feat: hot-swap LLM backend + media keys + codebase Q&A + scheduler cancel-by-text + TTS pre-warm

Single biggest user-facing change: you can now switch local↔cloud LLM by voice
without restarting. Plus four new packs and a perf nudge.

Shared LLM global (crates/jarvis-core/src/llm/mod.rs)
  - GLOBAL: Lazy<RwLock<Option<Arc<LlmClient>>>>. Single source of truth.
  - init_global() / current() / swap_to(LlmBackend) / current_backend_name() / parse_backend(str).
  - llm_fallback + llm_router migrated: they no longer own a separate LlmClient,
    they read from the global on each request. Hot-swap is now LIVE — change
    backend, next chat call uses the new one.
  - parse_backend accepts both english and russian aliases:
      "groq"/"cloud"/"облако"/"клауд"        → Groq
      "ollama"/"local"/"локал"/"локальный"  → Ollama

Voice + Lua LLM switcher (resources/commands/llm_switch/, 3 commands)
  - "переключись на локальный" / "перейди на оллама"  → swap to Ollama
  - "переключись на облако" / "используй грок"        → swap to Groq
  - "какой у тебя мозг" / "облако или локально"       → status query

  Underlying Lua API (registered globally):
    jarvis.llm({messages}, opts)   -- as before, now uses global client
    jarvis.llm_status()            -> "groq" | "ollama" | "none"
    jarvis.llm_switch(name)        -> name on success, nil on failure

Media keys pack (resources/commands/media_keys/, 4 commands)
  - "пауза" / "плей"           → VK_MEDIA_PLAY_PAUSE
  - "следующий трек"           → VK_MEDIA_NEXT_TRACK
  - "предыдущий трек"          → VK_MEDIA_PREV_TRACK
  - "стоп музыка"              → VK_MEDIA_STOP
  - Helper _media.ps1 uses user32.keybd_event (P/Invoke through Add-Type).
  - Works with anything that listens to global media keys: Spotify, Yandex Music,
    YouTube (focused tab), Foobar2000, Winamp, etc. No OAuth, no API keys.

Codebase Q&A pack (resources/commands/codebase_qa/, 3 commands)
  - "укажи проект <path>"             → stores path in jarvis.memory under codebase.root
  - "какой проект сейчас"             → speaks current path
  - "что делает функция X" / "найди в коде Y" / "объясни код"
    → walks the folder (depth 3, 30 files cap, 4KB per file, 50KB total),
      filters by source extensions (rs/py/ts/lua/go/...), feeds digest to LLM
      with a "senior reviewer" system prompt, speaks 3-5 sentence answer.

Scheduler cancel-by-text (crates/jarvis-core/src/scheduler.rs)
  - new pub fn find_by_text(query, limit) -> Vec<ScheduledTask>
  - new pub fn remove_by_text(query) -> usize (count removed)
  - new pure helper find_by_text_in(tasks, query, limit) for tests
  - Lua: jarvis.scheduler.{find_by_text, remove_by_text}
  - Voice: "отмени напоминание про воду" / "удали задачу про разминку"
  - 3 new unit tests (52 total in jarvis-core).

TTS pre-warm (crates/jarvis-app/src/main.rs)
  - Call jarvis_core::tts::backend() once on startup so the first speak()
    doesn't pay Piper binary discovery + voice loading cost. Cuts first-speak
    latency by ~150-300ms depending on backend.

Build: cargo build --release -p jarvis-app -p jarvis-gui both green.
Tests: 52/52 jarvis-core unit tests pass.
This commit is contained in:
Bossiara13 2026-05-15 16:15:59 +03:00
parent 6225198821
commit 5c7245012e
19 changed files with 709 additions and 34 deletions

View file

@ -2,7 +2,65 @@ mod client;
mod history;
pub use client::{
ChatMessage, ConfigError, LlmClient, LlmError,
ChatMessage, ConfigError, LlmBackend, LlmClient, LlmError,
DEFAULT_BASE_URL, DEFAULT_MODEL, ENV_BASE_URL, ENV_MODEL, ENV_TOKEN,
OLLAMA_DEFAULT_BASE_URL, OLLAMA_DEFAULT_MODEL,
};
pub use history::ConversationHistory;
use once_cell::sync::Lazy;
use parking_lot::RwLock;
use std::sync::Arc;
/// Shared mutable LLM client. Modules read via `current()`; the user can
/// hot-swap backends at runtime via `swap_to(LlmBackend::Ollama)` etc.
///
/// `None` means LLM is disabled (no GROQ_TOKEN and no Ollama reachable, or
/// `init_global()` was never called).
static GLOBAL: Lazy<RwLock<Option<Arc<LlmClient>>>> = Lazy::new(|| RwLock::new(None));
/// Initialise the global client from env vars (`JARVIS_LLM`, `GROQ_TOKEN`, ...).
/// Idempotent — replaces any previous global. Returns Ok even when the chosen
/// backend can't be probed: an Ollama server may not be running yet but we
/// stash the client anyway and let calls fail on first request.
pub fn init_global() -> Result<(), ConfigError> {
let c = LlmClient::from_env()?;
log::info!("LLM global initialised: backend={}, model={}", c.backend().name(), c.model());
*GLOBAL.write() = Some(Arc::new(c));
Ok(())
}
/// Return a clone of the active client. Cheap (Arc).
pub fn current() -> Option<Arc<LlmClient>> {
GLOBAL.read().clone()
}
/// Name of the currently active backend, or "none" if not initialised.
pub fn current_backend_name() -> &'static str {
match GLOBAL.read().as_ref().map(|c| c.backend()) {
Some(LlmBackend::Groq) => "groq",
Some(LlmBackend::Ollama) => "ollama",
None => "none",
}
}
/// Hot-swap to a different backend. Returns the new backend's name on success.
pub fn swap_to(backend: LlmBackend) -> Result<&'static str, ConfigError> {
let c = match backend {
LlmBackend::Groq => LlmClient::groq()?,
LlmBackend::Ollama => LlmClient::ollama(),
};
let name = c.backend().name();
log::info!("LLM swapped → backend={}, model={}", name, c.model());
*GLOBAL.write() = Some(Arc::new(c));
Ok(name)
}
/// Parse a backend name (case-insensitive) into the enum.
pub fn parse_backend(name: &str) -> Option<LlmBackend> {
match name.trim().to_lowercase().as_str() {
"groq" | "cloud" | "облако" | "клауд" => Some(LlmBackend::Groq),
"ollama" | "local" | "локал" | "локальн" | "локальный" => Some(LlmBackend::Ollama),
_ => None,
}
}

View file

@ -1,15 +1,18 @@
use mlua::{Lua, Table, Value};
use crate::llm::{ChatMessage, LlmClient};
use crate::llm::{self, ChatMessage};
// Lua-callable wrapper around jarvis_core::llm::LlmClient.
// Centralises the Groq plumbing so command scripts can just say:
// Lua-callable wrapper around jarvis_core::llm.
// Command scripts call:
// local reply = jarvis.llm(
// { {role='system', content='...'}, {role='user', content='...'} },
// { temperature = 0.0, max_tokens = 64 }
// )
// Returns the assistant text (string) or nil on any failure (missing token,
// network error, empty response, etc.) — callers should always check `if reply`.
// Returns assistant text (string) or nil on any failure.
//
// Backend selection is global (see `jarvis_core::llm::swap_to`). Lua scripts can:
// jarvis.llm_status() -> "groq" | "ollama" | "none"
// jarvis.llm_switch("ollama") -> name string on success, nil on failure
pub fn register(lua: &Lua, jarvis: &Table) -> mlua::Result<()> {
let chat_fn = lua.create_function(|lua, (messages, opts): (Table, Option<Table>)| {
let mut chat_messages = Vec::new();
@ -39,11 +42,18 @@ pub fn register(lua: &Lua, jarvis: &Table) -> mlua::Result<()> {
if let Ok(v) = o.get::<f32>("top_p") { top_p = v; }
}
let client = match LlmClient::from_env() {
Ok(c) => c,
Err(e) => {
log::warn!("[Lua llm] no client: {}", e);
return Ok(Value::Nil);
// Prefer the live global client (so runtime backend swap takes effect).
// Fall back to a one-shot client from env if global is not initialised.
let client = match llm::current() {
Some(c) => c,
None => {
match llm::LlmClient::from_env() {
Ok(c) => std::sync::Arc::new(c),
Err(e) => {
log::warn!("[Lua llm] no client: {}", e);
return Ok(Value::Nil);
}
}
}
};
@ -58,7 +68,35 @@ pub fn register(lua: &Lua, jarvis: &Table) -> mlua::Result<()> {
}
}
})?;
jarvis.set("llm", chat_fn)?;
// Current backend name ("groq" | "ollama" | "none").
let status_fn = lua.create_function(|_, ()| {
Ok(llm::current_backend_name().to_string())
})?;
jarvis.set("llm_status", status_fn)?;
// Hot-swap backend. Accepts "groq"/"ollama"/"local"/"cloud"/"локальный"/"облако".
// Returns the new backend name on success, nil on failure.
let switch_fn = lua.create_function(|lua, name: String| {
match llm::parse_backend(&name) {
Some(b) => match llm::swap_to(b) {
Ok(actual) => {
let s = lua.create_string(actual)?;
Ok(Value::String(s))
}
Err(e) => {
log::warn!("[Lua llm_switch] swap failed: {}", e);
Ok(Value::Nil)
}
},
None => {
log::warn!("[Lua llm_switch] unknown backend name: {}", name);
Ok(Value::Nil)
}
}
})?;
jarvis.set("llm_switch", switch_fn)?;
Ok(())
}

View file

@ -73,6 +73,23 @@ pub fn register(lua: &Lua, jarvis: &Table) -> mlua::Result<()> {
})?;
sched.set("clear", clear_fn)?;
// Remove all tasks whose name OR speak text matches `query` (substring,
// case-insensitive). Returns count removed.
let remove_by_text_fn = lua.create_function(|_, query: String| {
Ok(scheduler::remove_by_text(&query))
})?;
sched.set("remove_by_text", remove_by_text_fn)?;
// Find (don't remove) tasks matching query. Returns array of tasks.
let find_by_text_fn = lua.create_function(|lua, (query, n): (String, Option<usize>)| {
let arr = lua.create_table()?;
for (i, t) in scheduler::find_by_text(&query, n.unwrap_or(5)).iter().enumerate() {
arr.set(i + 1, task_to_lua(lua, t)?)?;
}
Ok(arr)
})?;
sched.set("find_by_text", find_by_text_fn)?;
let list_fn = lua.create_function(|lua, ()| {
let arr = lua.create_table()?;
for (i, t) in scheduler::list().iter().enumerate() {

View file

@ -293,6 +293,54 @@ pub fn list() -> Vec<ScheduledTask> {
state.store.read().tasks.clone()
}
/// Case-insensitive substring search across `name` + any `Speak` action text.
/// Returns matched tasks (clones) up to `limit`.
pub fn find_by_text(query: &str, limit: usize) -> Vec<ScheduledTask> {
find_by_text_in(&list(), query, limit)
}
/// Pure logic split out for testing.
pub(crate) fn find_by_text_in(tasks: &[ScheduledTask], query: &str, limit: usize) -> Vec<ScheduledTask> {
let nq = query.trim().to_lowercase();
if nq.is_empty() { return Vec::new(); }
let mut hits: Vec<ScheduledTask> = tasks.iter()
.filter(|t| {
if t.name.to_lowercase().contains(&nq) { return true; }
if let Action::Speak { text } = &t.action {
if text.to_lowercase().contains(&nq) { return true; }
}
false
})
.cloned()
.collect();
hits.truncate(limit.max(1));
hits
}
/// Remove all tasks whose name or speak text matches `query`. Returns count.
pub fn remove_by_text(query: &str) -> usize {
let Some(state) = STATE.get() else { return 0; };
let nq = query.trim().to_lowercase();
if nq.is_empty() { return 0; }
let removed = {
let mut store = state.store.write();
let before = store.tasks.len();
store.tasks.retain(|t| {
let name_match = t.name.to_lowercase().contains(&nq);
let text_match = if let Action::Speak { text } = &t.action {
text.to_lowercase().contains(&nq)
} else { false };
!(name_match || text_match)
});
before - store.tasks.len()
};
if removed > 0 {
persist(state);
info!("Scheduler: removed {} tasks matching '{}'", removed, query);
}
removed
}
pub fn find(id: &str) -> Option<ScheduledTask> {
let state = STATE.get()?;
state.store.read().tasks.iter().find(|t| t.id == id).cloned()
@ -446,4 +494,45 @@ mod tests {
assert_eq!(s.next_fire(None, 1000), Some(500));
assert_eq!(s.next_fire(Some(500), 1000), None);
}
fn speak_task(id: &str, name: &str, text: &str) -> ScheduledTask {
ScheduledTask {
id: id.into(),
name: name.into(),
schedule: Schedule::Interval { seconds: 60 },
action: Action::Speak { text: text.into() },
last_fired: None,
enabled: true,
created_at: 0,
}
}
#[test]
fn find_by_text_matches_name() {
let tasks = vec![
speak_task("a", "Drink water", "Попей воды."),
speak_task("b", "Stretch", "Разомнись."),
];
let hits = find_by_text_in(&tasks, "water", 5);
assert_eq!(hits.len(), 1);
assert_eq!(hits[0].id, "a");
}
#[test]
fn find_by_text_matches_speak_text_case_insensitive() {
let tasks = vec![
speak_task("a", "X", "Попей воды."),
speak_task("b", "Y", "Разомнись."),
];
let hits = find_by_text_in(&tasks, "ВОД", 5);
assert_eq!(hits.len(), 1);
assert_eq!(hits[0].id, "a");
}
#[test]
fn find_by_text_empty_query_returns_nothing() {
let tasks = vec![speak_task("a", "X", "Y")];
assert!(find_by_text_in(&tasks, "", 5).is_empty());
assert!(find_by_text_in(&tasks, " ", 5).is_empty());
}
}