ZIm/crates/agent2/src/thread.rs
2025-08-18 21:54:35 +00:00

1608 lines
54 KiB
Rust

use crate::{ContextServerRegistry, SystemPromptTemplate, Template, Templates};
use acp_thread::{MentionUri, UserMessageId};
use action_log::ActionLog;
use agent_client_protocol as acp;
use agent_settings::{AgentProfileId, AgentSettings, CompletionMode};
use anyhow::{Context as _, Result, anyhow};
use assistant_tool::adapt_schema_to_format;
use cloud_llm_client::{CompletionIntent, CompletionRequestStatus};
use collections::IndexMap;
use fs::Fs;
use futures::{
channel::{mpsc, oneshot},
stream::FuturesUnordered,
};
use gpui::{App, Context, Entity, SharedString, Task};
use language_model::{
LanguageModel, LanguageModelCompletionEvent, LanguageModelImage, LanguageModelProviderId,
LanguageModelRequest, LanguageModelRequestMessage, LanguageModelRequestTool,
LanguageModelToolResult, LanguageModelToolResultContent, LanguageModelToolSchemaFormat,
LanguageModelToolUse, LanguageModelToolUseId, Role, StopReason,
};
use project::Project;
use prompt_store::ProjectContext;
use schemars::{JsonSchema, Schema};
use serde::{Deserialize, Serialize};
use settings::{Settings, update_settings_file};
use smol::stream::StreamExt;
use std::{cell::RefCell, collections::BTreeMap, path::Path, rc::Rc, sync::Arc};
use std::{fmt::Write, ops::Range};
use util::{ResultExt, markdown::MarkdownCodeBlock};
use uuid::Uuid;
#[derive(
Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Clone, Serialize, Deserialize, JsonSchema,
)]
pub struct ThreadId(Arc<str>);
impl ThreadId {
pub fn new() -> Self {
Self(Uuid::new_v4().to_string().into())
}
}
impl std::fmt::Display for ThreadId {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
write!(f, "{}", self.0)
}
}
impl From<&str> for ThreadId {
fn from(value: &str) -> Self {
Self(value.into())
}
}
/// The ID of the user prompt that initiated a request.
///
/// This equates to the user physically submitting a message to the model (e.g., by pressing the Enter key).
#[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Clone, Serialize, Deserialize)]
pub struct PromptId(Arc<str>);
impl PromptId {
pub fn new() -> Self {
Self(Uuid::new_v4().to_string().into())
}
}
impl std::fmt::Display for PromptId {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
write!(f, "{}", self.0)
}
}
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum Message {
User(UserMessage),
Agent(AgentMessage),
Resume,
}
impl Message {
pub fn as_agent_message(&self) -> Option<&AgentMessage> {
match self {
Message::Agent(agent_message) => Some(agent_message),
_ => None,
}
}
pub fn to_markdown(&self) -> String {
match self {
Message::User(message) => message.to_markdown(),
Message::Agent(message) => message.to_markdown(),
Message::Resume => "[resumed after tool use limit was reached]".into(),
}
}
}
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct UserMessage {
pub id: UserMessageId,
pub content: Vec<UserMessageContent>,
}
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum UserMessageContent {
Text(String),
Mention { uri: MentionUri, content: String },
Image(LanguageModelImage),
}
impl UserMessage {
pub fn to_markdown(&self) -> String {
let mut markdown = String::from("## User\n\n");
for content in &self.content {
match content {
UserMessageContent::Text(text) => {
markdown.push_str(text);
markdown.push('\n');
}
UserMessageContent::Image(_) => {
markdown.push_str("<image />\n");
}
UserMessageContent::Mention { uri, content } => {
if !content.is_empty() {
let _ = write!(&mut markdown, "{}\n\n{}\n", uri.as_link(), content);
} else {
let _ = write!(&mut markdown, "{}\n", uri.as_link());
}
}
}
}
markdown
}
fn to_request(&self) -> LanguageModelRequestMessage {
let mut message = LanguageModelRequestMessage {
role: Role::User,
content: Vec::with_capacity(self.content.len()),
cache: false,
};
const OPEN_CONTEXT: &str = "<context>\n\
The following items were attached by the user. \
They are up-to-date and don't need to be re-read.\n\n";
const OPEN_FILES_TAG: &str = "<files>";
const OPEN_SYMBOLS_TAG: &str = "<symbols>";
const OPEN_THREADS_TAG: &str = "<threads>";
const OPEN_FETCH_TAG: &str = "<fetched_urls>";
const OPEN_RULES_TAG: &str =
"<rules>\nThe user has specified the following rules that should be applied:\n";
let mut file_context = OPEN_FILES_TAG.to_string();
let mut symbol_context = OPEN_SYMBOLS_TAG.to_string();
let mut thread_context = OPEN_THREADS_TAG.to_string();
let mut fetch_context = OPEN_FETCH_TAG.to_string();
let mut rules_context = OPEN_RULES_TAG.to_string();
for chunk in &self.content {
let chunk = match chunk {
UserMessageContent::Text(text) => {
language_model::MessageContent::Text(text.clone())
}
UserMessageContent::Image(value) => {
language_model::MessageContent::Image(value.clone())
}
UserMessageContent::Mention { uri, content } => {
match uri {
MentionUri::File { abs_path, .. } => {
write!(
&mut symbol_context,
"\n{}",
MarkdownCodeBlock {
tag: &codeblock_tag(abs_path, None),
text: &content.to_string(),
}
)
.ok();
}
MentionUri::Symbol {
path, line_range, ..
}
| MentionUri::Selection {
path, line_range, ..
} => {
write!(
&mut rules_context,
"\n{}",
MarkdownCodeBlock {
tag: &codeblock_tag(path, Some(line_range)),
text: content
}
)
.ok();
}
MentionUri::Thread { .. } => {
write!(&mut thread_context, "\n{}\n", content).ok();
}
MentionUri::TextThread { .. } => {
write!(&mut thread_context, "\n{}\n", content).ok();
}
MentionUri::Rule { .. } => {
write!(
&mut rules_context,
"\n{}",
MarkdownCodeBlock {
tag: "",
text: content
}
)
.ok();
}
MentionUri::Fetch { url } => {
write!(&mut fetch_context, "\nFetch: {}\n\n{}", url, content).ok();
}
}
language_model::MessageContent::Text(uri.as_link().to_string())
}
};
message.content.push(chunk);
}
let len_before_context = message.content.len();
if file_context.len() > OPEN_FILES_TAG.len() {
file_context.push_str("</files>\n");
message
.content
.push(language_model::MessageContent::Text(file_context));
}
if symbol_context.len() > OPEN_SYMBOLS_TAG.len() {
symbol_context.push_str("</symbols>\n");
message
.content
.push(language_model::MessageContent::Text(symbol_context));
}
if thread_context.len() > OPEN_THREADS_TAG.len() {
thread_context.push_str("</threads>\n");
message
.content
.push(language_model::MessageContent::Text(thread_context));
}
if fetch_context.len() > OPEN_FETCH_TAG.len() {
fetch_context.push_str("</fetched_urls>\n");
message
.content
.push(language_model::MessageContent::Text(fetch_context));
}
if rules_context.len() > OPEN_RULES_TAG.len() {
rules_context.push_str("</user_rules>\n");
message
.content
.push(language_model::MessageContent::Text(rules_context));
}
if message.content.len() > len_before_context {
message.content.insert(
len_before_context,
language_model::MessageContent::Text(OPEN_CONTEXT.into()),
);
message
.content
.push(language_model::MessageContent::Text("</context>".into()));
}
message
}
}
fn codeblock_tag(full_path: &Path, line_range: Option<&Range<u32>>) -> String {
let mut result = String::new();
if let Some(extension) = full_path.extension().and_then(|ext| ext.to_str()) {
let _ = write!(result, "{} ", extension);
}
let _ = write!(result, "{}", full_path.display());
if let Some(range) = line_range {
if range.start == range.end {
let _ = write!(result, ":{}", range.start + 1);
} else {
let _ = write!(result, ":{}-{}", range.start + 1, range.end + 1);
}
}
result
}
impl AgentMessage {
pub fn to_markdown(&self) -> String {
let mut markdown = String::from("## Assistant\n\n");
for content in &self.content {
match content {
AgentMessageContent::Text(text) => {
markdown.push_str(text);
markdown.push('\n');
}
AgentMessageContent::Thinking { text, .. } => {
markdown.push_str("<think>");
markdown.push_str(text);
markdown.push_str("</think>\n");
}
AgentMessageContent::RedactedThinking(_) => {
markdown.push_str("<redacted_thinking />\n")
}
AgentMessageContent::Image(_) => {
markdown.push_str("<image />\n");
}
AgentMessageContent::ToolUse(tool_use) => {
markdown.push_str(&format!(
"**Tool Use**: {} (ID: {})\n",
tool_use.name, tool_use.id
));
markdown.push_str(&format!(
"{}\n",
MarkdownCodeBlock {
tag: "json",
text: &format!("{:#}", tool_use.input)
}
));
}
}
}
for tool_result in self.tool_results.values() {
markdown.push_str(&format!(
"**Tool Result**: {} (ID: {})\n\n",
tool_result.tool_name, tool_result.tool_use_id
));
if tool_result.is_error {
markdown.push_str("**ERROR:**\n");
}
match &tool_result.content {
LanguageModelToolResultContent::Text(text) => {
writeln!(markdown, "{text}\n").ok();
}
LanguageModelToolResultContent::Image(_) => {
writeln!(markdown, "<image />\n").ok();
}
}
if let Some(output) = tool_result.output.as_ref() {
writeln!(
markdown,
"**Debug Output**:\n\n```json\n{}\n```\n",
serde_json::to_string_pretty(output).unwrap()
)
.unwrap();
}
}
markdown
}
pub fn to_request(&self) -> Vec<LanguageModelRequestMessage> {
let mut assistant_message = LanguageModelRequestMessage {
role: Role::Assistant,
content: Vec::with_capacity(self.content.len()),
cache: false,
};
for chunk in &self.content {
let chunk = match chunk {
AgentMessageContent::Text(text) => {
language_model::MessageContent::Text(text.clone())
}
AgentMessageContent::Thinking { text, signature } => {
language_model::MessageContent::Thinking {
text: text.clone(),
signature: signature.clone(),
}
}
AgentMessageContent::RedactedThinking(value) => {
language_model::MessageContent::RedactedThinking(value.clone())
}
AgentMessageContent::ToolUse(value) => {
language_model::MessageContent::ToolUse(value.clone())
}
AgentMessageContent::Image(value) => {
language_model::MessageContent::Image(value.clone())
}
};
assistant_message.content.push(chunk);
}
let mut user_message = LanguageModelRequestMessage {
role: Role::User,
content: Vec::new(),
cache: false,
};
for tool_result in self.tool_results.values() {
user_message
.content
.push(language_model::MessageContent::ToolResult(
tool_result.clone(),
));
}
let mut messages = Vec::new();
if !assistant_message.content.is_empty() {
messages.push(assistant_message);
}
if !user_message.content.is_empty() {
messages.push(user_message);
}
messages
}
}
#[derive(Default, Debug, Clone, PartialEq, Eq)]
pub struct AgentMessage {
pub content: Vec<AgentMessageContent>,
pub tool_results: IndexMap<LanguageModelToolUseId, LanguageModelToolResult>,
}
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum AgentMessageContent {
Text(String),
Thinking {
text: String,
signature: Option<String>,
},
RedactedThinking(String),
Image(LanguageModelImage),
ToolUse(LanguageModelToolUse),
}
#[derive(Debug)]
pub enum AgentResponseEvent {
Text(String),
Thinking(String),
ToolCall(acp::ToolCall),
ToolCallUpdate(acp_thread::ToolCallUpdate),
ToolCallAuthorization(ToolCallAuthorization),
Stop(acp::StopReason),
}
#[derive(Debug)]
pub struct ToolCallAuthorization {
pub tool_call: acp::ToolCallUpdate,
pub options: Vec<acp::PermissionOption>,
pub response: oneshot::Sender<acp::PermissionOptionId>,
}
pub struct Thread {
id: ThreadId,
prompt_id: PromptId,
messages: Vec<Message>,
completion_mode: CompletionMode,
/// Holds the task that handles agent interaction until the end of the turn.
/// Survives across multiple requests as the model performs tool calls and
/// we run tools, report their results.
running_turn: Option<RunningTurn>,
pending_message: Option<AgentMessage>,
tools: BTreeMap<SharedString, Arc<dyn AnyAgentTool>>,
tool_use_limit_reached: bool,
context_server_registry: Entity<ContextServerRegistry>,
profile_id: AgentProfileId,
project_context: Rc<RefCell<ProjectContext>>,
templates: Arc<Templates>,
model: Option<Arc<dyn LanguageModel>>,
project: Entity<Project>,
action_log: Entity<ActionLog>,
}
impl Thread {
pub fn new(
project: Entity<Project>,
project_context: Rc<RefCell<ProjectContext>>,
context_server_registry: Entity<ContextServerRegistry>,
action_log: Entity<ActionLog>,
templates: Arc<Templates>,
model: Option<Arc<dyn LanguageModel>>,
cx: &mut Context<Self>,
) -> Self {
let profile_id = AgentSettings::get_global(cx).default_profile.clone();
Self {
id: ThreadId::new(),
prompt_id: PromptId::new(),
messages: Vec::new(),
completion_mode: CompletionMode::Normal,
running_turn: None,
pending_message: None,
tools: BTreeMap::default(),
tool_use_limit_reached: false,
context_server_registry,
profile_id,
project_context,
templates,
model,
project,
action_log,
}
}
pub fn project(&self) -> &Entity<Project> {
&self.project
}
pub fn action_log(&self) -> &Entity<ActionLog> {
&self.action_log
}
pub fn model(&self) -> Option<&Arc<dyn LanguageModel>> {
self.model.as_ref()
}
pub fn set_model(&mut self, model: Arc<dyn LanguageModel>) {
self.model = Some(model);
}
pub fn completion_mode(&self) -> CompletionMode {
self.completion_mode
}
pub fn set_completion_mode(&mut self, mode: CompletionMode) {
self.completion_mode = mode;
}
#[cfg(any(test, feature = "test-support"))]
pub fn last_message(&self) -> Option<Message> {
if let Some(message) = self.pending_message.clone() {
Some(Message::Agent(message))
} else {
self.messages.last().cloned()
}
}
pub fn add_tool(&mut self, tool: impl AgentTool) {
self.tools.insert(tool.name(), tool.erase());
}
pub fn remove_tool(&mut self, name: &str) -> bool {
self.tools.remove(name).is_some()
}
pub fn profile(&self) -> &AgentProfileId {
&self.profile_id
}
pub fn set_profile(&mut self, profile_id: AgentProfileId) {
self.profile_id = profile_id;
}
pub fn cancel(&mut self) {
if let Some(running_turn) = self.running_turn.take() {
running_turn.cancel();
}
self.flush_pending_message();
}
pub fn truncate(&mut self, message_id: UserMessageId) -> Result<()> {
self.cancel();
let Some(position) = self.messages.iter().position(
|msg| matches!(msg, Message::User(UserMessage { id, .. }) if id == &message_id),
) else {
return Err(anyhow!("Message not found"));
};
self.messages.truncate(position);
Ok(())
}
pub fn resume(
&mut self,
cx: &mut Context<Self>,
) -> Result<mpsc::UnboundedReceiver<Result<AgentResponseEvent>>> {
anyhow::ensure!(self.model.is_some(), "Model not set");
anyhow::ensure!(
self.tool_use_limit_reached,
"can only resume after tool use limit is reached"
);
self.messages.push(Message::Resume);
cx.notify();
log::info!("Total messages in thread: {}", self.messages.len());
self.run_turn(cx)
}
/// Sending a message results in the model streaming a response, which could include tool calls.
/// After calling tools, the model will stops and waits for any outstanding tool calls to be completed and their results sent.
/// The returned channel will report all the occurrences in which the model stops before erroring or ending its turn.
pub fn send<T>(
&mut self,
id: UserMessageId,
content: impl IntoIterator<Item = T>,
cx: &mut Context<Self>,
) -> Result<mpsc::UnboundedReceiver<Result<AgentResponseEvent>>>
where
T: Into<UserMessageContent>,
{
let model = self.model().context("No language model configured")?;
log::info!("Thread::send called with model: {:?}", model.name());
self.advance_prompt_id();
let content = content.into_iter().map(Into::into).collect::<Vec<_>>();
log::debug!("Thread::send content: {:?}", content);
self.messages
.push(Message::User(UserMessage { id, content }));
cx.notify();
log::info!("Total messages in thread: {}", self.messages.len());
self.run_turn(cx)
}
fn run_turn(
&mut self,
cx: &mut Context<Self>,
) -> Result<mpsc::UnboundedReceiver<Result<AgentResponseEvent>>> {
self.cancel();
let model = self
.model()
.cloned()
.context("No language model configured")?;
let (events_tx, events_rx) = mpsc::unbounded::<Result<AgentResponseEvent>>();
let event_stream = AgentResponseEventStream(events_tx);
let message_ix = self.messages.len().saturating_sub(1);
self.tool_use_limit_reached = false;
self.running_turn = Some(RunningTurn {
event_stream: event_stream.clone(),
_task: cx.spawn(async move |this, cx| {
log::info!("Starting agent turn execution");
let turn_result: Result<()> = async {
let mut completion_intent = CompletionIntent::UserPrompt;
loop {
log::debug!(
"Building completion request with intent: {:?}",
completion_intent
);
let request = this.update(cx, |this, cx| {
this.build_completion_request(completion_intent, cx)
})??;
log::info!("Calling model.stream_completion");
let mut events = model.stream_completion(request, cx).await?;
log::debug!("Stream completion started successfully");
let mut tool_use_limit_reached = false;
let mut tool_uses = FuturesUnordered::new();
while let Some(event) = events.next().await {
match event? {
LanguageModelCompletionEvent::StatusUpdate(
CompletionRequestStatus::ToolUseLimitReached,
) => {
tool_use_limit_reached = true;
}
LanguageModelCompletionEvent::Stop(reason) => {
event_stream.send_stop(reason);
if reason == StopReason::Refusal {
this.update(cx, |this, _cx| {
this.flush_pending_message();
this.messages.truncate(message_ix);
})?;
return Ok(());
}
}
event => {
log::trace!("Received completion event: {:?}", event);
this.update(cx, |this, cx| {
tool_uses.extend(this.handle_streamed_completion_event(
event,
&event_stream,
cx,
));
})
.ok();
}
}
}
let used_tools = tool_uses.is_empty();
while let Some(tool_result) = tool_uses.next().await {
log::info!("Tool finished {:?}", tool_result);
event_stream.update_tool_call_fields(
&tool_result.tool_use_id,
acp::ToolCallUpdateFields {
status: Some(if tool_result.is_error {
acp::ToolCallStatus::Failed
} else {
acp::ToolCallStatus::Completed
}),
raw_output: tool_result.output.clone(),
..Default::default()
},
);
this.update(cx, |this, _cx| {
this.pending_message()
.tool_results
.insert(tool_result.tool_use_id.clone(), tool_result);
})
.ok();
}
if tool_use_limit_reached {
log::info!("Tool use limit reached, completing turn");
this.update(cx, |this, _cx| this.tool_use_limit_reached = true)?;
return Err(language_model::ToolUseLimitReachedError.into());
} else if used_tools {
log::info!("No tool uses found, completing turn");
return Ok(());
} else {
this.update(cx, |this, _| this.flush_pending_message())?;
completion_intent = CompletionIntent::ToolResults;
}
}
}
.await;
if let Err(error) = turn_result {
log::error!("Turn execution failed: {:?}", error);
event_stream.send_error(error);
} else {
log::info!("Turn execution completed successfully");
}
this.update(cx, |this, _| {
this.flush_pending_message();
this.running_turn.take();
})
.ok();
}),
});
Ok(events_rx)
}
pub fn build_system_message(&self) -> LanguageModelRequestMessage {
log::debug!("Building system message");
let prompt = SystemPromptTemplate {
project: &self.project_context.borrow(),
available_tools: self.tools.keys().cloned().collect(),
}
.render(&self.templates)
.context("failed to build system prompt")
.expect("Invalid template");
log::debug!("System message built");
LanguageModelRequestMessage {
role: Role::System,
content: vec![prompt.into()],
cache: true,
}
}
/// A helper method that's called on every streamed completion event.
/// Returns an optional tool result task, which the main agentic loop in
/// send will send back to the model when it resolves.
fn handle_streamed_completion_event(
&mut self,
event: LanguageModelCompletionEvent,
event_stream: &AgentResponseEventStream,
cx: &mut Context<Self>,
) -> Option<Task<LanguageModelToolResult>> {
log::trace!("Handling streamed completion event: {:?}", event);
use LanguageModelCompletionEvent::*;
match event {
StartMessage { .. } => {
self.flush_pending_message();
self.pending_message = Some(AgentMessage::default());
}
Text(new_text) => self.handle_text_event(new_text, event_stream, cx),
Thinking { text, signature } => {
self.handle_thinking_event(text, signature, event_stream, cx)
}
RedactedThinking { data } => self.handle_redacted_thinking_event(data, cx),
ToolUse(tool_use) => {
return self.handle_tool_use_event(tool_use, event_stream, cx);
}
ToolUseJsonParseError {
id,
tool_name,
raw_input,
json_parse_error,
} => {
return Some(Task::ready(self.handle_tool_use_json_parse_error_event(
id,
tool_name,
raw_input,
json_parse_error,
)));
}
UsageUpdate(_) | StatusUpdate(_) => {}
Stop(_) => unreachable!(),
}
None
}
fn handle_text_event(
&mut self,
new_text: String,
event_stream: &AgentResponseEventStream,
cx: &mut Context<Self>,
) {
event_stream.send_text(&new_text);
let last_message = self.pending_message();
if let Some(AgentMessageContent::Text(text)) = last_message.content.last_mut() {
text.push_str(&new_text);
} else {
last_message
.content
.push(AgentMessageContent::Text(new_text));
}
cx.notify();
}
fn handle_thinking_event(
&mut self,
new_text: String,
new_signature: Option<String>,
event_stream: &AgentResponseEventStream,
cx: &mut Context<Self>,
) {
event_stream.send_thinking(&new_text);
let last_message = self.pending_message();
if let Some(AgentMessageContent::Thinking { text, signature }) =
last_message.content.last_mut()
{
text.push_str(&new_text);
*signature = new_signature.or(signature.take());
} else {
last_message.content.push(AgentMessageContent::Thinking {
text: new_text,
signature: new_signature,
});
}
cx.notify();
}
fn handle_redacted_thinking_event(&mut self, data: String, cx: &mut Context<Self>) {
let last_message = self.pending_message();
last_message
.content
.push(AgentMessageContent::RedactedThinking(data));
cx.notify();
}
fn handle_tool_use_event(
&mut self,
tool_use: LanguageModelToolUse,
event_stream: &AgentResponseEventStream,
cx: &mut Context<Self>,
) -> Option<Task<LanguageModelToolResult>> {
cx.notify();
let tool = self.tools.get(tool_use.name.as_ref()).cloned();
let mut title = SharedString::from(&tool_use.name);
let mut kind = acp::ToolKind::Other;
if let Some(tool) = tool.as_ref() {
title = tool.initial_title(tool_use.input.clone());
kind = tool.kind();
}
// Ensure the last message ends in the current tool use
let last_message = self.pending_message();
let push_new_tool_use = last_message.content.last_mut().map_or(true, |content| {
if let AgentMessageContent::ToolUse(last_tool_use) = content {
if last_tool_use.id == tool_use.id {
*last_tool_use = tool_use.clone();
false
} else {
true
}
} else {
true
}
});
if push_new_tool_use {
event_stream.send_tool_call(&tool_use.id, title, kind, tool_use.input.clone());
last_message
.content
.push(AgentMessageContent::ToolUse(tool_use.clone()));
} else {
event_stream.update_tool_call_fields(
&tool_use.id,
acp::ToolCallUpdateFields {
title: Some(title.into()),
kind: Some(kind),
raw_input: Some(tool_use.input.clone()),
..Default::default()
},
);
}
if !tool_use.is_input_complete {
return None;
}
let Some(tool) = tool else {
let content = format!("No tool named {} exists", tool_use.name);
return Some(Task::ready(LanguageModelToolResult {
content: LanguageModelToolResultContent::Text(Arc::from(content)),
tool_use_id: tool_use.id,
tool_name: tool_use.name,
is_error: true,
output: None,
}));
};
let fs = self.project.read(cx).fs().clone();
let tool_event_stream =
ToolCallEventStream::new(tool_use.id.clone(), event_stream.clone(), Some(fs));
tool_event_stream.update_fields(acp::ToolCallUpdateFields {
status: Some(acp::ToolCallStatus::InProgress),
..Default::default()
});
let supports_images = self.model().map_or(false, |model| model.supports_images());
let tool_result = tool.run(tool_use.input, tool_event_stream, cx);
log::info!("Running tool {}", tool_use.name);
Some(cx.foreground_executor().spawn(async move {
let tool_result = tool_result.await.and_then(|output| {
if let LanguageModelToolResultContent::Image(_) = &output.llm_output {
if !supports_images {
return Err(anyhow!(
"Attempted to read an image, but this model doesn't support it.",
));
}
}
Ok(output)
});
match tool_result {
Ok(output) => LanguageModelToolResult {
tool_use_id: tool_use.id,
tool_name: tool_use.name,
is_error: false,
content: output.llm_output,
output: Some(output.raw_output),
},
Err(error) => LanguageModelToolResult {
tool_use_id: tool_use.id,
tool_name: tool_use.name,
is_error: true,
content: LanguageModelToolResultContent::Text(Arc::from(error.to_string())),
output: None,
},
}
}))
}
fn handle_tool_use_json_parse_error_event(
&mut self,
tool_use_id: LanguageModelToolUseId,
tool_name: Arc<str>,
raw_input: Arc<str>,
json_parse_error: String,
) -> LanguageModelToolResult {
let tool_output = format!("Error parsing input JSON: {json_parse_error}");
LanguageModelToolResult {
tool_use_id,
tool_name,
is_error: true,
content: LanguageModelToolResultContent::Text(tool_output.into()),
output: Some(serde_json::Value::String(raw_input.to_string())),
}
}
fn pending_message(&mut self) -> &mut AgentMessage {
self.pending_message.get_or_insert_default()
}
fn flush_pending_message(&mut self) {
let Some(mut message) = self.pending_message.take() else {
return;
};
for content in &message.content {
let AgentMessageContent::ToolUse(tool_use) = content else {
continue;
};
if !message.tool_results.contains_key(&tool_use.id) {
message.tool_results.insert(
tool_use.id.clone(),
LanguageModelToolResult {
tool_use_id: tool_use.id.clone(),
tool_name: tool_use.name.clone(),
is_error: true,
content: LanguageModelToolResultContent::Text(
"Tool canceled by user".into(),
),
output: None,
},
);
}
}
self.messages.push(Message::Agent(message));
}
pub(crate) fn build_completion_request(
&self,
completion_intent: CompletionIntent,
cx: &mut App,
) -> Result<LanguageModelRequest> {
let model = self.model().context("No language model configured")?;
log::debug!("Building completion request");
log::debug!("Completion intent: {:?}", completion_intent);
log::debug!("Completion mode: {:?}", self.completion_mode);
let messages = self.build_request_messages();
log::info!("Request will include {} messages", messages.len());
let tools = if let Some(tools) = self.tools(cx).log_err() {
tools
.filter_map(|tool| {
let tool_name = tool.name().to_string();
log::trace!("Including tool: {}", tool_name);
Some(LanguageModelRequestTool {
name: tool_name,
description: tool.description().to_string(),
input_schema: tool.input_schema(model.tool_input_format()).log_err()?,
})
})
.collect()
} else {
Vec::new()
};
log::info!("Request includes {} tools", tools.len());
let request = LanguageModelRequest {
thread_id: Some(self.id.to_string()),
prompt_id: Some(self.prompt_id.to_string()),
intent: Some(completion_intent),
mode: Some(self.completion_mode.into()),
messages,
tools,
tool_choice: None,
stop: Vec::new(),
temperature: AgentSettings::temperature_for_model(model, cx),
thinking_allowed: true,
};
log::debug!("Completion request built successfully");
Ok(request)
}
fn tools<'a>(&'a self, cx: &'a App) -> Result<impl Iterator<Item = &'a Arc<dyn AnyAgentTool>>> {
let model = self.model().context("No language model configured")?;
let profile = AgentSettings::get_global(cx)
.profiles
.get(&self.profile_id)
.context("profile not found")?;
let provider_id = model.provider_id();
Ok(self
.tools
.iter()
.filter(move |(_, tool)| tool.supported_provider(&provider_id))
.filter_map(|(tool_name, tool)| {
if profile.is_tool_enabled(tool_name) {
Some(tool)
} else {
None
}
})
.chain(self.context_server_registry.read(cx).servers().flat_map(
|(server_id, tools)| {
tools.iter().filter_map(|(tool_name, tool)| {
if profile.is_context_server_tool_enabled(&server_id.0, tool_name) {
Some(tool)
} else {
None
}
})
},
)))
}
fn build_request_messages(&self) -> Vec<LanguageModelRequestMessage> {
log::trace!(
"Building request messages from {} thread messages",
self.messages.len()
);
let mut messages = vec![self.build_system_message()];
for message in &self.messages {
match message {
Message::User(message) => messages.push(message.to_request()),
Message::Agent(message) => messages.extend(message.to_request()),
Message::Resume => messages.push(LanguageModelRequestMessage {
role: Role::User,
content: vec!["Continue where you left off".into()],
cache: false,
}),
}
}
if let Some(message) = self.pending_message.as_ref() {
messages.extend(message.to_request());
}
if let Some(last_user_message) = messages
.iter_mut()
.rev()
.find(|message| message.role == Role::User)
{
last_user_message.cache = true;
}
messages
}
pub fn to_markdown(&self) -> String {
let mut markdown = String::new();
for (ix, message) in self.messages.iter().enumerate() {
if ix > 0 {
markdown.push('\n');
}
markdown.push_str(&message.to_markdown());
}
if let Some(message) = self.pending_message.as_ref() {
markdown.push('\n');
markdown.push_str(&message.to_markdown());
}
markdown
}
fn advance_prompt_id(&mut self) {
self.prompt_id = PromptId::new();
}
}
struct RunningTurn {
/// Holds the task that handles agent interaction until the end of the turn.
/// Survives across multiple requests as the model performs tool calls and
/// we run tools, report their results.
_task: Task<()>,
/// The current event stream for the running turn. Used to report a final
/// cancellation event if we cancel the turn.
event_stream: AgentResponseEventStream,
}
impl RunningTurn {
fn cancel(self) {
log::debug!("Cancelling in progress turn");
self.event_stream.send_canceled();
}
}
pub trait AgentTool
where
Self: 'static + Sized,
{
type Input: for<'de> Deserialize<'de> + Serialize + JsonSchema;
type Output: for<'de> Deserialize<'de> + Serialize + Into<LanguageModelToolResultContent>;
fn name(&self) -> SharedString;
fn description(&self) -> SharedString {
let schema = schemars::schema_for!(Self::Input);
SharedString::new(
schema
.get("description")
.and_then(|description| description.as_str())
.unwrap_or_default(),
)
}
fn kind(&self) -> acp::ToolKind;
/// The initial tool title to display. Can be updated during the tool run.
fn initial_title(&self, input: Result<Self::Input, serde_json::Value>) -> SharedString;
/// Returns the JSON schema that describes the tool's input.
fn input_schema(&self) -> Schema {
schemars::schema_for!(Self::Input)
}
/// Some tools rely on a provider for the underlying billing or other reasons.
/// Allow the tool to check if they are compatible, or should be filtered out.
fn supported_provider(&self, _provider: &LanguageModelProviderId) -> bool {
true
}
/// Runs the tool with the provided input.
fn run(
self: Arc<Self>,
input: Self::Input,
event_stream: ToolCallEventStream,
cx: &mut App,
) -> Task<Result<Self::Output>>;
fn erase(self) -> Arc<dyn AnyAgentTool> {
Arc::new(Erased(Arc::new(self)))
}
}
pub struct Erased<T>(T);
pub struct AgentToolOutput {
pub llm_output: LanguageModelToolResultContent,
pub raw_output: serde_json::Value,
}
pub trait AnyAgentTool {
fn name(&self) -> SharedString;
fn description(&self) -> SharedString;
fn kind(&self) -> acp::ToolKind;
fn initial_title(&self, input: serde_json::Value) -> SharedString;
fn input_schema(&self, format: LanguageModelToolSchemaFormat) -> Result<serde_json::Value>;
fn supported_provider(&self, _provider: &LanguageModelProviderId) -> bool {
true
}
fn run(
self: Arc<Self>,
input: serde_json::Value,
event_stream: ToolCallEventStream,
cx: &mut App,
) -> Task<Result<AgentToolOutput>>;
}
impl<T> AnyAgentTool for Erased<Arc<T>>
where
T: AgentTool,
{
fn name(&self) -> SharedString {
self.0.name()
}
fn description(&self) -> SharedString {
self.0.description()
}
fn kind(&self) -> agent_client_protocol::ToolKind {
self.0.kind()
}
fn initial_title(&self, input: serde_json::Value) -> SharedString {
let parsed_input = serde_json::from_value(input.clone()).map_err(|_| input);
self.0.initial_title(parsed_input)
}
fn input_schema(&self, format: LanguageModelToolSchemaFormat) -> Result<serde_json::Value> {
let mut json = serde_json::to_value(self.0.input_schema())?;
adapt_schema_to_format(&mut json, format)?;
Ok(json)
}
fn supported_provider(&self, provider: &LanguageModelProviderId) -> bool {
self.0.supported_provider(provider)
}
fn run(
self: Arc<Self>,
input: serde_json::Value,
event_stream: ToolCallEventStream,
cx: &mut App,
) -> Task<Result<AgentToolOutput>> {
cx.spawn(async move |cx| {
let input = serde_json::from_value(input)?;
let output = cx
.update(|cx| self.0.clone().run(input, event_stream, cx))?
.await?;
let raw_output = serde_json::to_value(&output)?;
Ok(AgentToolOutput {
llm_output: output.into(),
raw_output,
})
})
}
}
#[derive(Clone)]
struct AgentResponseEventStream(mpsc::UnboundedSender<Result<AgentResponseEvent>>);
impl AgentResponseEventStream {
fn send_text(&self, text: &str) {
self.0
.unbounded_send(Ok(AgentResponseEvent::Text(text.to_string())))
.ok();
}
fn send_thinking(&self, text: &str) {
self.0
.unbounded_send(Ok(AgentResponseEvent::Thinking(text.to_string())))
.ok();
}
fn send_tool_call(
&self,
id: &LanguageModelToolUseId,
title: SharedString,
kind: acp::ToolKind,
input: serde_json::Value,
) {
self.0
.unbounded_send(Ok(AgentResponseEvent::ToolCall(Self::initial_tool_call(
id,
title.to_string(),
kind,
input,
))))
.ok();
}
fn initial_tool_call(
id: &LanguageModelToolUseId,
title: String,
kind: acp::ToolKind,
input: serde_json::Value,
) -> acp::ToolCall {
acp::ToolCall {
id: acp::ToolCallId(id.to_string().into()),
title,
kind,
status: acp::ToolCallStatus::Pending,
content: vec![],
locations: vec![],
raw_input: Some(input),
raw_output: None,
}
}
fn update_tool_call_fields(
&self,
tool_use_id: &LanguageModelToolUseId,
fields: acp::ToolCallUpdateFields,
) {
self.0
.unbounded_send(Ok(AgentResponseEvent::ToolCallUpdate(
acp::ToolCallUpdate {
id: acp::ToolCallId(tool_use_id.to_string().into()),
fields,
}
.into(),
)))
.ok();
}
fn send_stop(&self, reason: StopReason) {
match reason {
StopReason::EndTurn => {
self.0
.unbounded_send(Ok(AgentResponseEvent::Stop(acp::StopReason::EndTurn)))
.ok();
}
StopReason::MaxTokens => {
self.0
.unbounded_send(Ok(AgentResponseEvent::Stop(acp::StopReason::MaxTokens)))
.ok();
}
StopReason::Refusal => {
self.0
.unbounded_send(Ok(AgentResponseEvent::Stop(acp::StopReason::Refusal)))
.ok();
}
StopReason::ToolUse => {}
}
}
fn send_canceled(&self) {
self.0
.unbounded_send(Ok(AgentResponseEvent::Stop(acp::StopReason::Canceled)))
.ok();
}
fn send_error(&self, error: impl Into<anyhow::Error>) {
self.0.unbounded_send(Err(error.into())).ok();
}
}
#[derive(Clone)]
pub struct ToolCallEventStream {
tool_use_id: LanguageModelToolUseId,
stream: AgentResponseEventStream,
fs: Option<Arc<dyn Fs>>,
}
impl ToolCallEventStream {
#[cfg(test)]
pub fn test() -> (Self, ToolCallEventStreamReceiver) {
let (events_tx, events_rx) = mpsc::unbounded::<Result<AgentResponseEvent>>();
let stream =
ToolCallEventStream::new("test_id".into(), AgentResponseEventStream(events_tx), None);
(stream, ToolCallEventStreamReceiver(events_rx))
}
fn new(
tool_use_id: LanguageModelToolUseId,
stream: AgentResponseEventStream,
fs: Option<Arc<dyn Fs>>,
) -> Self {
Self {
tool_use_id,
stream,
fs,
}
}
pub fn update_fields(&self, fields: acp::ToolCallUpdateFields) {
self.stream
.update_tool_call_fields(&self.tool_use_id, fields);
}
pub fn update_diff(&self, diff: Entity<acp_thread::Diff>) {
self.stream
.0
.unbounded_send(Ok(AgentResponseEvent::ToolCallUpdate(
acp_thread::ToolCallUpdateDiff {
id: acp::ToolCallId(self.tool_use_id.to_string().into()),
diff,
}
.into(),
)))
.ok();
}
pub fn update_terminal(&self, terminal: Entity<acp_thread::Terminal>) {
self.stream
.0
.unbounded_send(Ok(AgentResponseEvent::ToolCallUpdate(
acp_thread::ToolCallUpdateTerminal {
id: acp::ToolCallId(self.tool_use_id.to_string().into()),
terminal,
}
.into(),
)))
.ok();
}
pub fn authorize(&self, title: impl Into<String>, cx: &mut App) -> Task<Result<()>> {
if agent_settings::AgentSettings::get_global(cx).always_allow_tool_actions {
return Task::ready(Ok(()));
}
let (response_tx, response_rx) = oneshot::channel();
self.stream
.0
.unbounded_send(Ok(AgentResponseEvent::ToolCallAuthorization(
ToolCallAuthorization {
tool_call: acp::ToolCallUpdate {
id: acp::ToolCallId(self.tool_use_id.to_string().into()),
fields: acp::ToolCallUpdateFields {
title: Some(title.into()),
..Default::default()
},
},
options: vec![
acp::PermissionOption {
id: acp::PermissionOptionId("always_allow".into()),
name: "Always Allow".into(),
kind: acp::PermissionOptionKind::AllowAlways,
},
acp::PermissionOption {
id: acp::PermissionOptionId("allow".into()),
name: "Allow".into(),
kind: acp::PermissionOptionKind::AllowOnce,
},
acp::PermissionOption {
id: acp::PermissionOptionId("deny".into()),
name: "Deny".into(),
kind: acp::PermissionOptionKind::RejectOnce,
},
],
response: response_tx,
},
)))
.ok();
let fs = self.fs.clone();
cx.spawn(async move |cx| match response_rx.await?.0.as_ref() {
"always_allow" => {
if let Some(fs) = fs.clone() {
cx.update(|cx| {
update_settings_file::<AgentSettings>(fs, cx, |settings, _| {
settings.set_always_allow_tool_actions(true);
});
})?;
}
Ok(())
}
"allow" => Ok(()),
_ => Err(anyhow!("Permission to run tool denied by user")),
})
}
}
#[cfg(test)]
pub struct ToolCallEventStreamReceiver(mpsc::UnboundedReceiver<Result<AgentResponseEvent>>);
#[cfg(test)]
impl ToolCallEventStreamReceiver {
pub async fn expect_authorization(&mut self) -> ToolCallAuthorization {
let event = self.0.next().await;
if let Some(Ok(AgentResponseEvent::ToolCallAuthorization(auth))) = event {
auth
} else {
panic!("Expected ToolCallAuthorization but got: {:?}", event);
}
}
pub async fn expect_terminal(&mut self) -> Entity<acp_thread::Terminal> {
let event = self.0.next().await;
if let Some(Ok(AgentResponseEvent::ToolCallUpdate(
acp_thread::ToolCallUpdate::UpdateTerminal(update),
))) = event
{
update.terminal
} else {
panic!("Expected terminal but got: {:?}", event);
}
}
}
#[cfg(test)]
impl std::ops::Deref for ToolCallEventStreamReceiver {
type Target = mpsc::UnboundedReceiver<Result<AgentResponseEvent>>;
fn deref(&self) -> &Self::Target {
&self.0
}
}
#[cfg(test)]
impl std::ops::DerefMut for ToolCallEventStreamReceiver {
fn deref_mut(&mut self) -> &mut Self::Target {
&mut self.0
}
}
impl From<&str> for UserMessageContent {
fn from(text: &str) -> Self {
Self::Text(text.into())
}
}
impl From<acp::ContentBlock> for UserMessageContent {
fn from(value: acp::ContentBlock) -> Self {
match value {
acp::ContentBlock::Text(text_content) => Self::Text(text_content.text),
acp::ContentBlock::Image(image_content) => Self::Image(convert_image(image_content)),
acp::ContentBlock::Audio(_) => {
// TODO
Self::Text("[audio]".to_string())
}
acp::ContentBlock::ResourceLink(resource_link) => {
match MentionUri::parse(&resource_link.uri) {
Ok(uri) => Self::Mention {
uri,
content: String::new(),
},
Err(err) => {
log::error!("Failed to parse mention link: {}", err);
Self::Text(format!("[{}]({})", resource_link.name, resource_link.uri))
}
}
}
acp::ContentBlock::Resource(resource) => match resource.resource {
acp::EmbeddedResourceResource::TextResourceContents(resource) => {
match MentionUri::parse(&resource.uri) {
Ok(uri) => Self::Mention {
uri,
content: resource.text,
},
Err(err) => {
log::error!("Failed to parse mention link: {}", err);
Self::Text(
MarkdownCodeBlock {
tag: &resource.uri,
text: &resource.text,
}
.to_string(),
)
}
}
}
acp::EmbeddedResourceResource::BlobResourceContents(_) => {
// TODO
Self::Text("[blob]".to_string())
}
},
}
}
}
fn convert_image(image_content: acp::ImageContent) -> LanguageModelImage {
LanguageModelImage {
source: image_content.data.into(),
// TODO: make this optional?
size: gpui::Size::new(0.into(), 0.into()),
}
}