1use crate::{
2 ContextServerRegistry, CopyPathTool, CreateDirectoryTool, DbLanguageModel, DbThread,
3 DeletePathTool, DiagnosticsTool, EditFileTool, FetchTool, FindPathTool, GrepTool, HeadTool,
4 ListDirectoryTool, MovePathTool, NowTool, OpenTool, ProjectSnapshot, ReadFileTool,
5 SystemPromptTemplate, Template, Templates, TerminalTool, ThinkingTool, WebSearchTool,
6};
7use acp_thread::{MentionUri, UserMessageId};
8use action_log::ActionLog;
9
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 chrono::{DateTime, Utc};
17use client::{ModelRequestUsage, RequestUsage, UserStore};
18use cloud_llm_client::{CompletionIntent, CompletionRequestStatus, Plan, UsageLimit};
19use collections::{HashMap, HashSet, IndexMap};
20use fs::Fs;
21use futures::stream;
22use futures::{
23 FutureExt,
24 channel::{mpsc, oneshot},
25 future::Shared,
26 stream::FuturesUnordered,
27};
28use gpui::{
29 App, AppContext, AsyncApp, Context, Entity, EventEmitter, SharedString, Task, WeakEntity,
30};
31use language_model::{
32 LanguageModel, LanguageModelCompletionError, LanguageModelCompletionEvent, LanguageModelExt,
33 LanguageModelImage, LanguageModelProviderId, LanguageModelRegistry, LanguageModelRequest,
34 LanguageModelRequestMessage, LanguageModelRequestTool, LanguageModelToolResult,
35 LanguageModelToolResultContent, LanguageModelToolSchemaFormat, LanguageModelToolUse,
36 LanguageModelToolUseId, Role, SelectedModel, StopReason, TokenUsage, ZED_CLOUD_PROVIDER_ID,
37};
38use project::Project;
39use prompt_store::ProjectContext;
40use schemars::{JsonSchema, Schema};
41use serde::{Deserialize, Serialize};
42use settings::{Settings, update_settings_file};
43use smol::stream::StreamExt;
44use std::{
45 collections::BTreeMap,
46 ops::RangeInclusive,
47 path::Path,
48 rc::Rc,
49 sync::Arc,
50 time::{Duration, Instant},
51};
52use std::{fmt::Write, path::PathBuf};
53use util::{ResultExt, debug_panic, markdown::MarkdownCodeBlock, paths::PathStyle};
54use uuid::Uuid;
55
56const TOOL_CANCELED_MESSAGE: &str = "Tool canceled by user";
57pub const MAX_TOOL_NAME_LENGTH: usize = 64;
58
59/// The ID of the user prompt that initiated a request.
60///
61/// This equates to the user physically submitting a message to the model (e.g., by pressing the Enter key).
62#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Clone, Serialize, Deserialize)]
63pub struct PromptId(Arc<str>);
64
65impl PromptId {
66 pub fn new() -> Self {
67 Self(Uuid::new_v4().to_string().into())
68 }
69}
70
71impl std::fmt::Display for PromptId {
72 fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
73 write!(f, "{}", self.0)
74 }
75}
76
77pub(crate) const MAX_RETRY_ATTEMPTS: u8 = 4;
78pub(crate) const BASE_RETRY_DELAY: Duration = Duration::from_secs(5);
79
80#[derive(Debug, Clone)]
81enum RetryStrategy {
82 ExponentialBackoff {
83 initial_delay: Duration,
84 max_attempts: u8,
85 },
86 Fixed {
87 delay: Duration,
88 max_attempts: u8,
89 },
90}
91
92#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
93pub enum Message {
94 User(UserMessage),
95 Agent(AgentMessage),
96 Resume,
97}
98
99impl Message {
100 pub fn as_agent_message(&self) -> Option<&AgentMessage> {
101 match self {
102 Message::Agent(agent_message) => Some(agent_message),
103 _ => None,
104 }
105 }
106
107 pub fn to_request(&self) -> Vec<LanguageModelRequestMessage> {
108 match self {
109 Message::User(message) => vec![message.to_request()],
110 Message::Agent(message) => message.to_request(),
111 Message::Resume => vec![LanguageModelRequestMessage {
112 role: Role::User,
113 content: vec!["Continue where you left off".into()],
114 cache: false,
115 }],
116 }
117 }
118
119 pub fn to_markdown(&self) -> String {
120 match self {
121 Message::User(message) => message.to_markdown(),
122 Message::Agent(message) => message.to_markdown(),
123 Message::Resume => "[resume]\n".into(),
124 }
125 }
126
127 pub fn role(&self) -> Role {
128 match self {
129 Message::User(_) | Message::Resume => Role::User,
130 Message::Agent(_) => Role::Assistant,
131 }
132 }
133}
134
135#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
136pub struct UserMessage {
137 pub id: UserMessageId,
138 pub content: Vec<UserMessageContent>,
139}
140
141#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
142pub enum UserMessageContent {
143 Text(String),
144 Mention { uri: MentionUri, content: String },
145 Image(LanguageModelImage),
146}
147
148impl UserMessage {
149 pub fn to_markdown(&self) -> String {
150 let mut markdown = String::from("## User\n\n");
151
152 for content in &self.content {
153 match content {
154 UserMessageContent::Text(text) => {
155 markdown.push_str(text);
156 markdown.push('\n');
157 }
158 UserMessageContent::Image(_) => {
159 markdown.push_str("<image />\n");
160 }
161 UserMessageContent::Mention { uri, content } => {
162 if !content.is_empty() {
163 let _ = writeln!(&mut markdown, "{}\n\n{}", uri.as_link(), content);
164 } else {
165 let _ = writeln!(&mut markdown, "{}", uri.as_link());
166 }
167 }
168 }
169 }
170
171 markdown
172 }
173
174 fn to_request(&self) -> LanguageModelRequestMessage {
175 let mut message = LanguageModelRequestMessage {
176 role: Role::User,
177 content: Vec::with_capacity(self.content.len()),
178 cache: false,
179 };
180
181 const OPEN_CONTEXT: &str = "<context>\n\
182 The following items were attached by the user. \
183 They are up-to-date and don't need to be re-read.\n\n";
184
185 const OPEN_FILES_TAG: &str = "<files>";
186 const OPEN_DIRECTORIES_TAG: &str = "<directories>";
187 const OPEN_SYMBOLS_TAG: &str = "<symbols>";
188 const OPEN_SELECTIONS_TAG: &str = "<selections>";
189 const OPEN_THREADS_TAG: &str = "<threads>";
190 const OPEN_FETCH_TAG: &str = "<fetched_urls>";
191 const OPEN_RULES_TAG: &str =
192 "<rules>\nThe user has specified the following rules that should be applied:\n";
193
194 let mut file_context = OPEN_FILES_TAG.to_string();
195 let mut directory_context = OPEN_DIRECTORIES_TAG.to_string();
196 let mut symbol_context = OPEN_SYMBOLS_TAG.to_string();
197 let mut selection_context = OPEN_SELECTIONS_TAG.to_string();
198 let mut thread_context = OPEN_THREADS_TAG.to_string();
199 let mut fetch_context = OPEN_FETCH_TAG.to_string();
200 let mut rules_context = OPEN_RULES_TAG.to_string();
201
202 for chunk in &self.content {
203 let chunk = match chunk {
204 UserMessageContent::Text(text) => {
205 language_model::MessageContent::Text(text.clone())
206 }
207 UserMessageContent::Image(value) => {
208 language_model::MessageContent::Image(value.clone())
209 }
210 UserMessageContent::Mention { uri, content } => {
211 match uri {
212 MentionUri::File { abs_path } => {
213 write!(
214 &mut file_context,
215 "\n{}",
216 MarkdownCodeBlock {
217 tag: &codeblock_tag(abs_path, None),
218 text: &content.to_string(),
219 }
220 )
221 .ok();
222 }
223 MentionUri::PastedImage => {
224 debug_panic!("pasted image URI should not be used in mention content")
225 }
226 MentionUri::Directory { .. } => {
227 write!(&mut directory_context, "\n{}\n", content).ok();
228 }
229 MentionUri::Symbol {
230 abs_path: path,
231 line_range,
232 ..
233 } => {
234 write!(
235 &mut symbol_context,
236 "\n{}",
237 MarkdownCodeBlock {
238 tag: &codeblock_tag(path, Some(line_range)),
239 text: content
240 }
241 )
242 .ok();
243 }
244 MentionUri::Selection {
245 abs_path: path,
246 line_range,
247 ..
248 } => {
249 write!(
250 &mut selection_context,
251 "\n{}",
252 MarkdownCodeBlock {
253 tag: &codeblock_tag(
254 path.as_deref().unwrap_or("Untitled".as_ref()),
255 Some(line_range)
256 ),
257 text: content
258 }
259 )
260 .ok();
261 }
262 MentionUri::Thread { .. } => {
263 write!(&mut thread_context, "\n{}\n", content).ok();
264 }
265 MentionUri::TextThread { .. } => {
266 write!(&mut thread_context, "\n{}\n", content).ok();
267 }
268 MentionUri::Rule { .. } => {
269 write!(
270 &mut rules_context,
271 "\n{}",
272 MarkdownCodeBlock {
273 tag: "",
274 text: content
275 }
276 )
277 .ok();
278 }
279 MentionUri::Fetch { url } => {
280 write!(&mut fetch_context, "\nFetch: {}\n\n{}", url, content).ok();
281 }
282 }
283
284 language_model::MessageContent::Text(uri.as_link().to_string())
285 }
286 };
287
288 message.content.push(chunk);
289 }
290
291 let len_before_context = message.content.len();
292
293 if file_context.len() > OPEN_FILES_TAG.len() {
294 file_context.push_str("</files>\n");
295 message
296 .content
297 .push(language_model::MessageContent::Text(file_context));
298 }
299
300 if directory_context.len() > OPEN_DIRECTORIES_TAG.len() {
301 directory_context.push_str("</directories>\n");
302 message
303 .content
304 .push(language_model::MessageContent::Text(directory_context));
305 }
306
307 if symbol_context.len() > OPEN_SYMBOLS_TAG.len() {
308 symbol_context.push_str("</symbols>\n");
309 message
310 .content
311 .push(language_model::MessageContent::Text(symbol_context));
312 }
313
314 if selection_context.len() > OPEN_SELECTIONS_TAG.len() {
315 selection_context.push_str("</selections>\n");
316 message
317 .content
318 .push(language_model::MessageContent::Text(selection_context));
319 }
320
321 if thread_context.len() > OPEN_THREADS_TAG.len() {
322 thread_context.push_str("</threads>\n");
323 message
324 .content
325 .push(language_model::MessageContent::Text(thread_context));
326 }
327
328 if fetch_context.len() > OPEN_FETCH_TAG.len() {
329 fetch_context.push_str("</fetched_urls>\n");
330 message
331 .content
332 .push(language_model::MessageContent::Text(fetch_context));
333 }
334
335 if rules_context.len() > OPEN_RULES_TAG.len() {
336 rules_context.push_str("</user_rules>\n");
337 message
338 .content
339 .push(language_model::MessageContent::Text(rules_context));
340 }
341
342 if message.content.len() > len_before_context {
343 message.content.insert(
344 len_before_context,
345 language_model::MessageContent::Text(OPEN_CONTEXT.into()),
346 );
347 message
348 .content
349 .push(language_model::MessageContent::Text("</context>".into()));
350 }
351
352 message
353 }
354}
355
356fn codeblock_tag(full_path: &Path, line_range: Option<&RangeInclusive<u32>>) -> String {
357 let mut result = String::new();
358
359 if let Some(extension) = full_path.extension().and_then(|ext| ext.to_str()) {
360 let _ = write!(result, "{} ", extension);
361 }
362
363 let _ = write!(result, "{}", full_path.display());
364
365 if let Some(range) = line_range {
366 if range.start() == range.end() {
367 let _ = write!(result, ":{}", range.start() + 1);
368 } else {
369 let _ = write!(result, ":{}-{}", range.start() + 1, range.end() + 1);
370 }
371 }
372
373 result
374}
375
376impl AgentMessage {
377 pub fn to_markdown(&self) -> String {
378 let mut markdown = String::from("## Assistant\n\n");
379
380 for content in &self.content {
381 match content {
382 AgentMessageContent::Text(text) => {
383 markdown.push_str(text);
384 markdown.push('\n');
385 }
386 AgentMessageContent::Thinking { text, .. } => {
387 markdown.push_str("<think>");
388 markdown.push_str(text);
389 markdown.push_str("</think>\n");
390 }
391 AgentMessageContent::RedactedThinking(_) => {
392 markdown.push_str("<redacted_thinking />\n")
393 }
394 AgentMessageContent::ToolUse(tool_use) => {
395 markdown.push_str(&format!(
396 "**Tool Use**: {} (ID: {})\n",
397 tool_use.name, tool_use.id
398 ));
399 markdown.push_str(&format!(
400 "{}\n",
401 MarkdownCodeBlock {
402 tag: "json",
403 text: &format!("{:#}", tool_use.input)
404 }
405 ));
406 }
407 }
408 }
409
410 for tool_result in self.tool_results.values() {
411 markdown.push_str(&format!(
412 "**Tool Result**: {} (ID: {})\n\n",
413 tool_result.tool_name, tool_result.tool_use_id
414 ));
415 if tool_result.is_error {
416 markdown.push_str("**ERROR:**\n");
417 }
418
419 match &tool_result.content {
420 LanguageModelToolResultContent::Text(text) => {
421 writeln!(markdown, "{text}\n").ok();
422 }
423 LanguageModelToolResultContent::Image(_) => {
424 writeln!(markdown, "<image />\n").ok();
425 }
426 }
427
428 if let Some(output) = tool_result.output.as_ref() {
429 writeln!(
430 markdown,
431 "**Debug Output**:\n\n```json\n{}\n```\n",
432 serde_json::to_string_pretty(output).unwrap()
433 )
434 .unwrap();
435 }
436 }
437
438 markdown
439 }
440
441 pub fn to_request(&self) -> Vec<LanguageModelRequestMessage> {
442 let mut assistant_message = LanguageModelRequestMessage {
443 role: Role::Assistant,
444 content: Vec::with_capacity(self.content.len()),
445 cache: false,
446 };
447 for chunk in &self.content {
448 match chunk {
449 AgentMessageContent::Text(text) => {
450 assistant_message
451 .content
452 .push(language_model::MessageContent::Text(text.clone()));
453 }
454 AgentMessageContent::Thinking { text, signature } => {
455 assistant_message
456 .content
457 .push(language_model::MessageContent::Thinking {
458 text: text.clone(),
459 signature: signature.clone(),
460 });
461 }
462 AgentMessageContent::RedactedThinking(value) => {
463 assistant_message.content.push(
464 language_model::MessageContent::RedactedThinking(value.clone()),
465 );
466 }
467 AgentMessageContent::ToolUse(tool_use) => {
468 if self.tool_results.contains_key(&tool_use.id) {
469 assistant_message
470 .content
471 .push(language_model::MessageContent::ToolUse(tool_use.clone()));
472 }
473 }
474 };
475 }
476
477 let mut user_message = LanguageModelRequestMessage {
478 role: Role::User,
479 content: Vec::new(),
480 cache: false,
481 };
482
483 for tool_result in self.tool_results.values() {
484 let mut tool_result = tool_result.clone();
485 // Surprisingly, the API fails if we return an empty string here.
486 // It thinks we are sending a tool use without a tool result.
487 if tool_result.content.is_empty() {
488 tool_result.content = "<Tool returned an empty string>".into();
489 }
490 user_message
491 .content
492 .push(language_model::MessageContent::ToolResult(tool_result));
493 }
494
495 let mut messages = Vec::new();
496 if !assistant_message.content.is_empty() {
497 messages.push(assistant_message);
498 }
499 if !user_message.content.is_empty() {
500 messages.push(user_message);
501 }
502 messages
503 }
504}
505
506#[derive(Default, Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
507pub struct AgentMessage {
508 pub content: Vec<AgentMessageContent>,
509 pub tool_results: IndexMap<LanguageModelToolUseId, LanguageModelToolResult>,
510}
511
512#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
513pub enum AgentMessageContent {
514 Text(String),
515 Thinking {
516 text: String,
517 signature: Option<String>,
518 },
519 RedactedThinking(String),
520 ToolUse(LanguageModelToolUse),
521}
522
523pub trait TerminalHandle {
524 fn id(&self, cx: &AsyncApp) -> Result<acp::TerminalId>;
525 fn current_output(&self, cx: &AsyncApp) -> Result<acp::TerminalOutputResponse>;
526 fn wait_for_exit(&self, cx: &AsyncApp) -> Result<Shared<Task<acp::TerminalExitStatus>>>;
527}
528
529pub trait ThreadEnvironment {
530 fn create_terminal(
531 &self,
532 command: String,
533 cwd: Option<PathBuf>,
534 output_byte_limit: Option<u64>,
535 cx: &mut AsyncApp,
536 ) -> Task<Result<Rc<dyn TerminalHandle>>>;
537}
538
539#[derive(Debug)]
540pub enum ThreadEvent {
541 UserMessage(UserMessage),
542 AgentText(String),
543 AgentThinking(String),
544 ToolCall(acp::ToolCall),
545 ToolCallUpdate(acp_thread::ToolCallUpdate),
546 ToolCallAuthorization(ToolCallAuthorization),
547 Retry(acp_thread::RetryStatus),
548 Stop(acp::StopReason),
549}
550
551#[derive(Debug)]
552pub struct NewTerminal {
553 pub command: String,
554 pub output_byte_limit: Option<u64>,
555 pub cwd: Option<PathBuf>,
556 pub response: oneshot::Sender<Result<Entity<acp_thread::Terminal>>>,
557}
558
559#[derive(Debug)]
560pub struct ToolCallAuthorization {
561 pub tool_call: acp::ToolCallUpdate,
562 pub options: Vec<acp::PermissionOption>,
563 pub response: oneshot::Sender<acp::PermissionOptionId>,
564}
565
566#[derive(Debug, thiserror::Error)]
567enum CompletionError {
568 #[error("max tokens")]
569 MaxTokens,
570 #[error("refusal")]
571 Refusal,
572 #[error(transparent)]
573 Other(#[from] anyhow::Error),
574}
575
576pub struct Thread {
577 id: acp::SessionId,
578 prompt_id: PromptId,
579 updated_at: DateTime<Utc>,
580 title: Option<SharedString>,
581 pending_title_generation: Option<Task<()>>,
582 pending_summary_generation: Option<Shared<Task<Option<SharedString>>>>,
583 summary: Option<SharedString>,
584 messages: Vec<Message>,
585 user_store: Entity<UserStore>,
586 completion_mode: CompletionMode,
587 /// Holds the task that handles agent interaction until the end of the turn.
588 /// Survives across multiple requests as the model performs tool calls and
589 /// we run tools, report their results.
590 running_turn: Option<RunningTurn>,
591 pending_message: Option<AgentMessage>,
592 tools: BTreeMap<SharedString, Arc<dyn AnyAgentTool>>,
593 tool_use_limit_reached: bool,
594 request_token_usage: HashMap<UserMessageId, language_model::TokenUsage>,
595 #[allow(unused)]
596 cumulative_token_usage: TokenUsage,
597 #[allow(unused)]
598 initial_project_snapshot: Shared<Task<Option<Arc<ProjectSnapshot>>>>,
599 context_server_registry: Entity<ContextServerRegistry>,
600 profile_id: AgentProfileId,
601 project_context: Entity<ProjectContext>,
602 templates: Arc<Templates>,
603 model: Option<Arc<dyn LanguageModel>>,
604 summarization_model: Option<Arc<dyn LanguageModel>>,
605 prompt_capabilities_tx: watch::Sender<acp::PromptCapabilities>,
606 pub(crate) prompt_capabilities_rx: watch::Receiver<acp::PromptCapabilities>,
607 pub(crate) project: Entity<Project>,
608 pub(crate) action_log: Entity<ActionLog>,
609}
610
611impl Thread {
612 fn prompt_capabilities(model: Option<&dyn LanguageModel>) -> acp::PromptCapabilities {
613 let image = model.map_or(true, |model| model.supports_images());
614 acp::PromptCapabilities {
615 meta: None,
616 image,
617 audio: false,
618 embedded_context: true,
619 }
620 }
621
622 pub fn new(
623 project: Entity<Project>,
624 project_context: Entity<ProjectContext>,
625 context_server_registry: Entity<ContextServerRegistry>,
626 templates: Arc<Templates>,
627 model: Option<Arc<dyn LanguageModel>>,
628 cx: &mut Context<Self>,
629 ) -> Self {
630 let profile_id = AgentSettings::get_global(cx).default_profile.clone();
631 let action_log = cx.new(|_cx| ActionLog::new(project.clone()));
632 let (prompt_capabilities_tx, prompt_capabilities_rx) =
633 watch::channel(Self::prompt_capabilities(model.as_deref()));
634 Self {
635 id: acp::SessionId(uuid::Uuid::new_v4().to_string().into()),
636 prompt_id: PromptId::new(),
637 updated_at: Utc::now(),
638 title: None,
639 pending_title_generation: None,
640 pending_summary_generation: None,
641 summary: None,
642 messages: Vec::new(),
643 user_store: project.read(cx).user_store(),
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(
749 &tool_use.id,
750 &tool_use.name,
751 title,
752 kind,
753 tool_use.input.clone(),
754 );
755
756 let output = tool_result
757 .as_ref()
758 .and_then(|result| result.output.clone());
759 if let Some(output) = output.clone() {
760 let tool_event_stream = ToolCallEventStream::new(
761 tool_use.id.clone(),
762 stream.clone(),
763 Some(self.project.read(cx).fs().clone()),
764 );
765 tool.replay(tool_use.input.clone(), output, tool_event_stream, cx)
766 .log_err();
767 }
768
769 stream.update_tool_call_fields(
770 &tool_use.id,
771 acp::ToolCallUpdateFields {
772 status: Some(
773 tool_result
774 .as_ref()
775 .map_or(acp::ToolCallStatus::Failed, |result| {
776 if result.is_error {
777 acp::ToolCallStatus::Failed
778 } else {
779 acp::ToolCallStatus::Completed
780 }
781 }),
782 ),
783 raw_output: output,
784 ..Default::default()
785 },
786 );
787 }
788
789 pub fn from_db(
790 id: acp::SessionId,
791 db_thread: DbThread,
792 project: Entity<Project>,
793 project_context: Entity<ProjectContext>,
794 context_server_registry: Entity<ContextServerRegistry>,
795 templates: Arc<Templates>,
796 cx: &mut Context<Self>,
797 ) -> Self {
798 let profile_id = db_thread
799 .profile
800 .unwrap_or_else(|| AgentSettings::get_global(cx).default_profile.clone());
801 let model = LanguageModelRegistry::global(cx).update(cx, |registry, cx| {
802 db_thread
803 .model
804 .and_then(|model| {
805 let model = SelectedModel {
806 provider: model.provider.clone().into(),
807 model: model.model.into(),
808 };
809 registry.select_model(&model, cx)
810 })
811 .or_else(|| registry.default_model())
812 .map(|model| model.model)
813 });
814 let (prompt_capabilities_tx, prompt_capabilities_rx) =
815 watch::channel(Self::prompt_capabilities(model.as_deref()));
816
817 let action_log = cx.new(|_| ActionLog::new(project.clone()));
818
819 Self {
820 id,
821 prompt_id: PromptId::new(),
822 title: if db_thread.title.is_empty() {
823 None
824 } else {
825 Some(db_thread.title.clone())
826 },
827 pending_title_generation: None,
828 pending_summary_generation: None,
829 summary: db_thread.detailed_summary,
830 messages: db_thread.messages,
831 user_store: project.read(cx).user_store(),
832 completion_mode: db_thread.completion_mode.unwrap_or_default(),
833 running_turn: None,
834 pending_message: None,
835 tools: BTreeMap::default(),
836 tool_use_limit_reached: false,
837 request_token_usage: db_thread.request_token_usage.clone(),
838 cumulative_token_usage: db_thread.cumulative_token_usage,
839 initial_project_snapshot: Task::ready(db_thread.initial_project_snapshot).shared(),
840 context_server_registry,
841 profile_id,
842 project_context,
843 templates,
844 model,
845 summarization_model: None,
846 project,
847 action_log,
848 updated_at: db_thread.updated_at,
849 prompt_capabilities_tx,
850 prompt_capabilities_rx,
851 }
852 }
853
854 pub fn to_db(&self, cx: &App) -> Task<DbThread> {
855 let initial_project_snapshot = self.initial_project_snapshot.clone();
856 let mut thread = DbThread {
857 title: self.title(),
858 messages: self.messages.clone(),
859 updated_at: self.updated_at,
860 detailed_summary: self.summary.clone(),
861 initial_project_snapshot: None,
862 cumulative_token_usage: self.cumulative_token_usage,
863 request_token_usage: self.request_token_usage.clone(),
864 model: self.model.as_ref().map(|model| DbLanguageModel {
865 provider: model.provider_id().to_string(),
866 model: model.name().0.to_string(),
867 }),
868 completion_mode: Some(self.completion_mode),
869 profile: Some(self.profile_id.clone()),
870 };
871
872 cx.background_spawn(async move {
873 let initial_project_snapshot = initial_project_snapshot.await;
874 thread.initial_project_snapshot = initial_project_snapshot;
875 thread
876 })
877 }
878
879 /// Create a snapshot of the current project state including git information and unsaved buffers.
880 fn project_snapshot(
881 project: Entity<Project>,
882 cx: &mut Context<Self>,
883 ) -> Task<Arc<ProjectSnapshot>> {
884 let task = project::telemetry_snapshot::TelemetrySnapshot::new(&project, cx);
885 cx.spawn(async move |_, _| {
886 let snapshot = task.await;
887
888 Arc::new(ProjectSnapshot {
889 worktree_snapshots: snapshot.worktree_snapshots,
890 timestamp: Utc::now(),
891 })
892 })
893 }
894
895 pub fn project_context(&self) -> &Entity<ProjectContext> {
896 &self.project_context
897 }
898
899 pub fn project(&self) -> &Entity<Project> {
900 &self.project
901 }
902
903 pub fn action_log(&self) -> &Entity<ActionLog> {
904 &self.action_log
905 }
906
907 pub fn is_empty(&self) -> bool {
908 self.messages.is_empty() && self.title.is_none()
909 }
910
911 pub fn model(&self) -> Option<&Arc<dyn LanguageModel>> {
912 self.model.as_ref()
913 }
914
915 pub fn set_model(&mut self, model: Arc<dyn LanguageModel>, cx: &mut Context<Self>) {
916 let old_usage = self.latest_token_usage();
917 self.model = Some(model);
918 let new_caps = Self::prompt_capabilities(self.model.as_deref());
919 let new_usage = self.latest_token_usage();
920 if old_usage != new_usage {
921 cx.emit(TokenUsageUpdated(new_usage));
922 }
923 self.prompt_capabilities_tx.send(new_caps).log_err();
924 cx.notify()
925 }
926
927 pub fn summarization_model(&self) -> Option<&Arc<dyn LanguageModel>> {
928 self.summarization_model.as_ref()
929 }
930
931 pub fn set_summarization_model(
932 &mut self,
933 model: Option<Arc<dyn LanguageModel>>,
934 cx: &mut Context<Self>,
935 ) {
936 self.summarization_model = model;
937 cx.notify()
938 }
939
940 pub fn completion_mode(&self) -> CompletionMode {
941 self.completion_mode
942 }
943
944 pub fn set_completion_mode(&mut self, mode: CompletionMode, cx: &mut Context<Self>) {
945 let old_usage = self.latest_token_usage();
946 self.completion_mode = mode;
947 let new_usage = self.latest_token_usage();
948 if old_usage != new_usage {
949 cx.emit(TokenUsageUpdated(new_usage));
950 }
951 cx.notify()
952 }
953
954 #[cfg(any(test, feature = "test-support"))]
955 pub fn last_message(&self) -> Option<Message> {
956 if let Some(message) = self.pending_message.clone() {
957 Some(Message::Agent(message))
958 } else {
959 self.messages.last().cloned()
960 }
961 }
962
963 pub fn add_default_tools(
964 &mut self,
965 environment: Rc<dyn ThreadEnvironment>,
966 cx: &mut Context<Self>,
967 ) {
968 let language_registry = self.project.read(cx).languages().clone();
969 self.add_tool(CopyPathTool::new(self.project.clone()));
970 self.add_tool(CreateDirectoryTool::new(self.project.clone()));
971 self.add_tool(DeletePathTool::new(
972 self.project.clone(),
973 self.action_log.clone(),
974 ));
975 self.add_tool(DiagnosticsTool::new(self.project.clone()));
976 self.add_tool(EditFileTool::new(
977 self.project.clone(),
978 cx.weak_entity(),
979 language_registry,
980 Templates::new(),
981 ));
982 self.add_tool(FetchTool::new(self.project.read(cx).client().http_client()));
983 self.add_tool(FindPathTool::new(self.project.clone()));
984 self.add_tool(GrepTool::new(self.project.clone()));
985 self.add_tool(HeadTool::new(self.project.clone()));
986 self.add_tool(ListDirectoryTool::new(self.project.clone()));
987 self.add_tool(MovePathTool::new(self.project.clone()));
988 self.add_tool(NowTool);
989 self.add_tool(OpenTool::new(self.project.clone()));
990 self.add_tool(ReadFileTool::new(
991 self.project.clone(),
992 self.action_log.clone(),
993 ));
994 self.add_tool(TerminalTool::new(self.project.clone(), environment));
995 self.add_tool(ThinkingTool);
996 self.add_tool(WebSearchTool);
997 }
998
999 pub fn add_tool<T: AgentTool>(&mut self, tool: T) {
1000 self.tools.insert(T::name().into(), tool.erase());
1001 }
1002
1003 pub fn remove_tool(&mut self, name: &str) -> bool {
1004 self.tools.remove(name).is_some()
1005 }
1006
1007 pub fn profile(&self) -> &AgentProfileId {
1008 &self.profile_id
1009 }
1010
1011 pub fn set_profile(&mut self, profile_id: AgentProfileId) {
1012 self.profile_id = profile_id;
1013 }
1014
1015 pub fn cancel(&mut self, cx: &mut Context<Self>) {
1016 if let Some(running_turn) = self.running_turn.take() {
1017 running_turn.cancel();
1018 }
1019 self.flush_pending_message(cx);
1020 }
1021
1022 fn update_token_usage(&mut self, update: language_model::TokenUsage, cx: &mut Context<Self>) {
1023 let Some(last_user_message) = self.last_user_message() else {
1024 return;
1025 };
1026
1027 self.request_token_usage
1028 .insert(last_user_message.id.clone(), update);
1029 cx.emit(TokenUsageUpdated(self.latest_token_usage()));
1030 cx.notify();
1031 }
1032
1033 pub fn truncate(&mut self, message_id: UserMessageId, cx: &mut Context<Self>) -> Result<()> {
1034 self.cancel(cx);
1035 let Some(position) = self.messages.iter().position(
1036 |msg| matches!(msg, Message::User(UserMessage { id, .. }) if id == &message_id),
1037 ) else {
1038 return Err(anyhow!("Message not found"));
1039 };
1040
1041 for message in self.messages.drain(position..) {
1042 match message {
1043 Message::User(message) => {
1044 self.request_token_usage.remove(&message.id);
1045 }
1046 Message::Agent(_) | Message::Resume => {}
1047 }
1048 }
1049 self.clear_summary();
1050 cx.notify();
1051 Ok(())
1052 }
1053
1054 pub fn latest_request_token_usage(&self) -> Option<language_model::TokenUsage> {
1055 let last_user_message = self.last_user_message()?;
1056 let tokens = self.request_token_usage.get(&last_user_message.id)?;
1057 Some(*tokens)
1058 }
1059
1060 pub fn latest_token_usage(&self) -> Option<acp_thread::TokenUsage> {
1061 let usage = self.latest_request_token_usage()?;
1062 let model = self.model.clone()?;
1063 Some(acp_thread::TokenUsage {
1064 max_tokens: model.max_token_count_for_mode(self.completion_mode.into()),
1065 used_tokens: usage.total_tokens(),
1066 })
1067 }
1068
1069 pub fn resume(
1070 &mut self,
1071 cx: &mut Context<Self>,
1072 ) -> Result<mpsc::UnboundedReceiver<Result<ThreadEvent>>> {
1073 self.messages.push(Message::Resume);
1074 cx.notify();
1075
1076 log::debug!("Total messages in thread: {}", self.messages.len());
1077 self.run_turn(cx)
1078 }
1079
1080 /// Sending a message results in the model streaming a response, which could include tool calls.
1081 /// After calling tools, the model will stops and waits for any outstanding tool calls to be completed and their results sent.
1082 /// The returned channel will report all the occurrences in which the model stops before erroring or ending its turn.
1083 pub fn send<T>(
1084 &mut self,
1085 id: UserMessageId,
1086 content: impl IntoIterator<Item = T>,
1087 cx: &mut Context<Self>,
1088 ) -> Result<mpsc::UnboundedReceiver<Result<ThreadEvent>>>
1089 where
1090 T: Into<UserMessageContent>,
1091 {
1092 let model = self.model().context("No language model configured")?;
1093
1094 log::info!("Thread::send called with model: {}", model.name().0);
1095 self.advance_prompt_id();
1096
1097 let content = content.into_iter().map(Into::into).collect::<Vec<_>>();
1098 log::debug!("Thread::send content: {:?}", content);
1099
1100 self.messages
1101 .push(Message::User(UserMessage { id, content }));
1102 cx.notify();
1103
1104 log::debug!("Total messages in thread: {}", self.messages.len());
1105 self.run_turn(cx)
1106 }
1107
1108 #[cfg(feature = "eval")]
1109 pub fn proceed(
1110 &mut self,
1111 cx: &mut Context<Self>,
1112 ) -> Result<mpsc::UnboundedReceiver<Result<ThreadEvent>>> {
1113 self.run_turn(cx)
1114 }
1115
1116 fn run_turn(
1117 &mut self,
1118 cx: &mut Context<Self>,
1119 ) -> Result<mpsc::UnboundedReceiver<Result<ThreadEvent>>> {
1120 self.cancel(cx);
1121
1122 let model = self.model.clone().context("No language model configured")?;
1123 let profile = AgentSettings::get_global(cx)
1124 .profiles
1125 .get(&self.profile_id)
1126 .context("Profile not found")?;
1127 let (events_tx, events_rx) = mpsc::unbounded::<Result<ThreadEvent>>();
1128 let event_stream = ThreadEventStream(events_tx);
1129 let message_ix = self.messages.len().saturating_sub(1);
1130 self.tool_use_limit_reached = false;
1131 self.clear_summary();
1132 self.running_turn = Some(RunningTurn {
1133 event_stream: event_stream.clone(),
1134 tools: self.enabled_tools(profile, &model, cx),
1135 _task: cx.spawn(async move |this, cx| {
1136 log::debug!("Starting agent turn execution");
1137
1138 let turn_result = Self::run_turn_internal(&this, model, &event_stream, cx).await;
1139 _ = this.update(cx, |this, cx| this.flush_pending_message(cx));
1140
1141 match turn_result {
1142 Ok(()) => {
1143 log::debug!("Turn execution completed");
1144 event_stream.send_stop(acp::StopReason::EndTurn);
1145 }
1146 Err(error) => {
1147 log::error!("Turn execution failed: {:?}", error);
1148 match error.downcast::<CompletionError>() {
1149 Ok(CompletionError::Refusal) => {
1150 event_stream.send_stop(acp::StopReason::Refusal);
1151 _ = this.update(cx, |this, _| this.messages.truncate(message_ix));
1152 }
1153 Ok(CompletionError::MaxTokens) => {
1154 event_stream.send_stop(acp::StopReason::MaxTokens);
1155 }
1156 Ok(CompletionError::Other(error)) | Err(error) => {
1157 event_stream.send_error(error);
1158 }
1159 }
1160 }
1161 }
1162
1163 _ = this.update(cx, |this, _| this.running_turn.take());
1164 }),
1165 });
1166 Ok(events_rx)
1167 }
1168
1169 async fn run_turn_internal(
1170 this: &WeakEntity<Self>,
1171 model: Arc<dyn LanguageModel>,
1172 event_stream: &ThreadEventStream,
1173 cx: &mut AsyncApp,
1174 ) -> Result<()> {
1175 let mut attempt = 0;
1176 let mut intent = CompletionIntent::UserPrompt;
1177 loop {
1178 let request =
1179 this.update(cx, |this, cx| this.build_completion_request(intent, cx))??;
1180
1181 telemetry::event!(
1182 "Agent Thread Completion",
1183 thread_id = this.read_with(cx, |this, _| this.id.to_string())?,
1184 prompt_id = this.read_with(cx, |this, _| this.prompt_id.to_string())?,
1185 model = model.telemetry_id(),
1186 model_provider = model.provider_id().to_string(),
1187 attempt
1188 );
1189
1190 log::debug!("Calling model.stream_completion, attempt {}", attempt);
1191
1192 let (mut events, mut error) = match model.stream_completion(request, cx).await {
1193 Ok(events) => (events, None),
1194 Err(err) => (stream::empty().boxed(), Some(err)),
1195 };
1196 let mut tool_results = FuturesUnordered::new();
1197 while let Some(event) = events.next().await {
1198 log::trace!("Received completion event: {:?}", event);
1199 match event {
1200 Ok(event) => {
1201 tool_results.extend(this.update(cx, |this, cx| {
1202 this.handle_completion_event(event, event_stream, cx)
1203 })??);
1204 }
1205 Err(err) => {
1206 error = Some(err);
1207 break;
1208 }
1209 }
1210 }
1211
1212 let end_turn = tool_results.is_empty();
1213 while let Some(tool_result) = tool_results.next().await {
1214 log::debug!("Tool finished {:?}", tool_result);
1215
1216 event_stream.update_tool_call_fields(
1217 &tool_result.tool_use_id,
1218 acp::ToolCallUpdateFields {
1219 status: Some(if tool_result.is_error {
1220 acp::ToolCallStatus::Failed
1221 } else {
1222 acp::ToolCallStatus::Completed
1223 }),
1224 raw_output: tool_result.output.clone(),
1225 ..Default::default()
1226 },
1227 );
1228 this.update(cx, |this, _cx| {
1229 this.pending_message()
1230 .tool_results
1231 .insert(tool_result.tool_use_id.clone(), tool_result);
1232 })?;
1233 }
1234
1235 this.update(cx, |this, cx| {
1236 this.flush_pending_message(cx);
1237 if this.title.is_none() && this.pending_title_generation.is_none() {
1238 this.generate_title(cx);
1239 }
1240 })?;
1241
1242 if let Some(error) = error {
1243 attempt += 1;
1244 let retry = this.update(cx, |this, cx| {
1245 let user_store = this.user_store.read(cx);
1246 this.handle_completion_error(error, attempt, user_store.plan())
1247 })??;
1248 let timer = cx.background_executor().timer(retry.duration);
1249 event_stream.send_retry(retry);
1250 timer.await;
1251 this.update(cx, |this, _cx| {
1252 if let Some(Message::Agent(message)) = this.messages.last() {
1253 if message.tool_results.is_empty() {
1254 intent = CompletionIntent::UserPrompt;
1255 this.messages.push(Message::Resume);
1256 }
1257 }
1258 })?;
1259 } else if this.read_with(cx, |this, _| this.tool_use_limit_reached)? {
1260 return Err(language_model::ToolUseLimitReachedError.into());
1261 } else if end_turn {
1262 return Ok(());
1263 } else {
1264 intent = CompletionIntent::ToolResults;
1265 attempt = 0;
1266 }
1267 }
1268 }
1269
1270 fn handle_completion_error(
1271 &mut self,
1272 error: LanguageModelCompletionError,
1273 attempt: u8,
1274 plan: Option<Plan>,
1275 ) -> Result<acp_thread::RetryStatus> {
1276 let Some(model) = self.model.as_ref() else {
1277 return Err(anyhow!(error));
1278 };
1279
1280 let auto_retry = if model.provider_id() == ZED_CLOUD_PROVIDER_ID {
1281 match plan {
1282 Some(Plan::V2(_)) => true,
1283 Some(Plan::V1(_)) => self.completion_mode == CompletionMode::Burn,
1284 None => false,
1285 }
1286 } else {
1287 true
1288 };
1289
1290 if !auto_retry {
1291 return Err(anyhow!(error));
1292 }
1293
1294 let Some(strategy) = Self::retry_strategy_for(&error) else {
1295 return Err(anyhow!(error));
1296 };
1297
1298 let max_attempts = match &strategy {
1299 RetryStrategy::ExponentialBackoff { max_attempts, .. } => *max_attempts,
1300 RetryStrategy::Fixed { max_attempts, .. } => *max_attempts,
1301 };
1302
1303 if attempt > max_attempts {
1304 return Err(anyhow!(error));
1305 }
1306
1307 let delay = match &strategy {
1308 RetryStrategy::ExponentialBackoff { initial_delay, .. } => {
1309 let delay_secs = initial_delay.as_secs() * 2u64.pow((attempt - 1) as u32);
1310 Duration::from_secs(delay_secs)
1311 }
1312 RetryStrategy::Fixed { delay, .. } => *delay,
1313 };
1314 log::debug!("Retry attempt {attempt} with delay {delay:?}");
1315
1316 Ok(acp_thread::RetryStatus {
1317 last_error: error.to_string().into(),
1318 attempt: attempt as usize,
1319 max_attempts: max_attempts as usize,
1320 started_at: Instant::now(),
1321 duration: delay,
1322 })
1323 }
1324
1325 /// A helper method that's called on every streamed completion event.
1326 /// Returns an optional tool result task, which the main agentic loop will
1327 /// send back to the model when it resolves.
1328 fn handle_completion_event(
1329 &mut self,
1330 event: LanguageModelCompletionEvent,
1331 event_stream: &ThreadEventStream,
1332 cx: &mut Context<Self>,
1333 ) -> Result<Option<Task<LanguageModelToolResult>>> {
1334 log::trace!("Handling streamed completion event: {:?}", event);
1335 use LanguageModelCompletionEvent::*;
1336
1337 match event {
1338 StartMessage { .. } => {
1339 self.flush_pending_message(cx);
1340 self.pending_message = Some(AgentMessage::default());
1341 }
1342 Text(new_text) => self.handle_text_event(new_text, event_stream, cx),
1343 Thinking { text, signature } => {
1344 self.handle_thinking_event(text, signature, event_stream, cx)
1345 }
1346 RedactedThinking { data } => self.handle_redacted_thinking_event(data, cx),
1347 ToolUse(tool_use) => {
1348 return Ok(self.handle_tool_use_event(tool_use, event_stream, cx));
1349 }
1350 ToolUseJsonParseError {
1351 id,
1352 tool_name,
1353 raw_input,
1354 json_parse_error,
1355 } => {
1356 return Ok(Some(Task::ready(
1357 self.handle_tool_use_json_parse_error_event(
1358 id,
1359 tool_name,
1360 raw_input,
1361 json_parse_error,
1362 ),
1363 )));
1364 }
1365 UsageUpdate(usage) => {
1366 telemetry::event!(
1367 "Agent Thread Completion Usage Updated",
1368 thread_id = self.id.to_string(),
1369 prompt_id = self.prompt_id.to_string(),
1370 model = self.model.as_ref().map(|m| m.telemetry_id()),
1371 model_provider = self.model.as_ref().map(|m| m.provider_id().to_string()),
1372 input_tokens = usage.input_tokens,
1373 output_tokens = usage.output_tokens,
1374 cache_creation_input_tokens = usage.cache_creation_input_tokens,
1375 cache_read_input_tokens = usage.cache_read_input_tokens,
1376 );
1377 self.update_token_usage(usage, cx);
1378 }
1379 StatusUpdate(CompletionRequestStatus::UsageUpdated { amount, limit }) => {
1380 self.update_model_request_usage(amount, limit, cx);
1381 }
1382 StatusUpdate(
1383 CompletionRequestStatus::Started
1384 | CompletionRequestStatus::Queued { .. }
1385 | CompletionRequestStatus::Failed { .. },
1386 ) => {}
1387 StatusUpdate(CompletionRequestStatus::ToolUseLimitReached) => {
1388 self.tool_use_limit_reached = true;
1389 }
1390 Stop(StopReason::Refusal) => return Err(CompletionError::Refusal.into()),
1391 Stop(StopReason::MaxTokens) => return Err(CompletionError::MaxTokens.into()),
1392 Stop(StopReason::ToolUse | StopReason::EndTurn) => {}
1393 }
1394
1395 Ok(None)
1396 }
1397
1398 fn handle_text_event(
1399 &mut self,
1400 new_text: String,
1401 event_stream: &ThreadEventStream,
1402 cx: &mut Context<Self>,
1403 ) {
1404 event_stream.send_text(&new_text);
1405
1406 let last_message = self.pending_message();
1407 if let Some(AgentMessageContent::Text(text)) = last_message.content.last_mut() {
1408 text.push_str(&new_text);
1409 } else {
1410 last_message
1411 .content
1412 .push(AgentMessageContent::Text(new_text));
1413 }
1414
1415 cx.notify();
1416 }
1417
1418 fn handle_thinking_event(
1419 &mut self,
1420 new_text: String,
1421 new_signature: Option<String>,
1422 event_stream: &ThreadEventStream,
1423 cx: &mut Context<Self>,
1424 ) {
1425 event_stream.send_thinking(&new_text);
1426
1427 let last_message = self.pending_message();
1428 if let Some(AgentMessageContent::Thinking { text, signature }) =
1429 last_message.content.last_mut()
1430 {
1431 text.push_str(&new_text);
1432 *signature = new_signature.or(signature.take());
1433 } else {
1434 last_message.content.push(AgentMessageContent::Thinking {
1435 text: new_text,
1436 signature: new_signature,
1437 });
1438 }
1439
1440 cx.notify();
1441 }
1442
1443 fn handle_redacted_thinking_event(&mut self, data: String, cx: &mut Context<Self>) {
1444 let last_message = self.pending_message();
1445 last_message
1446 .content
1447 .push(AgentMessageContent::RedactedThinking(data));
1448 cx.notify();
1449 }
1450
1451 fn handle_tool_use_event(
1452 &mut self,
1453 tool_use: LanguageModelToolUse,
1454 event_stream: &ThreadEventStream,
1455 cx: &mut Context<Self>,
1456 ) -> Option<Task<LanguageModelToolResult>> {
1457 cx.notify();
1458
1459 let tool = self.tool(tool_use.name.as_ref());
1460 let mut title = SharedString::from(&tool_use.name);
1461 let mut kind = acp::ToolKind::Other;
1462 if let Some(tool) = tool.as_ref() {
1463 title = tool.initial_title(tool_use.input.clone(), cx);
1464 kind = tool.kind();
1465 }
1466
1467 // Ensure the last message ends in the current tool use
1468 let last_message = self.pending_message();
1469 let push_new_tool_use = last_message.content.last_mut().is_none_or(|content| {
1470 if let AgentMessageContent::ToolUse(last_tool_use) = content {
1471 if last_tool_use.id == tool_use.id {
1472 *last_tool_use = tool_use.clone();
1473 false
1474 } else {
1475 true
1476 }
1477 } else {
1478 true
1479 }
1480 });
1481
1482 if push_new_tool_use {
1483 event_stream.send_tool_call(
1484 &tool_use.id,
1485 &tool_use.name,
1486 title,
1487 kind,
1488 tool_use.input.clone(),
1489 );
1490 last_message
1491 .content
1492 .push(AgentMessageContent::ToolUse(tool_use.clone()));
1493 } else {
1494 event_stream.update_tool_call_fields(
1495 &tool_use.id,
1496 acp::ToolCallUpdateFields {
1497 title: Some(title.into()),
1498 kind: Some(kind),
1499 raw_input: Some(tool_use.input.clone()),
1500 ..Default::default()
1501 },
1502 );
1503 }
1504
1505 if !tool_use.is_input_complete {
1506 return None;
1507 }
1508
1509 let Some(tool) = tool else {
1510 let content = format!("No tool named {} exists", tool_use.name);
1511 return Some(Task::ready(LanguageModelToolResult {
1512 content: LanguageModelToolResultContent::Text(Arc::from(content)),
1513 tool_use_id: tool_use.id,
1514 tool_name: tool_use.name,
1515 is_error: true,
1516 output: None,
1517 }));
1518 };
1519
1520 let fs = self.project.read(cx).fs().clone();
1521 let tool_event_stream =
1522 ToolCallEventStream::new(tool_use.id.clone(), event_stream.clone(), Some(fs));
1523 tool_event_stream.update_fields(acp::ToolCallUpdateFields {
1524 status: Some(acp::ToolCallStatus::InProgress),
1525 ..Default::default()
1526 });
1527 let supports_images = self.model().is_some_and(|model| model.supports_images());
1528 let tool_result = tool.run(tool_use.input, tool_event_stream, cx);
1529 log::debug!("Running tool {}", tool_use.name);
1530 Some(cx.foreground_executor().spawn(async move {
1531 let tool_result = tool_result.await.and_then(|output| {
1532 if let LanguageModelToolResultContent::Image(_) = &output.llm_output
1533 && !supports_images
1534 {
1535 return Err(anyhow!(
1536 "Attempted to read an image, but this model doesn't support it.",
1537 ));
1538 }
1539 Ok(output)
1540 });
1541
1542 match tool_result {
1543 Ok(output) => LanguageModelToolResult {
1544 tool_use_id: tool_use.id,
1545 tool_name: tool_use.name,
1546 is_error: false,
1547 content: output.llm_output,
1548 output: Some(output.raw_output),
1549 },
1550 Err(error) => LanguageModelToolResult {
1551 tool_use_id: tool_use.id,
1552 tool_name: tool_use.name,
1553 is_error: true,
1554 content: LanguageModelToolResultContent::Text(Arc::from(error.to_string())),
1555 output: Some(error.to_string().into()),
1556 },
1557 }
1558 }))
1559 }
1560
1561 fn handle_tool_use_json_parse_error_event(
1562 &mut self,
1563 tool_use_id: LanguageModelToolUseId,
1564 tool_name: Arc<str>,
1565 raw_input: Arc<str>,
1566 json_parse_error: String,
1567 ) -> LanguageModelToolResult {
1568 let tool_output = format!("Error parsing input JSON: {json_parse_error}");
1569 LanguageModelToolResult {
1570 tool_use_id,
1571 tool_name,
1572 is_error: true,
1573 content: LanguageModelToolResultContent::Text(tool_output.into()),
1574 output: Some(serde_json::Value::String(raw_input.to_string())),
1575 }
1576 }
1577
1578 fn update_model_request_usage(&self, amount: usize, limit: UsageLimit, cx: &mut Context<Self>) {
1579 self.project
1580 .read(cx)
1581 .user_store()
1582 .update(cx, |user_store, cx| {
1583 user_store.update_model_request_usage(
1584 ModelRequestUsage(RequestUsage {
1585 amount: amount as i32,
1586 limit,
1587 }),
1588 cx,
1589 )
1590 });
1591 }
1592
1593 pub fn title(&self) -> SharedString {
1594 self.title.clone().unwrap_or("New Thread".into())
1595 }
1596
1597 pub fn is_generating_summary(&self) -> bool {
1598 self.pending_summary_generation.is_some()
1599 }
1600
1601 pub fn summary(&mut self, cx: &mut Context<Self>) -> Shared<Task<Option<SharedString>>> {
1602 if let Some(summary) = self.summary.as_ref() {
1603 return Task::ready(Some(summary.clone())).shared();
1604 }
1605 if let Some(task) = self.pending_summary_generation.clone() {
1606 return task;
1607 }
1608 let Some(model) = self.summarization_model.clone() else {
1609 log::error!("No summarization model available");
1610 return Task::ready(None).shared();
1611 };
1612 let mut request = LanguageModelRequest {
1613 intent: Some(CompletionIntent::ThreadContextSummarization),
1614 temperature: AgentSettings::temperature_for_model(&model, cx),
1615 ..Default::default()
1616 };
1617
1618 for message in &self.messages {
1619 request.messages.extend(message.to_request());
1620 }
1621
1622 request.messages.push(LanguageModelRequestMessage {
1623 role: Role::User,
1624 content: vec![SUMMARIZE_THREAD_DETAILED_PROMPT.into()],
1625 cache: false,
1626 });
1627
1628 let task = cx
1629 .spawn(async move |this, cx| {
1630 let mut summary = String::new();
1631 let mut messages = model.stream_completion(request, cx).await.log_err()?;
1632 while let Some(event) = messages.next().await {
1633 let event = event.log_err()?;
1634 let text = match event {
1635 LanguageModelCompletionEvent::Text(text) => text,
1636 LanguageModelCompletionEvent::StatusUpdate(
1637 CompletionRequestStatus::UsageUpdated { amount, limit },
1638 ) => {
1639 this.update(cx, |thread, cx| {
1640 thread.update_model_request_usage(amount, limit, cx);
1641 })
1642 .ok()?;
1643 continue;
1644 }
1645 _ => continue,
1646 };
1647
1648 let mut lines = text.lines();
1649 summary.extend(lines.next());
1650 }
1651
1652 log::debug!("Setting summary: {}", summary);
1653 let summary = SharedString::from(summary);
1654
1655 this.update(cx, |this, cx| {
1656 this.summary = Some(summary.clone());
1657 this.pending_summary_generation = None;
1658 cx.notify()
1659 })
1660 .ok()?;
1661
1662 Some(summary)
1663 })
1664 .shared();
1665 self.pending_summary_generation = Some(task.clone());
1666 task
1667 }
1668
1669 fn generate_title(&mut self, cx: &mut Context<Self>) {
1670 let Some(model) = self.summarization_model.clone() else {
1671 return;
1672 };
1673
1674 log::debug!(
1675 "Generating title with model: {:?}",
1676 self.summarization_model.as_ref().map(|model| model.name())
1677 );
1678 let mut request = LanguageModelRequest {
1679 intent: Some(CompletionIntent::ThreadSummarization),
1680 temperature: AgentSettings::temperature_for_model(&model, cx),
1681 ..Default::default()
1682 };
1683
1684 for message in &self.messages {
1685 request.messages.extend(message.to_request());
1686 }
1687
1688 request.messages.push(LanguageModelRequestMessage {
1689 role: Role::User,
1690 content: vec![SUMMARIZE_THREAD_PROMPT.into()],
1691 cache: false,
1692 });
1693 self.pending_title_generation = Some(cx.spawn(async move |this, cx| {
1694 let mut title = String::new();
1695
1696 let generate = async {
1697 let mut messages = model.stream_completion(request, cx).await?;
1698 while let Some(event) = messages.next().await {
1699 let event = event?;
1700 let text = match event {
1701 LanguageModelCompletionEvent::Text(text) => text,
1702 LanguageModelCompletionEvent::StatusUpdate(
1703 CompletionRequestStatus::UsageUpdated { amount, limit },
1704 ) => {
1705 this.update(cx, |thread, cx| {
1706 thread.update_model_request_usage(amount, limit, cx);
1707 })?;
1708 continue;
1709 }
1710 _ => continue,
1711 };
1712
1713 let mut lines = text.lines();
1714 title.extend(lines.next());
1715
1716 // Stop if the LLM generated multiple lines.
1717 if lines.next().is_some() {
1718 break;
1719 }
1720 }
1721 anyhow::Ok(())
1722 };
1723
1724 if generate.await.context("failed to generate title").is_ok() {
1725 _ = this.update(cx, |this, cx| this.set_title(title.into(), cx));
1726 }
1727 _ = this.update(cx, |this, _| this.pending_title_generation = None);
1728 }));
1729 }
1730
1731 pub fn set_title(&mut self, title: SharedString, cx: &mut Context<Self>) {
1732 self.pending_title_generation = None;
1733 if Some(&title) != self.title.as_ref() {
1734 self.title = Some(title);
1735 cx.emit(TitleUpdated);
1736 cx.notify();
1737 }
1738 }
1739
1740 fn clear_summary(&mut self) {
1741 self.summary = None;
1742 self.pending_summary_generation = None;
1743 }
1744
1745 fn last_user_message(&self) -> Option<&UserMessage> {
1746 self.messages
1747 .iter()
1748 .rev()
1749 .find_map(|message| match message {
1750 Message::User(user_message) => Some(user_message),
1751 Message::Agent(_) => None,
1752 Message::Resume => None,
1753 })
1754 }
1755
1756 fn pending_message(&mut self) -> &mut AgentMessage {
1757 self.pending_message.get_or_insert_default()
1758 }
1759
1760 fn flush_pending_message(&mut self, cx: &mut Context<Self>) {
1761 let Some(mut message) = self.pending_message.take() else {
1762 return;
1763 };
1764
1765 if message.content.is_empty() {
1766 return;
1767 }
1768
1769 for content in &message.content {
1770 let AgentMessageContent::ToolUse(tool_use) = content else {
1771 continue;
1772 };
1773
1774 if !message.tool_results.contains_key(&tool_use.id) {
1775 message.tool_results.insert(
1776 tool_use.id.clone(),
1777 LanguageModelToolResult {
1778 tool_use_id: tool_use.id.clone(),
1779 tool_name: tool_use.name.clone(),
1780 is_error: true,
1781 content: LanguageModelToolResultContent::Text(TOOL_CANCELED_MESSAGE.into()),
1782 output: None,
1783 },
1784 );
1785 }
1786 }
1787
1788 self.messages.push(Message::Agent(message));
1789 self.updated_at = Utc::now();
1790 self.clear_summary();
1791 cx.notify()
1792 }
1793
1794 pub(crate) fn build_completion_request(
1795 &self,
1796 completion_intent: CompletionIntent,
1797 cx: &App,
1798 ) -> Result<LanguageModelRequest> {
1799 let model = self.model().context("No language model configured")?;
1800 let tools = if let Some(turn) = self.running_turn.as_ref() {
1801 turn.tools
1802 .iter()
1803 .filter_map(|(tool_name, tool)| {
1804 log::trace!("Including tool: {}", tool_name);
1805 Some(LanguageModelRequestTool {
1806 name: tool_name.to_string(),
1807 description: tool.description().to_string(),
1808 input_schema: tool.input_schema(model.tool_input_format()).log_err()?,
1809 })
1810 })
1811 .collect::<Vec<_>>()
1812 } else {
1813 Vec::new()
1814 };
1815
1816 log::debug!("Building completion request");
1817 log::debug!("Completion intent: {:?}", completion_intent);
1818 log::debug!("Completion mode: {:?}", self.completion_mode);
1819
1820 let available_tools: Vec<_> = self
1821 .running_turn
1822 .as_ref()
1823 .map(|turn| turn.tools.keys().cloned().collect())
1824 .unwrap_or_default();
1825
1826 log::debug!("Request includes {} tools", available_tools.len());
1827 let messages = self.build_request_messages(available_tools, cx);
1828 log::debug!("Request will include {} messages", messages.len());
1829
1830 let request = LanguageModelRequest {
1831 thread_id: Some(self.id.to_string()),
1832 prompt_id: Some(self.prompt_id.to_string()),
1833 intent: Some(completion_intent),
1834 mode: Some(self.completion_mode.into()),
1835 messages,
1836 tools,
1837 tool_choice: None,
1838 stop: Vec::new(),
1839 temperature: AgentSettings::temperature_for_model(model, cx),
1840 thinking_allowed: true,
1841 };
1842
1843 log::debug!("Completion request built successfully");
1844 Ok(request)
1845 }
1846
1847 fn enabled_tools(
1848 &self,
1849 profile: &AgentProfileSettings,
1850 model: &Arc<dyn LanguageModel>,
1851 cx: &App,
1852 ) -> BTreeMap<SharedString, Arc<dyn AnyAgentTool>> {
1853 fn truncate(tool_name: &SharedString) -> SharedString {
1854 if tool_name.len() > MAX_TOOL_NAME_LENGTH {
1855 let mut truncated = tool_name.to_string();
1856 truncated.truncate(MAX_TOOL_NAME_LENGTH);
1857 truncated.into()
1858 } else {
1859 tool_name.clone()
1860 }
1861 }
1862
1863 let mut tools = self
1864 .tools
1865 .iter()
1866 .filter_map(|(tool_name, tool)| {
1867 if tool.supports_provider(&model.provider_id())
1868 && profile.is_tool_enabled(tool_name)
1869 {
1870 Some((truncate(tool_name), tool.clone()))
1871 } else {
1872 None
1873 }
1874 })
1875 .collect::<BTreeMap<_, _>>();
1876
1877 let mut context_server_tools = Vec::new();
1878 let mut seen_tools = tools.keys().cloned().collect::<HashSet<_>>();
1879 let mut duplicate_tool_names = HashSet::default();
1880 for (server_id, server_tools) in self.context_server_registry.read(cx).servers() {
1881 for (tool_name, tool) in server_tools {
1882 if profile.is_context_server_tool_enabled(&server_id.0, &tool_name) {
1883 let tool_name = truncate(tool_name);
1884 if !seen_tools.insert(tool_name.clone()) {
1885 duplicate_tool_names.insert(tool_name.clone());
1886 }
1887 context_server_tools.push((server_id.clone(), tool_name, tool.clone()));
1888 }
1889 }
1890 }
1891
1892 // When there are duplicate tool names, disambiguate by prefixing them
1893 // with the server ID. In the rare case there isn't enough space for the
1894 // disambiguated tool name, keep only the last tool with this name.
1895 for (server_id, tool_name, tool) in context_server_tools {
1896 if duplicate_tool_names.contains(&tool_name) {
1897 let available = MAX_TOOL_NAME_LENGTH.saturating_sub(tool_name.len());
1898 if available >= 2 {
1899 let mut disambiguated = server_id.0.to_string();
1900 disambiguated.truncate(available - 1);
1901 disambiguated.push('_');
1902 disambiguated.push_str(&tool_name);
1903 tools.insert(disambiguated.into(), tool.clone());
1904 } else {
1905 tools.insert(tool_name, tool.clone());
1906 }
1907 } else {
1908 tools.insert(tool_name, tool.clone());
1909 }
1910 }
1911
1912 tools
1913 }
1914
1915 fn tool(&self, name: &str) -> Option<Arc<dyn AnyAgentTool>> {
1916 self.running_turn.as_ref()?.tools.get(name).cloned()
1917 }
1918
1919 fn build_request_messages(
1920 &self,
1921 available_tools: Vec<SharedString>,
1922 cx: &App,
1923 ) -> Vec<LanguageModelRequestMessage> {
1924 log::trace!(
1925 "Building request messages from {} thread messages",
1926 self.messages.len()
1927 );
1928
1929 let system_prompt = SystemPromptTemplate {
1930 project: self.project_context.read(cx),
1931 available_tools,
1932 model_name: self.model.as_ref().map(|m| m.name().0.to_string()),
1933 }
1934 .render(&self.templates)
1935 .context("failed to build system prompt")
1936 .expect("Invalid template");
1937 let mut messages = vec![LanguageModelRequestMessage {
1938 role: Role::System,
1939 content: vec![system_prompt.into()],
1940 cache: false,
1941 }];
1942 for message in &self.messages {
1943 messages.extend(message.to_request());
1944 }
1945
1946 if let Some(last_message) = messages.last_mut() {
1947 last_message.cache = true;
1948 }
1949
1950 if let Some(message) = self.pending_message.as_ref() {
1951 messages.extend(message.to_request());
1952 }
1953
1954 messages
1955 }
1956
1957 pub fn to_markdown(&self) -> String {
1958 let mut markdown = String::new();
1959 for (ix, message) in self.messages.iter().enumerate() {
1960 if ix > 0 {
1961 markdown.push('\n');
1962 }
1963 markdown.push_str(&message.to_markdown());
1964 }
1965
1966 if let Some(message) = self.pending_message.as_ref() {
1967 markdown.push('\n');
1968 markdown.push_str(&message.to_markdown());
1969 }
1970
1971 markdown
1972 }
1973
1974 fn advance_prompt_id(&mut self) {
1975 self.prompt_id = PromptId::new();
1976 }
1977
1978 fn retry_strategy_for(error: &LanguageModelCompletionError) -> Option<RetryStrategy> {
1979 use LanguageModelCompletionError::*;
1980 use http_client::StatusCode;
1981
1982 // General strategy here:
1983 // - If retrying won't help (e.g. invalid API key or payload too large), return None so we don't retry at all.
1984 // - If it's a time-based issue (e.g. server overloaded, rate limit exceeded), retry up to 4 times with exponential backoff.
1985 // - If it's an issue that *might* be fixed by retrying (e.g. internal server error), retry up to 3 times.
1986 match error {
1987 HttpResponseError {
1988 status_code: StatusCode::TOO_MANY_REQUESTS,
1989 ..
1990 } => Some(RetryStrategy::ExponentialBackoff {
1991 initial_delay: BASE_RETRY_DELAY,
1992 max_attempts: MAX_RETRY_ATTEMPTS,
1993 }),
1994 ServerOverloaded { retry_after, .. } | RateLimitExceeded { retry_after, .. } => {
1995 Some(RetryStrategy::Fixed {
1996 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
1997 max_attempts: MAX_RETRY_ATTEMPTS,
1998 })
1999 }
2000 UpstreamProviderError {
2001 status,
2002 retry_after,
2003 ..
2004 } => match *status {
2005 StatusCode::TOO_MANY_REQUESTS | StatusCode::SERVICE_UNAVAILABLE => {
2006 Some(RetryStrategy::Fixed {
2007 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
2008 max_attempts: MAX_RETRY_ATTEMPTS,
2009 })
2010 }
2011 StatusCode::INTERNAL_SERVER_ERROR => Some(RetryStrategy::Fixed {
2012 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
2013 // Internal Server Error could be anything, retry up to 3 times.
2014 max_attempts: 3,
2015 }),
2016 status => {
2017 // There is no StatusCode variant for the unofficial HTTP 529 ("The service is overloaded"),
2018 // but we frequently get them in practice. See https://http.dev/529
2019 if status.as_u16() == 529 {
2020 Some(RetryStrategy::Fixed {
2021 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
2022 max_attempts: MAX_RETRY_ATTEMPTS,
2023 })
2024 } else {
2025 Some(RetryStrategy::Fixed {
2026 delay: retry_after.unwrap_or(BASE_RETRY_DELAY),
2027 max_attempts: 2,
2028 })
2029 }
2030 }
2031 },
2032 ApiInternalServerError { .. } => Some(RetryStrategy::Fixed {
2033 delay: BASE_RETRY_DELAY,
2034 max_attempts: 3,
2035 }),
2036 ApiReadResponseError { .. }
2037 | HttpSend { .. }
2038 | DeserializeResponse { .. }
2039 | BadRequestFormat { .. } => Some(RetryStrategy::Fixed {
2040 delay: BASE_RETRY_DELAY,
2041 max_attempts: 3,
2042 }),
2043 // Retrying these errors definitely shouldn't help.
2044 HttpResponseError {
2045 status_code:
2046 StatusCode::PAYLOAD_TOO_LARGE | StatusCode::FORBIDDEN | StatusCode::UNAUTHORIZED,
2047 ..
2048 }
2049 | AuthenticationError { .. }
2050 | PermissionError { .. }
2051 | NoApiKey { .. }
2052 | ApiEndpointNotFound { .. }
2053 | PromptTooLarge { .. } => None,
2054 // These errors might be transient, so retry them
2055 SerializeRequest { .. } | BuildRequestBody { .. } => Some(RetryStrategy::Fixed {
2056 delay: BASE_RETRY_DELAY,
2057 max_attempts: 1,
2058 }),
2059 // Retry all other 4xx and 5xx errors once.
2060 HttpResponseError { status_code, .. }
2061 if status_code.is_client_error() || status_code.is_server_error() =>
2062 {
2063 Some(RetryStrategy::Fixed {
2064 delay: BASE_RETRY_DELAY,
2065 max_attempts: 3,
2066 })
2067 }
2068 Other(err)
2069 if err.is::<language_model::PaymentRequiredError>()
2070 || err.is::<language_model::ModelRequestLimitReachedError>() =>
2071 {
2072 // Retrying won't help for Payment Required or Model Request Limit errors (where
2073 // the user must upgrade to usage-based billing to get more requests, or else wait
2074 // for a significant amount of time for the request limit to reset).
2075 None
2076 }
2077 // Conservatively assume that any other errors are non-retryable
2078 HttpResponseError { .. } | Other(..) => Some(RetryStrategy::Fixed {
2079 delay: BASE_RETRY_DELAY,
2080 max_attempts: 2,
2081 }),
2082 }
2083 }
2084}
2085
2086struct RunningTurn {
2087 /// Holds the task that handles agent interaction until the end of the turn.
2088 /// Survives across multiple requests as the model performs tool calls and
2089 /// we run tools, report their results.
2090 _task: Task<()>,
2091 /// The current event stream for the running turn. Used to report a final
2092 /// cancellation event if we cancel the turn.
2093 event_stream: ThreadEventStream,
2094 /// The tools that were enabled for this turn.
2095 tools: BTreeMap<SharedString, Arc<dyn AnyAgentTool>>,
2096}
2097
2098impl RunningTurn {
2099 fn cancel(self) {
2100 log::debug!("Cancelling in progress turn");
2101 self.event_stream.send_canceled();
2102 }
2103}
2104
2105pub struct TokenUsageUpdated(pub Option<acp_thread::TokenUsage>);
2106
2107impl EventEmitter<TokenUsageUpdated> for Thread {}
2108
2109pub struct TitleUpdated;
2110
2111impl EventEmitter<TitleUpdated> for Thread {}
2112
2113pub trait AgentTool
2114where
2115 Self: 'static + Sized,
2116{
2117 type Input: for<'de> Deserialize<'de> + Serialize + JsonSchema;
2118 type Output: for<'de> Deserialize<'de> + Serialize + Into<LanguageModelToolResultContent>;
2119
2120 fn name() -> &'static str;
2121
2122 fn description() -> SharedString {
2123 let schema = schemars::schema_for!(Self::Input);
2124 SharedString::new(
2125 schema
2126 .get("description")
2127 .and_then(|description| description.as_str())
2128 .unwrap_or_default(),
2129 )
2130 }
2131
2132 fn kind() -> acp::ToolKind;
2133
2134 /// The initial tool title to display. Can be updated during the tool run.
2135 fn initial_title(
2136 &self,
2137 input: Result<Self::Input, serde_json::Value>,
2138 cx: &mut App,
2139 ) -> SharedString;
2140
2141 /// Returns the JSON schema that describes the tool's input.
2142 fn input_schema(format: LanguageModelToolSchemaFormat) -> Schema {
2143 language_model::tool_schema::root_schema_for::<Self::Input>(format)
2144 }
2145
2146 /// Some tools rely on a provider for the underlying billing or other reasons.
2147 /// Allow the tool to check if they are compatible, or should be filtered out.
2148 fn supports_provider(_provider: &LanguageModelProviderId) -> bool {
2149 true
2150 }
2151
2152 /// Runs the tool with the provided input.
2153 fn run(
2154 self: Arc<Self>,
2155 input: Self::Input,
2156 event_stream: ToolCallEventStream,
2157 cx: &mut App,
2158 ) -> Task<Result<Self::Output>>;
2159
2160 /// Emits events for a previous execution of the tool.
2161 fn replay(
2162 &self,
2163 _input: Self::Input,
2164 _output: Self::Output,
2165 _event_stream: ToolCallEventStream,
2166 _cx: &mut App,
2167 ) -> Result<()> {
2168 Ok(())
2169 }
2170
2171 fn erase(self) -> Arc<dyn AnyAgentTool> {
2172 Arc::new(Erased(Arc::new(self)))
2173 }
2174}
2175
2176pub struct Erased<T>(T);
2177
2178pub struct AgentToolOutput {
2179 pub llm_output: LanguageModelToolResultContent,
2180 pub raw_output: serde_json::Value,
2181}
2182
2183pub trait AnyAgentTool {
2184 fn name(&self) -> SharedString;
2185 fn description(&self) -> SharedString;
2186 fn kind(&self) -> acp::ToolKind;
2187 fn initial_title(&self, input: serde_json::Value, _cx: &mut App) -> SharedString;
2188 fn input_schema(&self, format: LanguageModelToolSchemaFormat) -> Result<serde_json::Value>;
2189 fn supports_provider(&self, _provider: &LanguageModelProviderId) -> bool {
2190 true
2191 }
2192 fn run(
2193 self: Arc<Self>,
2194 input: serde_json::Value,
2195 event_stream: ToolCallEventStream,
2196 cx: &mut App,
2197 ) -> Task<Result<AgentToolOutput>>;
2198 fn replay(
2199 &self,
2200 input: serde_json::Value,
2201 output: serde_json::Value,
2202 event_stream: ToolCallEventStream,
2203 cx: &mut App,
2204 ) -> Result<()>;
2205}
2206
2207impl<T> AnyAgentTool for Erased<Arc<T>>
2208where
2209 T: AgentTool,
2210{
2211 fn name(&self) -> SharedString {
2212 T::name().into()
2213 }
2214
2215 fn description(&self) -> SharedString {
2216 T::description()
2217 }
2218
2219 fn kind(&self) -> agent_client_protocol::ToolKind {
2220 T::kind()
2221 }
2222
2223 fn initial_title(&self, input: serde_json::Value, _cx: &mut App) -> SharedString {
2224 let parsed_input = serde_json::from_value(input.clone()).map_err(|_| input);
2225 self.0.initial_title(parsed_input, _cx)
2226 }
2227
2228 fn input_schema(&self, format: LanguageModelToolSchemaFormat) -> Result<serde_json::Value> {
2229 let mut json = serde_json::to_value(T::input_schema(format))?;
2230 language_model::tool_schema::adapt_schema_to_format(&mut json, format)?;
2231 Ok(json)
2232 }
2233
2234 fn supports_provider(&self, provider: &LanguageModelProviderId) -> bool {
2235 T::supports_provider(provider)
2236 }
2237
2238 fn run(
2239 self: Arc<Self>,
2240 input: serde_json::Value,
2241 event_stream: ToolCallEventStream,
2242 cx: &mut App,
2243 ) -> Task<Result<AgentToolOutput>> {
2244 cx.spawn(async move |cx| {
2245 let input = serde_json::from_value(input)?;
2246 let output = cx
2247 .update(|cx| self.0.clone().run(input, event_stream, cx))?
2248 .await?;
2249 let raw_output = serde_json::to_value(&output)?;
2250 Ok(AgentToolOutput {
2251 llm_output: output.into(),
2252 raw_output,
2253 })
2254 })
2255 }
2256
2257 fn replay(
2258 &self,
2259 input: serde_json::Value,
2260 output: serde_json::Value,
2261 event_stream: ToolCallEventStream,
2262 cx: &mut App,
2263 ) -> Result<()> {
2264 let input = serde_json::from_value(input)?;
2265 let output = serde_json::from_value(output)?;
2266 self.0.replay(input, output, event_stream, cx)
2267 }
2268}
2269
2270#[derive(Clone)]
2271struct ThreadEventStream(mpsc::UnboundedSender<Result<ThreadEvent>>);
2272
2273impl ThreadEventStream {
2274 fn send_user_message(&self, message: &UserMessage) {
2275 self.0
2276 .unbounded_send(Ok(ThreadEvent::UserMessage(message.clone())))
2277 .ok();
2278 }
2279
2280 fn send_text(&self, text: &str) {
2281 self.0
2282 .unbounded_send(Ok(ThreadEvent::AgentText(text.to_string())))
2283 .ok();
2284 }
2285
2286 fn send_thinking(&self, text: &str) {
2287 self.0
2288 .unbounded_send(Ok(ThreadEvent::AgentThinking(text.to_string())))
2289 .ok();
2290 }
2291
2292 fn send_tool_call(
2293 &self,
2294 id: &LanguageModelToolUseId,
2295 tool_name: &str,
2296 title: SharedString,
2297 kind: acp::ToolKind,
2298 input: serde_json::Value,
2299 ) {
2300 self.0
2301 .unbounded_send(Ok(ThreadEvent::ToolCall(Self::initial_tool_call(
2302 id,
2303 tool_name,
2304 title.to_string(),
2305 kind,
2306 input,
2307 ))))
2308 .ok();
2309 }
2310
2311 fn initial_tool_call(
2312 id: &LanguageModelToolUseId,
2313 tool_name: &str,
2314 title: String,
2315 kind: acp::ToolKind,
2316 input: serde_json::Value,
2317 ) -> acp::ToolCall {
2318 acp::ToolCall {
2319 meta: Some(serde_json::json!({
2320 "tool_name": tool_name
2321 })),
2322 id: acp::ToolCallId(id.to_string().into()),
2323 title,
2324 kind,
2325 status: acp::ToolCallStatus::Pending,
2326 content: vec![],
2327 locations: vec![],
2328 raw_input: Some(input),
2329 raw_output: None,
2330 }
2331 }
2332
2333 fn update_tool_call_fields(
2334 &self,
2335 tool_use_id: &LanguageModelToolUseId,
2336 fields: acp::ToolCallUpdateFields,
2337 ) {
2338 self.0
2339 .unbounded_send(Ok(ThreadEvent::ToolCallUpdate(
2340 acp::ToolCallUpdate {
2341 meta: None,
2342 id: acp::ToolCallId(tool_use_id.to_string().into()),
2343 fields,
2344 }
2345 .into(),
2346 )))
2347 .ok();
2348 }
2349
2350 fn send_retry(&self, status: acp_thread::RetryStatus) {
2351 self.0.unbounded_send(Ok(ThreadEvent::Retry(status))).ok();
2352 }
2353
2354 fn send_stop(&self, reason: acp::StopReason) {
2355 self.0.unbounded_send(Ok(ThreadEvent::Stop(reason))).ok();
2356 }
2357
2358 fn send_canceled(&self) {
2359 self.0
2360 .unbounded_send(Ok(ThreadEvent::Stop(acp::StopReason::Cancelled)))
2361 .ok();
2362 }
2363
2364 fn send_error(&self, error: impl Into<anyhow::Error>) {
2365 self.0.unbounded_send(Err(error.into())).ok();
2366 }
2367}
2368
2369#[derive(Clone)]
2370pub struct ToolCallEventStream {
2371 tool_use_id: LanguageModelToolUseId,
2372 stream: ThreadEventStream,
2373 fs: Option<Arc<dyn Fs>>,
2374}
2375
2376impl ToolCallEventStream {
2377 #[cfg(any(test, feature = "test-support"))]
2378 pub fn test() -> (Self, ToolCallEventStreamReceiver) {
2379 let (events_tx, events_rx) = mpsc::unbounded::<Result<ThreadEvent>>();
2380
2381 let stream = ToolCallEventStream::new("test_id".into(), ThreadEventStream(events_tx), None);
2382
2383 (stream, ToolCallEventStreamReceiver(events_rx))
2384 }
2385
2386 fn new(
2387 tool_use_id: LanguageModelToolUseId,
2388 stream: ThreadEventStream,
2389 fs: Option<Arc<dyn Fs>>,
2390 ) -> Self {
2391 Self {
2392 tool_use_id,
2393 stream,
2394 fs,
2395 }
2396 }
2397
2398 pub fn update_fields(&self, fields: acp::ToolCallUpdateFields) {
2399 self.stream
2400 .update_tool_call_fields(&self.tool_use_id, fields);
2401 }
2402
2403 pub fn update_diff(&self, diff: Entity<acp_thread::Diff>) {
2404 self.stream
2405 .0
2406 .unbounded_send(Ok(ThreadEvent::ToolCallUpdate(
2407 acp_thread::ToolCallUpdateDiff {
2408 id: acp::ToolCallId(self.tool_use_id.to_string().into()),
2409 diff,
2410 }
2411 .into(),
2412 )))
2413 .ok();
2414 }
2415
2416 pub fn authorize(&self, title: impl Into<String>, cx: &mut App) -> Task<Result<()>> {
2417 if agent_settings::AgentSettings::get_global(cx).always_allow_tool_actions {
2418 return Task::ready(Ok(()));
2419 }
2420
2421 let (response_tx, response_rx) = oneshot::channel();
2422 self.stream
2423 .0
2424 .unbounded_send(Ok(ThreadEvent::ToolCallAuthorization(
2425 ToolCallAuthorization {
2426 tool_call: acp::ToolCallUpdate {
2427 meta: None,
2428 id: acp::ToolCallId(self.tool_use_id.to_string().into()),
2429 fields: acp::ToolCallUpdateFields {
2430 title: Some(title.into()),
2431 ..Default::default()
2432 },
2433 },
2434 options: vec![
2435 acp::PermissionOption {
2436 id: acp::PermissionOptionId("always_allow".into()),
2437 name: "Always Allow".into(),
2438 kind: acp::PermissionOptionKind::AllowAlways,
2439 meta: None,
2440 },
2441 acp::PermissionOption {
2442 id: acp::PermissionOptionId("allow".into()),
2443 name: "Allow".into(),
2444 kind: acp::PermissionOptionKind::AllowOnce,
2445 meta: None,
2446 },
2447 acp::PermissionOption {
2448 id: acp::PermissionOptionId("deny".into()),
2449 name: "Deny".into(),
2450 kind: acp::PermissionOptionKind::RejectOnce,
2451 meta: None,
2452 },
2453 ],
2454 response: response_tx,
2455 },
2456 )))
2457 .ok();
2458 let fs = self.fs.clone();
2459 cx.spawn(async move |cx| match response_rx.await?.0.as_ref() {
2460 "always_allow" => {
2461 if let Some(fs) = fs.clone() {
2462 cx.update(|cx| {
2463 update_settings_file(fs, cx, |settings, _| {
2464 settings
2465 .agent
2466 .get_or_insert_default()
2467 .set_always_allow_tool_actions(true);
2468 });
2469 })?;
2470 }
2471
2472 Ok(())
2473 }
2474 "allow" => Ok(()),
2475 _ => Err(anyhow!("Permission to run tool denied by user")),
2476 })
2477 }
2478}
2479
2480#[cfg(any(test, feature = "test-support"))]
2481pub struct ToolCallEventStreamReceiver(mpsc::UnboundedReceiver<Result<ThreadEvent>>);
2482
2483#[cfg(any(test, feature = "test-support"))]
2484impl ToolCallEventStreamReceiver {
2485 pub async fn expect_authorization(&mut self) -> ToolCallAuthorization {
2486 let event = self.0.next().await;
2487 if let Some(Ok(ThreadEvent::ToolCallAuthorization(auth))) = event {
2488 auth
2489 } else {
2490 panic!("Expected ToolCallAuthorization but got: {:?}", event);
2491 }
2492 }
2493
2494 pub async fn expect_update_fields(&mut self) -> acp::ToolCallUpdateFields {
2495 let event = self.0.next().await;
2496 if let Some(Ok(ThreadEvent::ToolCallUpdate(acp_thread::ToolCallUpdate::UpdateFields(
2497 update,
2498 )))) = event
2499 {
2500 update.fields
2501 } else {
2502 panic!("Expected update fields but got: {:?}", event);
2503 }
2504 }
2505
2506 pub async fn expect_diff(&mut self) -> Entity<acp_thread::Diff> {
2507 let event = self.0.next().await;
2508 if let Some(Ok(ThreadEvent::ToolCallUpdate(acp_thread::ToolCallUpdate::UpdateDiff(
2509 update,
2510 )))) = event
2511 {
2512 update.diff
2513 } else {
2514 panic!("Expected diff but got: {:?}", event);
2515 }
2516 }
2517
2518 pub async fn expect_terminal(&mut self) -> Entity<acp_thread::Terminal> {
2519 let event = self.0.next().await;
2520 if let Some(Ok(ThreadEvent::ToolCallUpdate(acp_thread::ToolCallUpdate::UpdateTerminal(
2521 update,
2522 )))) = event
2523 {
2524 update.terminal
2525 } else {
2526 panic!("Expected terminal but got: {:?}", event);
2527 }
2528 }
2529}
2530
2531#[cfg(any(test, feature = "test-support"))]
2532impl std::ops::Deref for ToolCallEventStreamReceiver {
2533 type Target = mpsc::UnboundedReceiver<Result<ThreadEvent>>;
2534
2535 fn deref(&self) -> &Self::Target {
2536 &self.0
2537 }
2538}
2539
2540#[cfg(any(test, feature = "test-support"))]
2541impl std::ops::DerefMut for ToolCallEventStreamReceiver {
2542 fn deref_mut(&mut self) -> &mut Self::Target {
2543 &mut self.0
2544 }
2545}
2546
2547impl From<&str> for UserMessageContent {
2548 fn from(text: &str) -> Self {
2549 Self::Text(text.into())
2550 }
2551}
2552
2553impl UserMessageContent {
2554 pub fn from_content_block(value: acp::ContentBlock, path_style: PathStyle) -> Self {
2555 match value {
2556 acp::ContentBlock::Text(text_content) => Self::Text(text_content.text),
2557 acp::ContentBlock::Image(image_content) => Self::Image(convert_image(image_content)),
2558 acp::ContentBlock::Audio(_) => {
2559 // TODO
2560 Self::Text("[audio]".to_string())
2561 }
2562 acp::ContentBlock::ResourceLink(resource_link) => {
2563 match MentionUri::parse(&resource_link.uri, path_style) {
2564 Ok(uri) => Self::Mention {
2565 uri,
2566 content: String::new(),
2567 },
2568 Err(err) => {
2569 log::error!("Failed to parse mention link: {}", err);
2570 Self::Text(format!("[{}]({})", resource_link.name, resource_link.uri))
2571 }
2572 }
2573 }
2574 acp::ContentBlock::Resource(resource) => match resource.resource {
2575 acp::EmbeddedResourceResource::TextResourceContents(resource) => {
2576 match MentionUri::parse(&resource.uri, path_style) {
2577 Ok(uri) => Self::Mention {
2578 uri,
2579 content: resource.text,
2580 },
2581 Err(err) => {
2582 log::error!("Failed to parse mention link: {}", err);
2583 Self::Text(
2584 MarkdownCodeBlock {
2585 tag: &resource.uri,
2586 text: &resource.text,
2587 }
2588 .to_string(),
2589 )
2590 }
2591 }
2592 }
2593 acp::EmbeddedResourceResource::BlobResourceContents(_) => {
2594 // TODO
2595 Self::Text("[blob]".to_string())
2596 }
2597 },
2598 }
2599 }
2600}
2601
2602impl From<UserMessageContent> for acp::ContentBlock {
2603 fn from(content: UserMessageContent) -> Self {
2604 match content {
2605 UserMessageContent::Text(text) => acp::ContentBlock::Text(acp::TextContent {
2606 text,
2607 annotations: None,
2608 meta: None,
2609 }),
2610 UserMessageContent::Image(image) => acp::ContentBlock::Image(acp::ImageContent {
2611 data: image.source.to_string(),
2612 mime_type: "image/png".to_string(),
2613 meta: None,
2614 annotations: None,
2615 uri: None,
2616 }),
2617 UserMessageContent::Mention { uri, content } => {
2618 acp::ContentBlock::Resource(acp::EmbeddedResource {
2619 meta: None,
2620 resource: acp::EmbeddedResourceResource::TextResourceContents(
2621 acp::TextResourceContents {
2622 meta: None,
2623 mime_type: None,
2624 text: content,
2625 uri: uri.to_uri().to_string(),
2626 },
2627 ),
2628 annotations: None,
2629 })
2630 }
2631 }
2632 }
2633}
2634
2635fn convert_image(image_content: acp::ImageContent) -> LanguageModelImage {
2636 LanguageModelImage {
2637 source: image_content.data.into(),
2638 // TODO: make this optional?
2639 size: gpui::Size::new(0.into(), 0.into()),
2640 }
2641}