kardo_core/llm/
mod.rs

1//! Local LLM integration for document classification.
2//!
3//! Supports multiple backends:
4//! - Ollama (default): HTTP API to local Ollama server
5//! - GGUF (optional): Direct inference via llama.cpp
6
7pub mod deep;
8pub mod engine;
9pub mod ollama;
10pub mod sidecar;
11
12pub use engine::{classify_by_rules, Classification, ClassificationEngine, ClassificationSource};
13use serde::{Deserialize, Serialize};
14
15/// Typed errors for LLM operations.
16#[derive(Debug, thiserror::Error)]
17pub enum LlmError {
18    #[error("HTTP request failed: {0}")]
19    Http(#[from] reqwest::Error),
20    #[error("LLM backend unavailable: {0}")]
21    Unavailable(String),
22    #[error("Parse error: {0}")]
23    Parse(String),
24    #[error("Model not found: {0}")]
25    ModelNotFound(String),
26}
27
28impl From<LlmError> for String {
29    fn from(e: LlmError) -> Self {
30        e.to_string()
31    }
32}
33
34/// Available LLM backends.
35#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
36pub enum LlmBackend {
37    /// Ollama server (default, requires Ollama installed)
38    Ollama,
39    /// Disabled — use rule-based classification only
40    Disabled,
41}
42
43/// LLM generation request.
44#[derive(Debug, Clone)]
45pub struct GenerateRequest {
46    pub prompt: String,
47    pub max_tokens: u32,
48    pub temperature: f64,
49}
50
51/// LLM generation response.
52#[derive(Debug, Clone)]
53pub struct GenerateResponse {
54    pub text: String,
55    pub tokens_used: u32,
56}
57
58/// LLM availability status.
59#[derive(Debug, Clone, Serialize)]
60pub struct LlmStatus {
61    pub available: bool,
62    pub backend: String,
63    pub model: Option<String>,
64    pub error: Option<String>,
65}
kardo_core/llm/mod.rs

kardo_core/llm/
mod.rs
