Improve chat rendering and generation summaries

This commit is contained in:
Georg Bauer
2026-07-27 14:04:14 +02:00
parent 6a541b51f9
commit 8a08cd7b52
10 changed files with 152 additions and 13 deletions

View File

@@ -55,6 +55,7 @@ pub(crate) struct ChatMessage {
pub(super) system: bool,
pub(super) compaction: bool,
pub(super) compaction_tail_start: Option<i32>,
pub(super) generation_stats: Option<GenerationStats>,
pub(super) reasoning: Option<String>,
pub(super) reasoning_complete: bool,
pub(super) reasoning_open: bool,
@@ -66,6 +67,14 @@ pub(crate) struct ChatMessage {
pub(super) a2ui_open_urls: Vec<String>,
}
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub(super) struct GenerationStats {
pub(super) duration_ms: u64,
pub(super) input_tokens: u32,
pub(super) cached_tokens: u32,
pub(super) output_tokens: u32,
}
const TOOL_PROTOCOL_CORRECTION: &str = "Tool protocol correction (no tool was executed): you emitted a complete tool call inside private reasoning. End reasoning, then emit the tool call again as assistant content using the required syntax; do not merely discuss it.";
const EMPTY_RESPONSE_CORRECTION: &str = "Protocol correction: your previous response was empty. Continue the task now with either a valid tool call or a final answer.";
@@ -122,6 +131,27 @@ fn correction_already_sent(conversation: &[ChatMessage], correction: &str) -> bo
impl From<StoredMessage> for ChatMessage {
fn from(message: StoredMessage) -> Self {
let generation_stats = match (
message.generation_duration_ms,
message.input_tokens,
message.cached_tokens,
message.output_tokens,
) {
(Some(duration_ms), Some(input_tokens), Some(cached_tokens), Some(output_tokens))
if duration_ms >= 0
&& input_tokens >= 0
&& cached_tokens >= 0
&& output_tokens >= 0 =>
{
Some(GenerationStats {
duration_ms: duration_ms as u64,
input_tokens: input_tokens as u32,
cached_tokens: cached_tokens as u32,
output_tokens: output_tokens as u32,
})
}
_ => None,
};
let mut message = Self {
id: message.id,
user: message.user,
@@ -129,6 +159,7 @@ impl From<StoredMessage> for ChatMessage {
system: message.system,
compaction: message.compaction,
compaction_tail_start: message.compaction_tail_start,
generation_stats,
reasoning: message.reasoning,
reasoning_complete: message.reasoning_complete,
reasoning_open: false,
@@ -661,6 +692,30 @@ impl App {
context_changed = true;
}
Ok(GenerationEvent::Finished(result)) => {
if let Ok(output) = &result {
let stats = GenerationStats {
duration_ms: u64::try_from(active.started_at.elapsed().as_millis())
.unwrap_or(u64::MAX),
input_tokens: output.prompt_tokens,
cached_tokens: output.cached_tokens,
output_tokens: output.completion_tokens,
};
if let Some(message) = self.conversation.last_mut() {
message.generation_stats = Some(stats);
if let Some(database) = &mut self.database
&& let Err(error) = database.update_message_generation_stats(
message.id,
stats.duration_ms,
stats.input_tokens,
stats.cached_tokens,
stats.output_tokens,
)
{
self.error =
Some(format!("Could not save generation summary: {error}"));
}
}
}
match result {
Ok(_) if self.stop_requested => {
self.generating = false;
@@ -1605,6 +1660,7 @@ mod tests {
system: false,
compaction: false,
compaction_tail_start: None,
generation_stats: None,
reasoning: reasoning.map(str::to_owned),
reasoning_complete: false,
reasoning_open: false,
@@ -1664,6 +1720,7 @@ mod tests {
system: false,
compaction: false,
compaction_tail_start: None,
generation_stats: None,
reasoning: None,
reasoning_complete: true,
reasoning_open: false,
@@ -1729,6 +1786,7 @@ mod tests {
system: compaction,
compaction,
compaction_tail_start: tail,
generation_stats: None,
reasoning: None,
reasoning_complete: true,
reasoning_open: false,
@@ -1765,6 +1823,7 @@ mod tests {
system,
compaction,
compaction_tail_start: None,
generation_stats: None,
reasoning: None,
reasoning_complete: true,
reasoning_open: false,