert(tool_name, tool.clone());
1936 }
1937 }
1938
1939 tools
1940 }
1941
1942 fn tool(&self, name: &str) -> Option<Arc<dyn AnyAgentTool>> {
1943 self.running_turn.as_ref()?.tools.get(name).cloned()
1944 }
1945
1946 fn build_request_messages(&self, cx: &App) -> Vec<LanguageModelRequestMessage> {
1947 log::trace!(
1948 "Building request messages from {} thread messages",
1949 self.messages.len()
1950 );
1951
1952 let system_prompt = SystemPromptTemplate {
1953 project: self.project_context.read(cx),
1954 available_tools: self.tools.keys().cloned().collect(),
1955 }
1956 .render(&self.templates)
1957 .context("failed to build system prompt")
1958 .expect("Invalid template");
1959 let mut messages = vec![LanguageModelRequestMessage {
1960 role: Role::System,
1961 content: vec![system_prompt.into()],
1962 cache: false,
1963 }];
1964 for message in &self.messages {
1965 messages.extend(message.to_request());
1966 }
1967
1968 if let Some(last_message) = messages.last_mut() {
1969 last_message.cache = true;
1970 }
1971
1972 if let Some(message) = self.pending_message.as_ref() {
1973 messages.extend(message.to_request());
1974 }
1975
1976 messages
1977 }
1978
1979 pub fn to_markdown(&self) -> String {
1980 let mut markdown = String::new();
1981 for (ix, message) in self.messages.iter().enumerate() {
1982 if ix > 0 {
1983 markdown.push('\n');
1984 }
1985 markdown.push_str(&message.to_markdown());
1986 }
1987
1988 if let Some(message) = self.pending_message.as_ref() {
1989 markdown.push('\n');
1990 markdown.push_str(&message.to_markdown());
1991 }
1992
1993 markdown
1994 }
1995
1996 fn advance_prompt_id(&mut self) {
1997 self.prompt_id = PromptId::new();
1998 }
1999
2000 fn retry_strategy_for(error: &LanguageModelCompletionError) -> Option<RetryStrategy> {
2001 use LanguageModelCompletionError::*;
2002 use http_client::StatusCode;
2003
2004 // General strategy here:
2005 // - If retrying won't help (e.g. invalid API key or payload too large), return None so we don't retry at all.
2006 // - If it's a time-based issue (e.g. server overloaded, rate limit exceeded), retry up to 4 times with exponential backoff.
2007 // - If it's an issue that *might* be fixed by retrying (e.g. internal server error), retry up to 3 times.
2008 match error {
2009 HttpResponseError {
2010 status_code: StatusCode::TOO_MANY_REQUESTS,
2011 ..
2012 } => Some(RetryStrategy::ExponentialBackoff {
2013 initial_delay: BASE_RETRY_DELAY,
2014 max_attempts: MAX_RETRY_ATTEMPTS,
2015 }),
2016 ServerOverloaded { retry_after, .. } | RateLimitExceeded { retry_after, .. } => {
2017 Some(RetryStrategy::Fixed {
2018 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
2019 max_attempts: MAX_RETRY_ATTEMPTS,
2020 })
2021 }
2022 UpstreamProviderError {
2023 status,
2024 retry_after,
2025 ..
2026 } => match *status {
2027 StatusCode::TOO_MANY_REQUESTS | StatusCode::SERVICE_UNAVAILABLE => {
2028 Some(RetryStrategy::Fixed {
2029 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
2030 max_attempts: MAX_RETRY_ATTEMPTS,
2031 })
2032 }
2033 StatusCode::INTERNAL_SERVER_ERROR => Some(RetryStrategy::Fixed {
2034 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
2035 // Internal Server Error could be anything, retry up to 3 times.
2036 max_attempts: 3,
2037 }),
2038 status => {
2039 // There is no StatusCode variant for the unofficial HTTP 529 ("The service is overloaded"),
2040 // but we frequently get them in practice. See https://http.dev/529
2041 if status.as_u16() == 529 {
2042 Some(RetryStrategy::Fixed {
2043 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
2044 max_attempts: MAX_RETRY_ATTEMPTS,
2045 })
2046 } else {
2047 Some(RetryStrategy::Fixed {
2048 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
2049 max_attempts: 2,
2050 })
2051 }
2052 }
2053 },
2054 ApiInternalServerError { .. } => Some(RetryStrategy::Fixed {
2055 delay: BASE_RETRY_DELAY,
2056 max_attempts: 3,
2057 }),
2058 ApiReadResponseError { .. }
2059 | HttpSend { .. }
2060 | DeserializeResponse { .. }
2061 | BadRequestFormat { .. } => Some(RetryStrategy::Fixed {
2062 delay: BASE_RETRY_DELAY,
2063 max_attempts: 3,
2064 }),
2065 // Retrying these errors definitely shouldn't help.
2066 HttpResponseError {
2067 status_code:
2068 StatusCode::PAYLOAD_TOO_LARGE | StatusCode::FORBIDDEN | StatusCode::UNAUTHORIZED,
2069 ..
2070 }
2071 | AuthenticationError { .. }
2072 | PermissionError { .. }
2073 | NoApiKey { .. }
2074 | ApiEndpointNotFound { .. }
2075 | PromptTooLarge { .. } => None,
2076 // These errors might be transient, so retry them
2077 SerializeRequest { .. } | BuildRequestBody { .. } => Some(RetryStrategy::Fixed {
2078 delay: BASE_RETRY_DELAY,
2079 max_attempts: 1,
2080 }),
2081 // Retry all other 4xx and 5xx errors once.
2082 HttpResponseError { status_code, .. }
2083 if status_code.is_client_error() || status_code.is_server_error() =>
2084 {
2085 Some(RetryStrategy::Fixed {
2086 delay: BASE_RETRY_DELAY,
2087 max_attempts: 3,
2088 })
2089 }
2090 Other(err)
2091 if err.is::<language_model::PaymentRequiredError>()
2092 || err.is::<language_model::ModelRequestLimitReachedError>() =>
2093 {
2094 // Retrying won't help for Payment Required or Model Request Limit errors (where
2095 // the user must upgrade to usage-based billing to get more requests, or else wait
2096 // for a significant amount of time for the request limit to reset).
2097 None
2098 }
2099 // Conservatively assume that any other errors are non-retryable
2100 HttpResponseError { .. } | Other(..) => Some(RetryStrategy::Fixed {
2101 delay: BASE_RETRY_DELAY,
2102 max_attempts: 2,
2103 }),
2104 }
2105 }
2106}
2107
2108struct RunningTurn {
2109 /// Holds the task that handles agent interaction until the end of the turn.
2110 /// Survives across multiple requests as the model performs tool calls and
2111 /// we run tools, report their results.
2112 _task: Task<()>,
2113 /// The current event stream for the running turn. Used to report a final
2114 /// cancellation event if we cancel the turn.
2115 event_stream: ThreadEventStream,
2116 /// The tools that were enabled for this turn.
2117 tools: BTreeMap<SharedString, Arc<dyn AnyAgentTool>>,
2118}
2119
2120impl RunningTurn {
2121 fn cancel(self) {
2122 log::debug!("Cancelling in progress turn");
2123 self.event_stream.send_canceled();
2124 }
2125}
2126
2127pub struct TokenUsageUpdated(pub Option<acp_thread::TokenUsage>);
2128
2129impl EventEmitter<TokenUsageUpdated> for Thread {}
2130
2131pub struct TitleUpdated;
2132
2133impl EventEmitter<TitleUpdated> for Thread {}
2134
2135pub trait AgentTool
2136where
2137 Self: 'static + Sized,
2138{
2139 type Input: for<'de> Deserialize<'de> + Serialize + JsonSchema;
2140 type Output: for<'de> Deserialize<'de> + Serialize + Into<LanguageModelToolResultContent>;
2141
2142 fn name() -> &'static str;
2143
2144 fn description(&self) -> SharedString {
2145 let schema = schemars::schema_for!(Self::Input);
2146 SharedString::new(
2147 schema
2148 .get("description")
2149 .and_then(|description| description.as_str())
2150 .unwrap_or_default(),
2151 )
2152 }
2153
2154 fn kind() -> acp::ToolKind;
2155
2156 /// The initial tool title to display. Can be updated during the tool run.
2157 fn initial_title(
2158 &self,
2159 input: Result<Self::Input, serde_json::Value>,
2160 cx: &mut App,
2161 ) -> SharedString;
2162
2163 /// Returns the JSON schema that describes the tool's input.
2164 fn input_schema(&self, format: LanguageModelToolSchemaFormat) -> Schema {
2165 crate::tool_schema::root_schema_for::<Self::Input>(format)
2166 }
2167
2168 /// Some tools rely on a provider for the underlying billing or other reasons.
2169 /// Allow the tool to check if they are compatible, or should be filtered out.
2170 fn supported_provider(&self, _provider: &LanguageModelProviderId) -> bool {
2171 true
2172 }
2173
2174 /// Runs the tool with the provided input.
2175 fn run(
2176 self: Arc<Self>,
2177 input: Self::Input,
2178 event_stream: ToolCallEventStream,
2179 cx: &mut App,
2180 ) -> Task<Result<Self::Output>>;
2181
2182 /// Emits events for a previous execution of the tool.
2183 fn replay(
2184 &self,
2185 _input: Self::Input,
2186 _output: Self::Output,
2187 _event_stream: ToolCallEventStream,
2188 _cx: &mut App,
2189 ) -> Result<()> {
2190 Ok(())
2191 }
2192
2193 fn erase(self) -> Arc<dyn AnyAgentTool> {
2194 Arc::new(Erased(Arc::new(self)))
2195 }
2196}
2197
2198pub struct Erased<T>(T);
2199
2200pub struct AgentToolOutput {
2201 pub llm_output: LanguageModelToolResultContent,
2202 pub raw_output: serde_json::Value,
2203}
2204
2205pub trait AnyAgentTool {
2206 fn name(&self) -> SharedString;
2207 fn description(&self) -> SharedString;
2208 fn kind(&self) -> acp::ToolKind;
2209 fn initial_title(&self, input: serde_json::Value, _cx: &mut App) -> SharedString;
2210 fn input_schema(&self, format: LanguageModelToolSchemaFormat) -> Result<serde_json::Value>;
2211 fn supported_provider(&self, _provider: &LanguageModelProviderId) -> bool {
2212 true
2213 }
2214 fn run(
2215 self: Arc<Self>,
2216 input: serde_json::Value,
2217 event_stream: ToolCallEventStream,
2218 cx: &mut App,
2219 ) -> Task<Result<AgentToolOutput>>;
2220 fn replay(
2221 &self,
2222 input: serde_json::Value,
2223 output: serde_json::Value,
2224 event_stream: ToolCallEventStream,
2225 cx: &mut App,
2226 ) -> Result<()>;
2227}
2228
2229impl<T> AnyAgentTool for Erased<Arc<T>>
2230where
2231 T: AgentTool,
2232{
2233 fn name(&self) -> SharedString {
2234 T::name().into()
2235 }
2236
2237 fn description(&self) -> SharedString {
2238 self.0.description()
2239 }
2240
2241 fn kind(&self) -> agent_client_protocol::ToolKind {
2242 T::kind()
2243 }
2244
2245 fn initial_title(&self, input: serde_json::Value, _cx: &mut App) -> SharedString {
2246 let parsed_input = serde_json::from_value(input.clone()).map_err(|_| input);
2247 self.0.initial_title(parsed_input, _cx)
2248 }
2249
2250 fn input_schema(&self, format: LanguageModelToolSchemaFormat) -> Result<serde_json::Value> {
2251 let mut json = serde_json::to_value(self.0.input_schema(format))?;
2252 adapt_schema_to_format(&mut json, format)?;
2253 Ok(json)
2254 }
2255
2256 fn supported_provider(&self, provider: &LanguageModelProviderId) -> bool {
2257 self.0.supported_provider(provider)
2258 }
2259
2260 fn run(
2261 self: Arc<Self>,
2262 input: serde_json::Value,
2263 event_stream: ToolCallEventStream,
2264 cx: &mut App,
2265 ) -> Task<Result<AgentToolOutput>> {
2266 cx.spawn(async move |cx| {
2267 let input = serde_json::from_value(input)?;
2268 let output = cx
2269 .update(|cx| self.0.clone().run(input, event_stream, cx))?
2270 .await?;
2271 let raw_output = serde_json::to_value(&output)?;
2272 Ok(AgentToolOutput {
2273 llm_output: output.into(),
2274 raw_output,
2275 })
2276 })
2277 }
2278
2279 fn replay(
2280 &self,
2281 input: serde_json::Value,
2282 output: serde_json::Value,
2283 event_stream: ToolCallEventStream,
2284 cx: &mut App,
2285 ) -> Result<()> {
2286 let input = serde_json::from_value(input)?;
2287 let output = serde_json::from_value(output)?;
2288 self.0.replay(input, output, event_stream, cx)
2289 }
2290}
2291
2292#[derive(Clone)]
2293struct ThreadEventStream(mpsc::UnboundedSender<Result<ThreadEvent>>);
2294
2295impl ThreadEventStream {
2296 fn send_user_message(&self, message: &UserMessage) {
2297 self.0
2298 .unbounded_send(Ok(ThreadEvent::UserMessage(message.clone())))
2299 .ok();
2300 }
2301
2302 fn send_text(&self, text: &str) {
2303 self.0
2304 .unbounded_send(Ok(ThreadEvent::AgentText(text.to_string())))
2305 .ok();
2306 }
2307
2308 fn send_thinking(&self, text: &str) {
2309 self.0
2310 .unbounded_send(Ok(ThreadEvent::AgentThinking(text.to_string())))
2311 .ok();
2312 }
2313
2314 fn send_tool_call(
2315 &self,
2316 id: &LanguageModelToolUseId,
2317 title: SharedString,
2318 kind: acp::ToolKind,
2319 input: serde_json::Value,
2320 ) {
2321 self.0
2322 .unbounded_send(Ok(ThreadEvent::ToolCall(Self::initial_tool_call(
2323 id,
2324 title.to_string(),
2325 kind,
2326 input,
2327 ))))
2328 .ok();
2329 }
2330
2331 fn initial_tool_call(
2332 id: &LanguageModelToolUseId,
2333 title: String,
2334 kind: acp::ToolKind,
2335 input: serde_json::Value,
2336 ) -> acp::ToolCall {
2337 acp::ToolCall {
2338 meta: None,
2339 id: acp::ToolCallId(id.to_string().into()),
2340 title,
2341 kind,
2342 status: acp::ToolCallStatus::Pending,
2343 content: vec![],
2344 locations: vec![],
2345 raw_input: Some(input),
2346 raw_output: None,
2347 }
2348 }
2349
2350 fn update_tool_call_fields(
2351 &self,
2352 tool_use_id: &LanguageModelToolUseId,
2353 fields: acp::ToolCallUpdateFields,
2354 ) {
2355 self.0
2356 .unbounded_send(Ok(ThreadEvent::ToolCallUpdate(
2357 acp::ToolCallUpdate {
2358 meta: None,
2359 id: acp::ToolCallId(tool_use_id.to_string().into()),
2360 fields,
2361 }
2362 .into(),
2363 )))
2364 .ok();
2365 }
2366
2367 fn send_retry(&self, status: acp_thread::RetryStatus) {
2368 self.0.unbounded_send(Ok(ThreadEvent::Retry(status))).ok();
2369 }
2370
2371 fn send_stop(&self, reason: acp::StopReason) {
2372 self.0.unbounded_send(Ok(ThreadEvent::Stop(reason))).ok();
2373 }
2374
2375 fn send_canceled(&self) {
2376 self.0
2377 .unbounded_send(Ok(ThreadEvent::Stop(acp::StopReason::Cancelled)))
2378 .ok();
2379 }
2380
2381 fn send_error(&self, error: impl Into<anyhow::Error>) {
2382 self.0.unbounded_send(Err(error.into())).ok();
2383 }
2384}
2385
2386#[derive(Clone)]
2387pub struct ToolCallEventStream {
2388 tool_use_id: LanguageModelToolUseId,
2389 stream: ThreadEventStream,
2390 fs: Option<Arc<dyn Fs>>,
2391}
2392
2393impl ToolCallEventStream {
2394 #[cfg(test)]
2395 pub fn test() -> (Self, ToolCallEventStreamReceiver) {
2396 let (events_tx, events_rx) = mpsc::unbounded::<Result<ThreadEvent>>();
2397
2398 let stream = ToolCallEventStream::new("test_id".into(), ThreadEventStream(events_tx), None);
2399
2400 (stream, ToolCallEventStreamReceiver(events_rx))
2401 }
2402
2403 fn new(
2404 tool_use_id: LanguageModelToolUseId,
2405 stream: ThreadEventStream,
2406 fs: Option<Arc<dyn Fs>>,
2407 ) -> Self {
2408 Self {
2409 tool_use_id,
2410 stream,
2411 fs,
2412 }
2413 }
2414
2415 pub fn update_fields(&self, fields: acp::ToolCallUpdateFields) {
2416 self.stream
2417 .update_tool_call_fields(&self.tool_use_id, fields);
2418 }
2419
2420 pub fn update_diff(&self, diff: Entity<acp_thread::Diff>) {
2421 self.stream
2422 .0
2423 .unbounded_send(Ok(ThreadEvent::ToolCallUpdate(
2424 acp_thread::ToolCallUpdateDiff {
2425 id: acp::ToolCallId(self.tool_use_id.to_string().into()),
2426 diff,
2427 }
2428 .into(),
2429 )))
2430 .ok();
2431 }
2432
2433 pub fn authorize(&self, title: impl Into<String>, cx: &mut App) -> Task<Result<()>> {
2434 if agent_settings::AgentSettings::get_global(cx).always_allow_tool_actions {
2435 return Task::ready(Ok(()));
2436 }
2437
2438 let (response_tx, response_rx) = oneshot::channel();
2439 self.stream
2440 .0
2441 .unbounded_send(Ok(ThreadEvent::ToolCallAuthorization(
2442 ToolCallAuthorization {
2443 tool_call: acp::ToolCallUpdate {
2444 meta: None,
2445 id: acp::ToolCallId(self.tool_use_id.to_string().into()),
2446 fields: acp::ToolCallUpdateFields {
2447 title: Some(title.into()),
2448 ..Default::default()
2449 },
2450 },
2451 options: vec![
2452 acp::PermissionOption {
2453 id: acp::PermissionOptionId("always_allow".into()),
2454 name: "Always Allow".into(),
2455 kind: acp::PermissionOptionKind::AllowAlways,
2456 meta: None,
2457 },
2458 acp::PermissionOption {
2459 id: acp::PermissionOptionId("allow".into()),
2460 name: "Allow".into(),
2461 kind: acp::PermissionOptionKind::AllowOnce,
2462 meta: None,
2463 },
2464 acp::PermissionOption {
2465 id: acp::PermissionOptionId("deny".into()),
2466 name: "Deny".into(),
2467 kind: acp::PermissionOptionKind::RejectOnce,
2468 meta: None,
2469 },
2470 ],
2471 response: response_tx,
2472 },
2473 )))
2474 .ok();
2475 let fs = self.fs.clone();
2476 cx.spawn(async move |cx| match response_rx.await?.0.as_ref() {
2477 "always_allow" => {
2478 if let Some(fs) = fs.clone() {
2479 cx.update(|cx| {
2480 update_settings_file(fs, cx, |settings, _| {
2481 settings
2482 .agent
2483 .get_or_insert_default()
2484 .set_always_allow_tool_actions(true);
2485 });
2486 })?;
2487 }
2488
2489 Ok(())
2490 }
2491 "allow" => Ok(()),
2492 _ => Err(anyhow!("Permission to run tool denied by user")),
2493 })
2494 }
2495}
2496
2497#[cfg(test)]
2498pub struct ToolCallEventStreamReceiver(mpsc::UnboundedReceiver<Result<ThreadEvent>>);
2499
2500#[cfg(test)]
2501impl ToolCallEventStreamReceiver {
2502 pub async fn expect_authorization(&mut self) -> ToolCallAuthorization {
2503 let event = self.0.next().await;
2504 if let Some(Ok(ThreadEvent::ToolCallAuthorization(auth))) = event {
2505 auth
2506 } else {
2507 panic!("Expected ToolCallAuthorization but got: {:?}", event);
2508 }
2509 }
2510
2511 pub async fn expect_update_fields(&mut self) -> acp::ToolCallUpdateFields {
2512 let event = self.0.next().await;
2513 if let Some(Ok(ThreadEvent::ToolCallUpdate(acp_thread::ToolCallUpdate::UpdateFields(
2514 update,
2515 )))) = event
2516 {
2517 update.fields
2518 } else {
2519 panic!("Expected update fields but got: {:?}", event);
2520 }
2521 }
2522
2523 pub async fn expect_diff(&mut self) -> Entity<acp_thread::Diff> {
2524 let event = self.0.next().await;
2525 if let Some(Ok(ThreadEvent::ToolCallUpdate(acp_thread::ToolCallUpdate::UpdateDiff(
2526 update,
2527 )))) = event
2528 {
2529 update.diff
2530 } else {
2531 panic!("Expected diff but got: {:?}", event);
2532 }
2533 }
2534
2535 pub async fn expect_terminal(&mut self) -> Entity<acp_thread::Terminal> {
2536 let event = self.0.next().await;
2537 if let Some(Ok(ThreadEvent::ToolCallUpdate(acp_thread::ToolCallUpdate::UpdateTerminal(
2538 update,
2539 )))) = event
2540 {
2541 update.terminal
2542 } else {
2543 panic!("Expected terminal but got: {:?}", event);
2544 }
2545 }
2546}
2547
2548#[cfg(test)]
2549impl std::ops::Deref for ToolCallEventStreamReceiver {
2550 type Target = mpsc::UnboundedReceiver<Result<ThreadEvent>>;
2551
2552 fn deref(&self) -> &Self::Target {
2553 &self.0
2554 }
2555}
2556
2557#[cfg(test)]
2558impl std::ops::DerefMut for ToolCallEventStreamReceiver {
2559 fn deref_mut(&mut self) -> &mut Self::Target {
2560 &mut self.0
2561 }
2562}
2563
2564impl From<&str> for UserMessageContent {
2565 fn from(text: &str) -> Self {
2566 Self::Text(text.into())
2567 }
2568}
2569
2570impl From<acp::ContentBlock> for UserMessageContent {
2571 fn from(value: acp::ContentBlock) -> Self {
2572 match value {
2573 acp::ContentBlock::Text(text_content) => Self::Text(text_content.text),
2574 acp::ContentBlock::Image(image_content) => Self::Image(convert_image(image_content)),
2575 acp::ContentBlock::Audio(_) => {
2576 // TODO
2577 Self::Text("[audio]".to_string())
2578 }
2579 acp::ContentBlock::ResourceLink(resource_link) => {
2580 match MentionUri::parse(&resource_link.uri) {
2581 Ok(uri) => Self::Mention {
2582 uri,
2583 content: String::new(),
2584 },
2585 Err(err) => {
2586 log::error!("Failed to parse mention link: {}", err);
2587 Self::Text(format!("[{}]({})", resource_link.name, resource_link.uri))
2588 }
2589 }
2590 }
2591 acp::ContentBlock::Resource(resource) => match resource.resource {
2592 acp::EmbeddedResourceResource::TextResourceContents(resource) => {
2593 match MentionUri::parse(&resource.uri) {
2594 Ok(uri) => Self::Mention {
2595 uri,
2596 content: resource.text,
2597 },
2598 Err(err) => {
2599 log::error!("Failed to parse mention link: {}", err);
2600 Self::Text(
2601 MarkdownCodeBlock {
2602 tag: &resource.uri,
2603 text: &resource.text,
2604 }
2605 .to_string(),
2606 )
2607 }
2608 }
2609 }
2610 acp::EmbeddedResourceResource::BlobResourceContents(_) => {
2611 // TODO
2612 Self::Text("[blob]".to_string())
2613 }
2614 },
2615 }
2616 }
2617}
2618
2619impl From<UserMessageContent> for acp::ContentBlock {
2620 fn from(content: UserMessageContent) -> Self {
2621 match content {
2622 UserMessageContent::Text(text) => acp::ContentBlock::Text(acp::TextContent {
2623 text,
2624 annotations: None,
2625 meta: None,
2626 }),
2627 UserMessageContent::Image(image) => acp::ContentBlock::Image(acp::ImageContent {
2628 data: image.source.to_string(),
2629 mime_type: "image/png".to_string(),
2630 meta: None,
2631 annotations: None,
2632 uri: None,
2633 }),
2634 UserMessageContent::Mention { uri, content } => {
2635 acp::ContentBlock::Resource(acp::EmbeddedResource {
2636 meta: None,
2637 resource: acp::EmbeddedResourceResource::TextResourceContents(
2638 acp::TextResourceContents {
2639 meta: None,
2640 mime_type: None,
2641 text: content,
2642 uri: uri.to_uri().to_string(),
2643 },
2644 ),
2645 annotations: None,
2646 })
2647 }
2648 }
2649 }
2650}
2651
2652fn convert_image(image_content: acp::ImageContent) -> LanguageModelImage {
2653 LanguageModelImage {
2654 source: image_content.data.into(),
2655 // TODO: make this optional?
2656 size: gpui::Size::new(0.into(), 0.into()),
2657 }
2658}