use crate::connection::DbHandle;
use async_trait::async_trait;
use chrono::Utc;
use klieo_core::error::MemoryError;
use klieo_core::ids::ThreadId;
use klieo_core::llm::{Message, Role, ToolCall};
use klieo_core::memory::ShortTermMemory;
pub struct SqliteShortTerm {
db: DbHandle,
}
impl SqliteShortTerm {
pub(crate) fn new(db: DbHandle) -> Self {
Self { db }
}
}
fn role_to_str(r: Role) -> &'static str {
match r {
Role::System => "system",
Role::User => "user",
Role::Assistant => "assistant",
Role::Tool => "tool",
_ => "user",
}
}
fn role_from_str(s: &str) -> Result<Role, MemoryError> {
match s {
"system" => Ok(Role::System),
"user" => Ok(Role::User),
"assistant" => Ok(Role::Assistant),
"tool" => Ok(Role::Tool),
other => Err(MemoryError::Serialization(format!("unknown role: {other}"))),
}
}
#[async_trait]
impl ShortTermMemory for SqliteShortTerm {
async fn append(&self, thread: ThreadId, msg: Message) -> Result<(), MemoryError> {
let tool_calls_json = serde_json::to_string(&msg.tool_calls)
.map_err(|e| MemoryError::Serialization(e.to_string()))?;
let role = role_to_str(msg.role);
let now = Utc::now().to_rfc3339();
self.db
.execute(move |conn| {
let tx = conn.transaction()?;
let next_seq: i64 = tx
.query_row(
"SELECT COALESCE(MAX(seq), 0) + 1 FROM short_term_messages WHERE thread_id = ?1",
rusqlite::params![&thread.0],
|r| r.get(0),
)?;
tx.execute(
"INSERT INTO short_term_messages (thread_id, seq, role, content, tool_calls, tool_call_id, created_at) VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7)",
rusqlite::params![
&thread.0,
next_seq,
role,
&msg.content,
&tool_calls_json,
&msg.tool_call_id,
&now,
],
)?;
tx.commit()?;
Ok(())
})
.await
}
async fn load(&self, thread: ThreadId, max_tokens: usize) -> Result<Vec<Message>, MemoryError> {
let rows: Vec<(String, String, String, Option<String>)> = self
.db
.execute(move |conn| {
let mut stmt = conn.prepare(
"SELECT role, content, tool_calls, tool_call_id FROM short_term_messages WHERE thread_id = ?1 ORDER BY seq ASC",
)?;
let iter = stmt.query_map(rusqlite::params![&thread.0], |row| {
Ok((
row.get::<_, String>(0)?,
row.get::<_, String>(1)?,
row.get::<_, String>(2)?,
row.get::<_, Option<String>>(3)?,
))
})?;
iter.collect::<Result<Vec<_>, _>>()
})
.await?;
let mut messages: Vec<Message> = rows
.into_iter()
.map(|(role, content, tool_calls_json, tool_call_id)| {
let tool_calls: Vec<ToolCall> = serde_json::from_str(&tool_calls_json)
.map_err(|e| MemoryError::Serialization(e.to_string()))?;
Ok(Message {
role: role_from_str(&role)?,
content,
tool_calls,
tool_call_id,
})
})
.collect::<Result<Vec<_>, MemoryError>>()?;
let total: usize = messages.iter().map(|m| (m.content.len() / 4).max(1)).sum();
if total > max_tokens {
let mut kept = 0usize;
let mut keep_from = messages.len();
for (idx, msg) in messages.iter().enumerate().rev() {
let cost = (msg.content.len() / 4).max(1);
if kept + cost > max_tokens {
break;
}
kept += cost;
keep_from = idx;
}
messages = messages.split_off(keep_from);
}
Ok(messages)
}
async fn clear(&self, thread: ThreadId) -> Result<(), MemoryError> {
self.db
.execute(move |conn| {
conn.execute(
"DELETE FROM short_term_messages WHERE thread_id = ?1",
rusqlite::params![&thread.0],
)?;
Ok(())
})
.await
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::connection::DbHandle;
fn user(text: &str) -> Message {
Message {
role: Role::User,
content: text.into(),
tool_calls: vec![],
tool_call_id: None,
}
}
async fn fresh() -> SqliteShortTerm {
let db = DbHandle::open(":memory:").await.unwrap();
SqliteShortTerm::new(db)
}
#[tokio::test]
async fn append_then_load_round_trips() {
let m = fresh().await;
let t = ThreadId::new("t1");
m.append(t.clone(), user("hello")).await.unwrap();
m.append(t.clone(), user("world")).await.unwrap();
let loaded = m.load(t, 10_000).await.unwrap();
assert_eq!(loaded.len(), 2);
assert_eq!(loaded[0].content, "hello");
assert_eq!(loaded[1].content, "world");
}
#[tokio::test]
async fn load_truncates_to_token_budget() {
let m = fresh().await;
let t = ThreadId::new("t1");
for i in 0..20 {
m.append(
t.clone(),
user(&format!("msg-{i:03}-padding-padding-padding")),
)
.await
.unwrap();
}
let loaded = m.load(t, 30).await.unwrap();
assert!(
loaded.len() <= 4 && !loaded.is_empty(),
"expected ~3 messages, got {}",
loaded.len()
);
assert!(
loaded.last().unwrap().content.contains("msg-019"),
"newest message must survive truncation"
);
}
#[tokio::test]
async fn clear_removes_thread() {
let m = fresh().await;
let t = ThreadId::new("t1");
m.append(t.clone(), user("hello")).await.unwrap();
m.clear(t.clone()).await.unwrap();
let loaded = m.load(t, 10_000).await.unwrap();
assert!(loaded.is_empty());
}
#[tokio::test]
async fn threads_are_isolated() {
let m = fresh().await;
m.append(ThreadId::new("a"), user("a-msg")).await.unwrap();
m.append(ThreadId::new("b"), user("b-msg")).await.unwrap();
let a = m.load(ThreadId::new("a"), 10_000).await.unwrap();
let b = m.load(ThreadId::new("b"), 10_000).await.unwrap();
assert_eq!(a.len(), 1);
assert_eq!(b.len(), 1);
assert_eq!(a[0].content, "a-msg");
assert_eq!(b[0].content, "b-msg");
}
#[tokio::test]
async fn role_round_trip_for_all_variants() {
let m = fresh().await;
let t = ThreadId::new("r");
for role in [Role::System, Role::User, Role::Assistant, Role::Tool] {
m.append(
t.clone(),
Message {
role,
content: "x".into(),
tool_calls: vec![],
tool_call_id: None,
},
)
.await
.unwrap();
}
let loaded = m.load(t, 10_000).await.unwrap();
assert_eq!(loaded.len(), 4);
assert_eq!(loaded[0].role, Role::System);
assert_eq!(loaded[3].role, Role::Tool);
}
}