Files
llmx/codex-rs/tui/src/chatwidget.rs

550 lines
20 KiB
Rust
Raw Normal View History

use std::collections::HashMap;
use std::path::PathBuf;
use std::sync::Arc;
use std::time::Duration;
use codex_core::codex_wrapper::CodexConversation;
use codex_core::codex_wrapper::init_codex;
use codex_core::config::Config;
use codex_core::protocol::AgentMessageDeltaEvent;
use codex_core::protocol::AgentMessageEvent;
use codex_core::protocol::AgentReasoningDeltaEvent;
use codex_core::protocol::AgentReasoningEvent;
use codex_core::protocol::ApplyPatchApprovalRequestEvent;
use codex_core::protocol::ErrorEvent;
use codex_core::protocol::Event;
use codex_core::protocol::EventMsg;
use codex_core::protocol::ExecApprovalRequestEvent;
use codex_core::protocol::ExecCommandBeginEvent;
use codex_core::protocol::ExecCommandEndEvent;
use codex_core::protocol::InputItem;
use codex_core::protocol::McpToolCallBeginEvent;
use codex_core::protocol::McpToolCallEndEvent;
use codex_core::protocol::Op;
use codex_core::protocol::PatchApplyBeginEvent;
use codex_core::protocol::TaskCompleteEvent;
feat: show number of tokens remaining in UI (#1388) When using the OpenAI Responses API, we now record the `usage` field for a `"response.completed"` event, which includes metrics about the number of tokens consumed. We also introduce `openai_model_info.rs`, which includes current data about the most common OpenAI models available via the API (specifically `context_window` and `max_output_tokens`). If Codex does not recognize the model, you can set `model_context_window` and `model_max_output_tokens` explicitly in `config.toml`. When then introduce a new event type to `protocol.rs`, `TokenCount`, which includes the `TokenUsage` for the most recent turn. Finally, we update the TUI to record the running sum of tokens used so the percentage of available context window remaining can be reported via the placeholder text for the composer: ![Screenshot 2025-06-25 at 11 20 55 PM](https://github.com/user-attachments/assets/6fd6982f-7247-4f14-84b2-2e600cb1fd49) We could certainly get much fancier with this (such as reporting the estimated cost of the conversation), but for now, we are just trying to achieve feature parity with the TypeScript CLI. Though arguably this improves upon the TypeScript CLI, as the TypeScript CLI uses heuristics to estimate the number of tokens used rather than using the `usage` information directly: https://github.com/openai/codex/blob/296996d74e345b1b05d8c3451a06ace21c5ada96/codex-cli/src/utils/approximate-tokens-used.ts#L3-L16 Fixes https://github.com/openai/codex/issues/1242
2025-06-25 23:31:11 -07:00
use codex_core::protocol::TokenUsage;
use crossterm::event::KeyEvent;
use crossterm::event::KeyEventKind;
use ratatui::buffer::Buffer;
use ratatui::layout::Rect;
use ratatui::widgets::Widget;
use ratatui::widgets::WidgetRef;
use tokio::sync::mpsc::UnboundedSender;
use tokio::sync::mpsc::unbounded_channel;
use crate::app_event::AppEvent;
use crate::app_event_sender::AppEventSender;
use crate::bottom_pane::BottomPane;
use crate::bottom_pane::BottomPaneParams;
use crate::bottom_pane::CancellationEvent;
use crate::bottom_pane::InputResult;
use crate::exec_command::strip_bash_lc_and_escape;
use crate::history_cell::CommandOutput;
use crate::history_cell::HistoryCell;
use crate::history_cell::PatchEventType;
use crate::user_approval_widget::ApprovalRequest;
use codex_file_search::FileMatch;
struct RunningCommand {
command: Vec<String>,
#[allow(dead_code)]
cwd: PathBuf,
}
pub(crate) struct ChatWidget<'a> {
app_event_tx: AppEventSender,
codex_op_tx: UnboundedSender<Op>,
bottom_pane: BottomPane<'a>,
config: Config,
initial_user_message: Option<UserMessage>,
feat: show number of tokens remaining in UI (#1388) When using the OpenAI Responses API, we now record the `usage` field for a `"response.completed"` event, which includes metrics about the number of tokens consumed. We also introduce `openai_model_info.rs`, which includes current data about the most common OpenAI models available via the API (specifically `context_window` and `max_output_tokens`). If Codex does not recognize the model, you can set `model_context_window` and `model_max_output_tokens` explicitly in `config.toml`. When then introduce a new event type to `protocol.rs`, `TokenCount`, which includes the `TokenUsage` for the most recent turn. Finally, we update the TUI to record the running sum of tokens used so the percentage of available context window remaining can be reported via the placeholder text for the composer: ![Screenshot 2025-06-25 at 11 20 55 PM](https://github.com/user-attachments/assets/6fd6982f-7247-4f14-84b2-2e600cb1fd49) We could certainly get much fancier with this (such as reporting the estimated cost of the conversation), but for now, we are just trying to achieve feature parity with the TypeScript CLI. Though arguably this improves upon the TypeScript CLI, as the TypeScript CLI uses heuristics to estimate the number of tokens used rather than using the `usage` information directly: https://github.com/openai/codex/blob/296996d74e345b1b05d8c3451a06ace21c5ada96/codex-cli/src/utils/approximate-tokens-used.ts#L3-L16 Fixes https://github.com/openai/codex/issues/1242
2025-06-25 23:31:11 -07:00
token_usage: TokenUsage,
reasoning_buffer: String,
// Buffer for streaming assistant answer text; we do not surface partial
// We wait for the final AgentMessage event and then emit the full text
// at once into scrollback so the history contains a single message.
answer_buffer: String,
running_commands: HashMap<String, RunningCommand>,
}
struct UserMessage {
text: String,
image_paths: Vec<PathBuf>,
}
impl From<String> for UserMessage {
fn from(text: String) -> Self {
Self {
text,
image_paths: Vec::new(),
}
}
}
fn create_initial_user_message(text: String, image_paths: Vec<PathBuf>) -> Option<UserMessage> {
if text.is_empty() && image_paths.is_empty() {
None
} else {
Some(UserMessage { text, image_paths })
}
}
impl ChatWidget<'_> {
pub(crate) fn new(
config: Config,
app_event_tx: AppEventSender,
initial_prompt: Option<String>,
initial_images: Vec<PathBuf>,
enhanced_keys_supported: bool,
) -> Self {
let (codex_op_tx, mut codex_op_rx) = unbounded_channel::<Op>();
let app_event_tx_clone = app_event_tx.clone();
// Create the Codex asynchronously so the UI loads as quickly as possible.
let config_for_agent_loop = config.clone();
tokio::spawn(async move {
let CodexConversation {
codex,
session_configured,
..
} = match init_codex(config_for_agent_loop).await {
Ok(vals) => vals,
Err(e) => {
// TODO: surface this error to the user.
tracing::error!("failed to initialize codex: {e}");
return;
}
};
// Forward the captured `SessionInitialized` event that was consumed
// inside `init_codex()` so it can be rendered in the UI.
app_event_tx_clone.send(AppEvent::CodexEvent(session_configured.clone()));
let codex = Arc::new(codex);
let codex_clone = codex.clone();
tokio::spawn(async move {
while let Some(op) = codex_op_rx.recv().await {
let id = codex_clone.submit(op).await;
if let Err(e) = id {
tracing::error!("failed to submit op: {e}");
}
}
});
while let Ok(event) = codex.next_event().await {
app_event_tx_clone.send(AppEvent::CodexEvent(event));
}
});
Self {
app_event_tx: app_event_tx.clone(),
codex_op_tx,
bottom_pane: BottomPane::new(BottomPaneParams {
app_event_tx,
has_input_focus: true,
enhanced_keys_supported,
}),
config,
initial_user_message: create_initial_user_message(
initial_prompt.unwrap_or_default(),
initial_images,
),
feat: show number of tokens remaining in UI (#1388) When using the OpenAI Responses API, we now record the `usage` field for a `"response.completed"` event, which includes metrics about the number of tokens consumed. We also introduce `openai_model_info.rs`, which includes current data about the most common OpenAI models available via the API (specifically `context_window` and `max_output_tokens`). If Codex does not recognize the model, you can set `model_context_window` and `model_max_output_tokens` explicitly in `config.toml`. When then introduce a new event type to `protocol.rs`, `TokenCount`, which includes the `TokenUsage` for the most recent turn. Finally, we update the TUI to record the running sum of tokens used so the percentage of available context window remaining can be reported via the placeholder text for the composer: ![Screenshot 2025-06-25 at 11 20 55 PM](https://github.com/user-attachments/assets/6fd6982f-7247-4f14-84b2-2e600cb1fd49) We could certainly get much fancier with this (such as reporting the estimated cost of the conversation), but for now, we are just trying to achieve feature parity with the TypeScript CLI. Though arguably this improves upon the TypeScript CLI, as the TypeScript CLI uses heuristics to estimate the number of tokens used rather than using the `usage` information directly: https://github.com/openai/codex/blob/296996d74e345b1b05d8c3451a06ace21c5ada96/codex-cli/src/utils/approximate-tokens-used.ts#L3-L16 Fixes https://github.com/openai/codex/issues/1242
2025-06-25 23:31:11 -07:00
token_usage: TokenUsage::default(),
reasoning_buffer: String::new(),
answer_buffer: String::new(),
running_commands: HashMap::new(),
}
}
pub fn desired_height(&self, width: u16) -> u16 {
self.bottom_pane.desired_height(width)
}
pub(crate) fn handle_key_event(&mut self, key_event: KeyEvent) {
if key_event.kind == KeyEventKind::Press {
self.bottom_pane.clear_ctrl_c_quit_hint();
}
match self.bottom_pane.handle_key_event(key_event) {
InputResult::Submitted(text) => {
self.submit_user_message(text.into());
}
InputResult::None => {}
}
}
pub(crate) fn handle_paste(&mut self, text: String) {
self.bottom_pane.handle_paste(text);
}
fn add_to_history(&mut self, cell: HistoryCell) {
self.app_event_tx
.send(AppEvent::InsertHistory(cell.plain_lines()));
}
fn submit_user_message(&mut self, user_message: UserMessage) {
let UserMessage { text, image_paths } = user_message;
let mut items: Vec<InputItem> = Vec::new();
if !text.is_empty() {
items.push(InputItem::Text { text: text.clone() });
}
for path in image_paths {
items.push(InputItem::LocalImage { path });
}
if items.is_empty() {
return;
}
self.codex_op_tx
.send(Op::UserInput { items })
.unwrap_or_else(|e| {
tracing::error!("failed to send message: {e}");
});
feat: record messages from user in ~/.codex/history.jsonl (#939) This is a large change to support a "history" feature like you would expect in a shell like Bash. History events are recorded in `$CODEX_HOME/history.jsonl`. Because it is a JSONL file, it is straightforward to append new entries (as opposed to the TypeScript file that uses `$CODEX_HOME/history.json`, so to be valid JSON, each new entry entails rewriting the entire file). Because it is possible for there to be multiple instances of Codex CLI writing to `history.jsonl` at once, we use advisory file locking when working with `history.jsonl` in `codex-rs/core/src/message_history.rs`. Because we believe history is a sufficiently useful feature, we enable it by default. Though to provide some safety, we set the file permissions of `history.jsonl` to be `o600` so that other users on the system cannot read the user's history. We do not yet support a default list of `SENSITIVE_PATTERNS` as the TypeScript CLI does: https://github.com/openai/codex/blob/3fdf9df1335ac9501e3fb0e61715359145711e8b/codex-cli/src/utils/storage/command-history.ts#L10-L17 We are going to take a more conservative approach to this list in the Rust CLI. For example, while `/\b[A-Za-z0-9-_]{20,}\b/` might exclude sensitive information like API tokens, it would also exclude valuable information such as references to Git commits. As noted in the updated documentation, users can opt-out of history by adding the following to `config.toml`: ```toml [history] persistence = "none" ``` Because `history.jsonl` could, in theory, be quite large, we take a[n arguably overly pedantic] approach in reading history entries into memory. Specifically, we start by telling the client the current number of entries in the history file (`history_entry_count`) as well as the inode (`history_log_id`) of `history.jsonl` (see the new fields on `SessionConfiguredEvent`). The client is responsible for keeping new entries in memory to create a "local history," but if the user hits up enough times to go "past" the end of local history, then the client should use the new `GetHistoryEntryRequest` in the protocol to fetch older entries. Specifically, it should pass the `history_log_id` it was given originally and work backwards from `history_entry_count`. (It should really fetch history in batches rather than one-at-a-time, but that is something we can improve upon in subsequent PRs.) The motivation behind this crazy scheme is that it is designed to defend against: * The `history.jsonl` being truncated during the session such that the index into the history is no longer consistent with what had been read up to that point. We do not yet have logic to enforce a `max_bytes` for `history.jsonl`, but once we do, we will aspire to implement it in a way that should result in a new inode for the file on most systems. * New items from concurrent Codex CLI sessions amending to the history. Because, in absence of truncation, `history.jsonl` is an append-only log, so long as the client reads backwards from `history_entry_count`, it should always get a consistent view of history. (That said, it will not be able to read _new_ commands from concurrent sessions, but perhaps we will introduce a `/` command to reload latest history or something down the road.) Admittedly, my testing of this feature thus far has been fairly light. I expect we will find bugs and introduce enhancements/fixes going forward.
2025-05-15 16:26:23 -07:00
// Persist the text to cross-session message history.
if !text.is_empty() {
self.codex_op_tx
.send(Op::AddToHistory { text: text.clone() })
.unwrap_or_else(|e| {
tracing::error!("failed to send AddHistory op: {e}");
});
}
// Only show text portion in conversation history for now.
if !text.is_empty() {
self.add_to_history(HistoryCell::new_user_prompt(text.clone()));
}
}
pub(crate) fn handle_codex_event(&mut self, event: Event) {
let Event { id, msg } = event;
match msg {
EventMsg::SessionConfigured(event) => {
feat: record messages from user in ~/.codex/history.jsonl (#939) This is a large change to support a "history" feature like you would expect in a shell like Bash. History events are recorded in `$CODEX_HOME/history.jsonl`. Because it is a JSONL file, it is straightforward to append new entries (as opposed to the TypeScript file that uses `$CODEX_HOME/history.json`, so to be valid JSON, each new entry entails rewriting the entire file). Because it is possible for there to be multiple instances of Codex CLI writing to `history.jsonl` at once, we use advisory file locking when working with `history.jsonl` in `codex-rs/core/src/message_history.rs`. Because we believe history is a sufficiently useful feature, we enable it by default. Though to provide some safety, we set the file permissions of `history.jsonl` to be `o600` so that other users on the system cannot read the user's history. We do not yet support a default list of `SENSITIVE_PATTERNS` as the TypeScript CLI does: https://github.com/openai/codex/blob/3fdf9df1335ac9501e3fb0e61715359145711e8b/codex-cli/src/utils/storage/command-history.ts#L10-L17 We are going to take a more conservative approach to this list in the Rust CLI. For example, while `/\b[A-Za-z0-9-_]{20,}\b/` might exclude sensitive information like API tokens, it would also exclude valuable information such as references to Git commits. As noted in the updated documentation, users can opt-out of history by adding the following to `config.toml`: ```toml [history] persistence = "none" ``` Because `history.jsonl` could, in theory, be quite large, we take a[n arguably overly pedantic] approach in reading history entries into memory. Specifically, we start by telling the client the current number of entries in the history file (`history_entry_count`) as well as the inode (`history_log_id`) of `history.jsonl` (see the new fields on `SessionConfiguredEvent`). The client is responsible for keeping new entries in memory to create a "local history," but if the user hits up enough times to go "past" the end of local history, then the client should use the new `GetHistoryEntryRequest` in the protocol to fetch older entries. Specifically, it should pass the `history_log_id` it was given originally and work backwards from `history_entry_count`. (It should really fetch history in batches rather than one-at-a-time, but that is something we can improve upon in subsequent PRs.) The motivation behind this crazy scheme is that it is designed to defend against: * The `history.jsonl` being truncated during the session such that the index into the history is no longer consistent with what had been read up to that point. We do not yet have logic to enforce a `max_bytes` for `history.jsonl`, but once we do, we will aspire to implement it in a way that should result in a new inode for the file on most systems. * New items from concurrent Codex CLI sessions amending to the history. Because, in absence of truncation, `history.jsonl` is an append-only log, so long as the client reads backwards from `history_entry_count`, it should always get a consistent view of history. (That said, it will not be able to read _new_ commands from concurrent sessions, but perhaps we will introduce a `/` command to reload latest history or something down the road.) Admittedly, my testing of this feature thus far has been fairly light. I expect we will find bugs and introduce enhancements/fixes going forward.
2025-05-15 16:26:23 -07:00
self.bottom_pane
.set_history_metadata(event.history_log_id, event.history_entry_count);
// Record session information at the top of the conversation.
self.add_to_history(HistoryCell::new_session_info(&self.config, event, true));
if let Some(user_message) = self.initial_user_message.take() {
// If the user provided an initial message, add it to the
// conversation history.
self.submit_user_message(user_message);
}
self.request_redraw();
}
EventMsg::AgentMessage(AgentMessageEvent { message }) => {
// Final assistant answer. Prefer the fully provided message
// from the event; if it is empty fall back to any accumulated
// delta buffer (some providers may only stream deltas and send
// an empty final message).
let full = if message.is_empty() {
std::mem::take(&mut self.answer_buffer)
} else {
self.answer_buffer.clear();
message
};
if !full.is_empty() {
self.add_to_history(HistoryCell::new_agent_message(&self.config, full));
}
self.request_redraw();
}
EventMsg::AgentMessageDelta(AgentMessageDeltaEvent { delta }) => {
// Buffer only do not emit partial lines. This avoids cases
// where long responses appear truncated if the terminal
// wrapped early. The full message is emitted on
// AgentMessage.
self.answer_buffer.push_str(&delta);
}
EventMsg::AgentReasoningDelta(AgentReasoningDeltaEvent { delta }) => {
// Buffer only disable incremental reasoning streaming so we
// avoid truncated intermediate lines. Full text emitted on
// AgentReasoning.
self.reasoning_buffer.push_str(&delta);
}
EventMsg::AgentReasoning(AgentReasoningEvent { text }) => {
// Emit full reasoning text once. Some providers might send
// final event with empty text if only deltas were used.
let full = if text.is_empty() {
std::mem::take(&mut self.reasoning_buffer)
} else {
self.reasoning_buffer.clear();
text
};
if !full.is_empty() {
self.add_to_history(HistoryCell::new_agent_reasoning(&self.config, full));
}
self.request_redraw();
}
EventMsg::TaskStarted => {
self.bottom_pane.clear_ctrl_c_quit_hint();
self.bottom_pane.set_task_running(true);
self.request_redraw();
}
EventMsg::TaskComplete(TaskCompleteEvent {
last_agent_message: _,
}) => {
self.bottom_pane.set_task_running(false);
self.request_redraw();
}
feat: show number of tokens remaining in UI (#1388) When using the OpenAI Responses API, we now record the `usage` field for a `"response.completed"` event, which includes metrics about the number of tokens consumed. We also introduce `openai_model_info.rs`, which includes current data about the most common OpenAI models available via the API (specifically `context_window` and `max_output_tokens`). If Codex does not recognize the model, you can set `model_context_window` and `model_max_output_tokens` explicitly in `config.toml`. When then introduce a new event type to `protocol.rs`, `TokenCount`, which includes the `TokenUsage` for the most recent turn. Finally, we update the TUI to record the running sum of tokens used so the percentage of available context window remaining can be reported via the placeholder text for the composer: ![Screenshot 2025-06-25 at 11 20 55 PM](https://github.com/user-attachments/assets/6fd6982f-7247-4f14-84b2-2e600cb1fd49) We could certainly get much fancier with this (such as reporting the estimated cost of the conversation), but for now, we are just trying to achieve feature parity with the TypeScript CLI. Though arguably this improves upon the TypeScript CLI, as the TypeScript CLI uses heuristics to estimate the number of tokens used rather than using the `usage` information directly: https://github.com/openai/codex/blob/296996d74e345b1b05d8c3451a06ace21c5ada96/codex-cli/src/utils/approximate-tokens-used.ts#L3-L16 Fixes https://github.com/openai/codex/issues/1242
2025-06-25 23:31:11 -07:00
EventMsg::TokenCount(token_usage) => {
self.token_usage = add_token_usage(&self.token_usage, &token_usage);
self.bottom_pane
.set_token_usage(self.token_usage.clone(), self.config.model_context_window);
}
EventMsg::Error(ErrorEvent { message }) => {
self.add_to_history(HistoryCell::new_error_event(message.clone()));
self.bottom_pane.set_task_running(false);
}
EventMsg::PlanUpdate(update) => {
self.add_to_history(HistoryCell::new_plan_update(update));
self.request_redraw();
}
EventMsg::ExecApprovalRequest(ExecApprovalRequestEvent {
call_id: _,
command,
cwd,
reason,
}) => {
// Print the command to the history so it is visible in the
// transcript *before* the modal asks for approval.
let cmdline = strip_bash_lc_and_escape(&command);
let text = format!(
"command requires approval:\n$ {cmdline}{reason}",
reason = reason
.as_ref()
.map(|r| format!("\n{r}"))
.unwrap_or_default()
);
self.add_to_history(HistoryCell::new_background_event(text));
let request = ApprovalRequest::Exec {
id,
command,
cwd,
reason,
};
self.bottom_pane.push_approval_request(request);
self.request_redraw();
}
EventMsg::ApplyPatchApprovalRequest(ApplyPatchApprovalRequestEvent {
call_id: _,
changes,
reason,
grant_root,
}) => {
// ------------------------------------------------------------------
// Before we even prompt the user for approval we surface the patch
// summary in the main conversation so that the dialog appears in a
// sensible chronological order:
// (1) codex → proposes patch (HistoryCell::PendingPatch)
// (2) UI → asks for approval (BottomPane)
// This mirrors how command execution is shown (command begins →
// approval dialog) and avoids surprising the user with a modal
// prompt before they have seen *what* is being requested.
// ------------------------------------------------------------------
self.add_to_history(HistoryCell::new_patch_event(
PatchEventType::ApprovalRequest,
changes,
));
// Now surface the approval request in the BottomPane as before.
let request = ApprovalRequest::ApplyPatch {
id,
reason,
grant_root,
};
self.bottom_pane.push_approval_request(request);
self.request_redraw();
}
EventMsg::ExecCommandBegin(ExecCommandBeginEvent {
call_id,
command,
cwd,
}) => {
self.running_commands.insert(
call_id,
RunningCommand {
command: command.clone(),
cwd: cwd.clone(),
},
);
self.add_to_history(HistoryCell::new_active_exec_command(command));
}
EventMsg::ExecCommandStdoutDelta(_) => {}
EventMsg::PatchApplyBegin(PatchApplyBeginEvent {
call_id: _,
auto_approved,
changes,
}) => {
// Even when a patch is autoapproved we still display the
// summary so the user can follow along.
self.add_to_history(HistoryCell::new_patch_event(
PatchEventType::ApplyBegin { auto_approved },
changes,
));
}
EventMsg::ExecCommandEnd(ExecCommandEndEvent {
call_id,
exit_code,
stdout,
stderr,
}) => {
let cmd = self.running_commands.remove(&call_id);
self.add_to_history(HistoryCell::new_completed_exec_command(
cmd.map(|cmd| cmd.command).unwrap_or_else(|| vec![call_id]),
CommandOutput {
exit_code,
stdout,
stderr,
duration: Duration::from_secs(0),
},
));
}
EventMsg::McpToolCallBegin(McpToolCallBeginEvent {
call_id: _,
invocation,
}) => {
self.add_to_history(HistoryCell::new_active_mcp_tool_call(invocation));
}
EventMsg::McpToolCallEnd(McpToolCallEndEvent {
call_id: _,
duration,
invocation,
result,
}) => {
self.add_to_history(HistoryCell::new_completed_mcp_tool_call(
80,
invocation,
duration,
result
.as_ref()
.map(|r| r.is_error.unwrap_or(false))
.unwrap_or(false),
result,
));
}
feat: record messages from user in ~/.codex/history.jsonl (#939) This is a large change to support a "history" feature like you would expect in a shell like Bash. History events are recorded in `$CODEX_HOME/history.jsonl`. Because it is a JSONL file, it is straightforward to append new entries (as opposed to the TypeScript file that uses `$CODEX_HOME/history.json`, so to be valid JSON, each new entry entails rewriting the entire file). Because it is possible for there to be multiple instances of Codex CLI writing to `history.jsonl` at once, we use advisory file locking when working with `history.jsonl` in `codex-rs/core/src/message_history.rs`. Because we believe history is a sufficiently useful feature, we enable it by default. Though to provide some safety, we set the file permissions of `history.jsonl` to be `o600` so that other users on the system cannot read the user's history. We do not yet support a default list of `SENSITIVE_PATTERNS` as the TypeScript CLI does: https://github.com/openai/codex/blob/3fdf9df1335ac9501e3fb0e61715359145711e8b/codex-cli/src/utils/storage/command-history.ts#L10-L17 We are going to take a more conservative approach to this list in the Rust CLI. For example, while `/\b[A-Za-z0-9-_]{20,}\b/` might exclude sensitive information like API tokens, it would also exclude valuable information such as references to Git commits. As noted in the updated documentation, users can opt-out of history by adding the following to `config.toml`: ```toml [history] persistence = "none" ``` Because `history.jsonl` could, in theory, be quite large, we take a[n arguably overly pedantic] approach in reading history entries into memory. Specifically, we start by telling the client the current number of entries in the history file (`history_entry_count`) as well as the inode (`history_log_id`) of `history.jsonl` (see the new fields on `SessionConfiguredEvent`). The client is responsible for keeping new entries in memory to create a "local history," but if the user hits up enough times to go "past" the end of local history, then the client should use the new `GetHistoryEntryRequest` in the protocol to fetch older entries. Specifically, it should pass the `history_log_id` it was given originally and work backwards from `history_entry_count`. (It should really fetch history in batches rather than one-at-a-time, but that is something we can improve upon in subsequent PRs.) The motivation behind this crazy scheme is that it is designed to defend against: * The `history.jsonl` being truncated during the session such that the index into the history is no longer consistent with what had been read up to that point. We do not yet have logic to enforce a `max_bytes` for `history.jsonl`, but once we do, we will aspire to implement it in a way that should result in a new inode for the file on most systems. * New items from concurrent Codex CLI sessions amending to the history. Because, in absence of truncation, `history.jsonl` is an append-only log, so long as the client reads backwards from `history_entry_count`, it should always get a consistent view of history. (That said, it will not be able to read _new_ commands from concurrent sessions, but perhaps we will introduce a `/` command to reload latest history or something down the road.) Admittedly, my testing of this feature thus far has been fairly light. I expect we will find bugs and introduce enhancements/fixes going forward.
2025-05-15 16:26:23 -07:00
EventMsg::GetHistoryEntryResponse(event) => {
let codex_core::protocol::GetHistoryEntryResponseEvent {
offset,
log_id,
entry,
} = event;
// Inform bottom pane / composer.
self.bottom_pane
.on_history_entry_response(log_id, offset, entry.map(|e| e.text));
}
EventMsg::ShutdownComplete => {
self.app_event_tx.send(AppEvent::ExitRequest);
}
event => {
self.add_to_history(HistoryCell::new_background_event(format!("{event:?}")));
}
}
}
/// Update the live log preview while a task is running.
pub(crate) fn update_latest_log(&mut self, line: String) {
// Forward only if we are currently showing the status indicator.
self.bottom_pane.update_status_text(line);
}
fn request_redraw(&mut self) {
self.app_event_tx.send(AppEvent::RequestRedraw);
}
pub(crate) fn add_diff_output(&mut self, diff_output: String) {
self.add_to_history(HistoryCell::new_diff_output(diff_output.clone()));
}
feat: add support for @ to do file search (#1401) Introduces support for `@` to trigger a fuzzy-filename search in the composer. Under the hood, this leverages https://crates.io/crates/nucleo-matcher to do the fuzzy matching and https://crates.io/crates/ignore to build up the list of file candidates (so that it respects `.gitignore`). For simplicity (at least for now), we do not do any caching between searches like VS Code does for its file search: https://github.com/microsoft/vscode/blob/1d89ed699b2e924d418c856318a3e12bca67ff3a/src/vs/workbench/services/search/node/rawSearchService.ts#L212-L218 Because we do not do any caching, I saw queries take up to three seconds on large repositories with hundreds of thousands of files. To that end, we do not perform searches synchronously on each keystroke, but instead dispatch an event to do the search on a background thread that asynchronously reports back to the UI when the results are available. This is largely handled by the `FileSearchManager` introduced in this PR, which also has logic for debouncing requests so there is at most one search in flight at a time. While we could potentially polish and tune this feature further, it may already be overengineered for how it will be used, in practice, so we can improve things going forward if it turns out that this is not "good enough" in the wild. Note this feature does not work like `@` in the TypeScript CLI, which was more like directory-based tab completion. In the Rust CLI, `@` triggers a full-repo fuzzy-filename search. Fixes https://github.com/openai/codex/issues/1261.
2025-06-28 13:47:42 -07:00
/// Forward file-search results to the bottom pane.
pub(crate) fn apply_file_search_result(&mut self, query: String, matches: Vec<FileMatch>) {
feat: add support for @ to do file search (#1401) Introduces support for `@` to trigger a fuzzy-filename search in the composer. Under the hood, this leverages https://crates.io/crates/nucleo-matcher to do the fuzzy matching and https://crates.io/crates/ignore to build up the list of file candidates (so that it respects `.gitignore`). For simplicity (at least for now), we do not do any caching between searches like VS Code does for its file search: https://github.com/microsoft/vscode/blob/1d89ed699b2e924d418c856318a3e12bca67ff3a/src/vs/workbench/services/search/node/rawSearchService.ts#L212-L218 Because we do not do any caching, I saw queries take up to three seconds on large repositories with hundreds of thousands of files. To that end, we do not perform searches synchronously on each keystroke, but instead dispatch an event to do the search on a background thread that asynchronously reports back to the UI when the results are available. This is largely handled by the `FileSearchManager` introduced in this PR, which also has logic for debouncing requests so there is at most one search in flight at a time. While we could potentially polish and tune this feature further, it may already be overengineered for how it will be used, in practice, so we can improve things going forward if it turns out that this is not "good enough" in the wild. Note this feature does not work like `@` in the TypeScript CLI, which was more like directory-based tab completion. In the Rust CLI, `@` triggers a full-repo fuzzy-filename search. Fixes https://github.com/openai/codex/issues/1261.
2025-06-28 13:47:42 -07:00
self.bottom_pane.on_file_search_result(query, matches);
}
/// Handle Ctrl-C key press.
/// Returns CancellationEvent::Handled if the event was consumed by the UI, or
/// CancellationEvent::Ignored if the caller should handle it (e.g. exit).
pub(crate) fn on_ctrl_c(&mut self) -> CancellationEvent {
match self.bottom_pane.on_ctrl_c() {
CancellationEvent::Handled => return CancellationEvent::Handled,
CancellationEvent::Ignored => {}
}
if self.bottom_pane.is_task_running() {
self.bottom_pane.clear_ctrl_c_quit_hint();
self.submit_op(Op::Interrupt);
self.answer_buffer.clear();
self.reasoning_buffer.clear();
CancellationEvent::Ignored
} else if self.bottom_pane.ctrl_c_quit_hint_visible() {
self.submit_op(Op::Shutdown);
CancellationEvent::Handled
} else {
self.bottom_pane.show_ctrl_c_quit_hint();
CancellationEvent::Ignored
}
}
pub(crate) fn composer_is_empty(&self) -> bool {
self.bottom_pane.composer_is_empty()
}
/// Forward an `Op` directly to codex.
pub(crate) fn submit_op(&self, op: Op) {
if let Err(e) = self.codex_op_tx.send(op) {
tracing::error!("failed to submit op: {e}");
}
}
pub(crate) fn token_usage(&self) -> &TokenUsage {
&self.token_usage
}
pub(crate) fn clear_token_usage(&mut self) {
self.token_usage = TokenUsage::default();
self.bottom_pane
.set_token_usage(self.token_usage.clone(), self.config.model_context_window);
}
}
impl WidgetRef for &ChatWidget<'_> {
fn render_ref(&self, area: Rect, buf: &mut Buffer) {
// In the hybrid inline viewport mode we only draw the interactive
// bottom pane; history entries are injected directly into scrollback
// via `Terminal::insert_before`.
(&self.bottom_pane).render(area, buf);
}
}
feat: show number of tokens remaining in UI (#1388) When using the OpenAI Responses API, we now record the `usage` field for a `"response.completed"` event, which includes metrics about the number of tokens consumed. We also introduce `openai_model_info.rs`, which includes current data about the most common OpenAI models available via the API (specifically `context_window` and `max_output_tokens`). If Codex does not recognize the model, you can set `model_context_window` and `model_max_output_tokens` explicitly in `config.toml`. When then introduce a new event type to `protocol.rs`, `TokenCount`, which includes the `TokenUsage` for the most recent turn. Finally, we update the TUI to record the running sum of tokens used so the percentage of available context window remaining can be reported via the placeholder text for the composer: ![Screenshot 2025-06-25 at 11 20 55 PM](https://github.com/user-attachments/assets/6fd6982f-7247-4f14-84b2-2e600cb1fd49) We could certainly get much fancier with this (such as reporting the estimated cost of the conversation), but for now, we are just trying to achieve feature parity with the TypeScript CLI. Though arguably this improves upon the TypeScript CLI, as the TypeScript CLI uses heuristics to estimate the number of tokens used rather than using the `usage` information directly: https://github.com/openai/codex/blob/296996d74e345b1b05d8c3451a06ace21c5ada96/codex-cli/src/utils/approximate-tokens-used.ts#L3-L16 Fixes https://github.com/openai/codex/issues/1242
2025-06-25 23:31:11 -07:00
fn add_token_usage(current_usage: &TokenUsage, new_usage: &TokenUsage) -> TokenUsage {
let cached_input_tokens = match (
current_usage.cached_input_tokens,
new_usage.cached_input_tokens,
) {
(Some(current), Some(new)) => Some(current + new),
(Some(current), None) => Some(current),
(None, Some(new)) => Some(new),
(None, None) => None,
};
let reasoning_output_tokens = match (
current_usage.reasoning_output_tokens,
new_usage.reasoning_output_tokens,
) {
(Some(current), Some(new)) => Some(current + new),
(Some(current), None) => Some(current),
(None, Some(new)) => Some(new),
(None, None) => None,
};
TokenUsage {
input_tokens: current_usage.input_tokens + new_usage.input_tokens,
cached_input_tokens,
output_tokens: current_usage.output_tokens + new_usage.output_tokens,
reasoning_output_tokens,
total_tokens: current_usage.total_tokens + new_usage.total_tokens,
}
}