offline_intelligence/api/
title_api.rs

1// Title generation API: summarize first prompt into 1-5 word chat title via model inference
2// Uses direct LLM integration with shared state for 1-hop architecture
3use axum::{
4    extract::{State, Json},
5    http::StatusCode,
6};
7use serde::{Deserialize, Serialize};
8use tracing::info;
9
10use crate::shared_state::UnifiedAppState;
11
12#[derive(Debug, Deserialize)]
13pub struct GenerateTitleRequest {
14    pub prompt: String,
15    #[serde(default = "default_max_tokens")]
16    pub max_tokens: u32,
17}
18
19fn default_max_tokens() -> u32 {
20    20
21}
22
23#[derive(Debug, Serialize)]
24pub struct GenerateTitleResponse {
25    pub title: String,
26}
27
28#[derive(Debug, Serialize)]
29pub struct ErrorResponse {
30    pub error: String,
31}
32
33/// Generate a concise chat title (1-5 words) from a user prompt using the LLM.
34/// Uses temperature 0.3 for consistent results, caps output at 20 tokens
35pub async fn generate_title(
36    State(state): State<UnifiedAppState>,
37    Json(req): Json<GenerateTitleRequest>,
38) -> Result<Json<GenerateTitleResponse>, (StatusCode, Json<ErrorResponse>)> {
39    info!("Generating title for prompt (length: {} chars)", req.prompt.len());
40
41    if req.prompt.is_empty() {
42        return Err((
43            StatusCode::BAD_REQUEST,
44            Json(ErrorResponse {
45                error: "Prompt cannot be empty".to_string(),
46            }),
47        ));
48    }
49
50    let llm_worker = state.llm_worker.clone();
51
52    // Improved instruction: prioritize clarity and let LLM naturally generate 1-5 word titles
53    let title_instruction = format!(
54        "User prompt: {}\n\n\
55         Create a short, meaningful chat title using 1-5 words maximum that captures the essence of this prompt.",
56        req.prompt
57    );
58
59    // Generate title using LLM worker directly (1-hop architecture)
60    match llm_worker.generate_title(&title_instruction, req.max_tokens.min(20)).await {
61        Ok(title) => {
62            let word_count = title.split_whitespace().count();
63            info!("Generated title: '{}' ({} words)", title, word_count);
64            Ok(Json(GenerateTitleResponse { title }))
65        }
66        Err(e) => {
67            info!("Title generation failed: {}", e);
68            Err((
69                StatusCode::INTERNAL_SERVER_ERROR,
70                Json(ErrorResponse {
71                    error: format!("Title generation failed: {}", e),
72                }),
73            ))
74        }
75    }
76}
offline_intelligence/api/title_api.rs

offline_intelligence/api/
title_api.rs
