Big push driven by user feedback ("делай имбу") and web research on what
voice assistants need to be the ideal:
TTS backend abstraction (P0.1)
- new module crates/jarvis-core/src/tts/{mod,sapi,piper,silero}.rs
- TtsBackend trait with SapiBackend (current PowerShell), PiperBackend
(rhasspy/piper, neural quality), SileroBackend (python subprocess)
- JARVIS_TTS env var picks (sapi|piper|silero). Auto-detect Piper if
binary + voice present in tools/piper/. Falls back to SAPI on missing.
- SpeakOpts {lang, detached, raw} replaces ad-hoc args. text_utils
sanitiser applied unless raw=true.
- llm_fallback + lua/api/tts both routed through tts::backend().
- tools/piper/install.ps1 downloads piper.exe + ru_RU-irina-medium.onnx
from rhasspy releases + huggingface. Smoke-test included.
- tools/silero/silero_tts.py helper (PyTorch); rust spawns it as subprocess.
IMBA-1 Agentic LLM router
- crates/jarvis-app/src/llm_router.rs
- When fuzzy/intent matcher fails, LLM picks the closest command from the
full registry. Returns JSON {command_id, confidence, reason}.
- Threshold-gated re-dispatch via substitute phrase. JARVIS_LLM_ROUTER=1
enables; JARVIS_LLM_ROUTER_THRESHOLD overrides 0.55 default.
- Inserted in app.rs::execute_command between "no match" and existing
llm_fallback chat fallback.
IMBA-2 Long-term memory
- crates/jarvis-core/src/long_term_memory.rs — JSON store at
APP_CONFIG_DIR/long_term_memory.json. Atomic write-through.
- remember/recall/search/forget/all/build_context API.
- Lua bindings: jarvis.memory.* (5 functions).
- llm_fallback auto-injects relevant facts (substring search of prompt)
into system message before LLM call.
- Pack resources/commands/memory_pack/ with 4 commands: remember, recall,
forget, list.
IMBA-3 Profile switching (work/game/sleep/driving/default)
- crates/jarvis-core/src/profiles.rs — JSON profiles at APP_CONFIG_DIR/profiles/
Auto-seeds 5 defaults on first run with personality + allow/deny lists +
greetings + emoji icons.
- active_profile.txt persists choice across restart.
- Lua bindings: jarvis.profile.{active,set,list,allows,active_name}.
- llm_fallback prepends profile personality to system prompt.
- Pack resources/commands/profile_switch/ with 6 voice triggers.
IMBA-4 Multimodal screenshot + vision LLM
- crates/jarvis-core/src/lua/api/vision.rs — gated on HTTP sandbox.
- jarvis.vision.screenshot() captures via PowerShell System.Drawing.
- jarvis.vision.describe(prompt?) sends base64 PNG to Groq vision model
(default llama-3.2-11b-vision-preview, override via GROQ_VISION_MODEL).
- Pack resources/commands/vision/ with 2 commands: describe + read_error.
P0.2 Continuous conversation grace window
- config::CONVERSATION_GRACE_MS = 30_000.
- app.rs: after command result, if grace_ms > 0 keep listening WITHOUT
re-wake for the grace duration. Existing CMS_WAIT_DELAY back-dated so
the existing timeout fires at start + grace_ms.
Tests: 24/24 jarvis-core unit tests pass (including 5 text_utils).
Build: cargo build --release -p jarvis-app and -p jarvis-gui both succeed
on Windows MSVC (VS 2026 Enterprise vcvars64).
Notes for setup:
- Piper voice install: pwsh tools/piper/install.ps1 (downloads ~90 MB).
- GROQ_TOKEN needed for IMBA-1 (router) and IMBA-4 (vision).
- All features are opt-in via env vars or auto-detect; existing SAPI +
fuzzy match path remains the default.
User complaint: SAPI reads every dotted acronym character-by-character with a
"точка" between letters. Unbearable on every speak. No real TTS analog
shipping yet (Silero/Coqui/Inworld in roadmap), so the immediate fix is text
preprocessing.
New module `jarvis-core::text_utils` with `sanitize_for_speech()`:
- Specific brand mapping: J.A.R.V.I.S. → Джарвис, U.S.A. → США, U.K. →
Британия, U.S. → США, S.O.S. → сос.
- Generic dotted-acronym collapse: any `[letter].[letter].[letter].(...)`
run of 3+ letters gets its dots stripped. "T.O.N." → "TON".
- URL stripping: anything starting with http/https gets replaced with the
word "ссылка". SAPI reading URLs char-by-char is unlistenable.
- Em/en-dash normalisation: — → -, « » " stripped.
- Whitespace collapse.
Wired in two places:
- `jarvis.speak(text, opts?)` Lua API runs sanitiser unless opts.raw=true.
- `jarvis-app/llm_fallback::speak_via_sapi` runs sanitiser before the SAPI
PowerShell shell-out for LLM auto-fallback replies.
5 unit tests in `text_utils::tests` covering jarvis-name collapse, generic
acronym pattern, URL stripping, dash normalisation, and "leave clean
russian sentences alone" — all green.
Real TTS upgrade (Silero v4 first cut) tracked as P0.1 in roadmap.
Symptom: jarvis-app.exe failed to start with a MessageBox saying
"Точка входа в процедуру TaskDialogIndirect не найдена в библиотеке
DLL ... jarvis-app.exe". Loader phase, no log lines ever written.
Cause: tray-icon 0.21 / winit 0.30 transitive Win32 imports include
TaskDialogIndirect from comctl32.dll, which is a Common-Controls v6
API. Without a side-by-side manifest declaring a dependency on
Microsoft.Windows.Common-Controls v6.0.0.0, Windows loads the legacy
v5 comctl32 (which does not export TaskDialogIndirect) and the
loader rejects the exe before main().
jarvis-gui does not hit this because tauri-build embeds its own
manifest as part of the Tauri compile step. jarvis-app had no
manifest at all.
Fix:
- app.manifest: Common-Controls v6 dependency, supportedOS for
Win7..Win11, PerMonitorV2 DPI awareness, UTF-8 active code page,
asInvoker execution level.
- app.manifest.rc: 3-line .rc that embeds the manifest with
CREATEPROCESS_MANIFEST_RESOURCE_ID (1) and RT_MANIFEST (24).
- build.rs: on cfg(windows), embed_resource::compile() compiles the
.rc and links the resulting .res into the exe. rerun-if-changed
on the manifest sources.
- Cargo.toml: target-windows build-dependency on embed-resource 3.
Verified: jarvis-app.exe now runs cleanly through all init steps:
commands parsed, audio init, recorder init found "Микрофон (5-
Fifine Microphone)", IPC server up on ws://127.0.0.1:9712, VAD
already flushing frames.
UTF-8 active code page in the manifest also helps russian-character
fidelity when win32 APIs are invoked from PowerShell helpers.
Iron Man icon regenerated from resources/icons/icon.png at sizes
16/24/32/48/64/128/256 via Pillow (make_ico.py committed alongside);
icon.ico is now a clean 83 KB multi-res for sharp display at any
scale instead of whatever placeholder Tauri originally emitted.
dotenvy added to workspace deps. jarvis-app/main.rs and jarvis-gui/
main.rs both walk up from current_exe() looking for dev.env (up to
5 parent levels), so the Desktop shortcut can launch the exe
directly without a wrapper .bat to pre-set GROQ_TOKEN. Falls back
to dotenvy::dotenv() (cwd lookup) if nothing found near the binary.
5 new portable Lua command packs. Bumps total from 21 to 26.
cargo test -p jarvis-core --lib commands::tests still passes 3/3.
reminders/ — set_reminder. Parses "напомни через N (секунд|минут|
часов) <body>" or English equivalents; understands one Russian word
numerals (один, две, пять, десять, пятнадцать, ...) for the count
when speech recognition returns words instead of digits. Defaults to
5 minutes if number/unit missing. Spawns a detached PowerShell that
Start-Sleeps then fires: BurntToast if installed, System.Speech
SAPI (ru-RU voice preference), and a WScript.Shell.Popup as the
guaranteed-visible last resort.
date_query/ — today / tomorrow / yesterday. Single answer.lua keyed
by command_id, computes the offset via PowerShell Get-Date.AddDays()
in ru-RU culture so the weekday/month come out in Russian, then
speaks via SAPI.
voice_type/ — type_text. Strips the trigger, drops the remainder to
clipboard via jarvis.system.clipboard.set, then synthesises Ctrl+V
through user32!keybd_event so it lands in whatever window currently
has focus. Works in any text field (note app, browser, IDE,
Telegram, etc.).
dice/ — coin_flip / roll_dice / random_number. Single roll.lua,
seeds math.random with os.time + jitter, speaks the result. "1-6"
for dice, "1-100" for random.
stopwatch/ — start / check / stop. Uses jarvis.state.get/set
(persisted via jarvis-core's settings DB) to remember the start
timestamp, computes elapsed via jarvis.context.time.timestamp,
formats as "X сек" / "X мин Y сек" / "X ч Y мин Z сек".
All new packs follow the established patterns (sandbox=full, PS-via-
exec helper where needed, USERPROFILE/SystemDrive everywhere, no
hardcoded paths).
Before: clicking "Запустить" in jarvis-gui spawned jarvis-app, which
silently exited with code 1 (or in release builds, looked like the
console window flashed and closed). The pv_recorder library returns
INVALID_ARGUMENT from pv_recorder_init when Windows Core Audio reports
zero capture endpoints (some other app holding the mic exclusively,
or all input devices disabled in mmsys.cpl). User saw no actionable
feedback.
Now: on recorder::init failure jarvis-app calls notify_mic_problem()
which (Windows-only):
- Fires a long-duration Windows toast titled "J.A.R.V.I.S.: микрофон
не найден" with a hint pointing at mmsys.cpl / Recording.
- Spawns "start ms-settings:sound" so the Sound settings page opens
automatically — user can re-enable the mic in two clicks.
Then the original app::close(1, ...) path runs to keep the same exit
behaviour the GUI's get_jarvis_app_stats poller expects.
Cargo.toml: jarvis-app now pulls winrt-notification (already in
workspace.dependencies via jarvis-core) for the toast.
Also incidentally fixed: the release-build C0000139 (entrypoint not
found) loader crash that was showing up before this change. It went
away after the workspace dep was added and the release relink ran.
Most likely the previous release exe had a stale import table from an
earlier partial rebuild; the clean relink resolves it.
Non-Windows builds get a no-op eprintln so the binary still compiles
for Linux/macOS.
GUI front-end does not subscribe to IpcEvent::LlmReply, so the LLM
auto-fallback was effectively silent — user got the OK chime but no
spoken answer. Backlog item resolved.
llm_fallback::handle now also invokes speak_via_sapi() after firing
IpcEvent::LlmReply (both for successful replies and the fallback
error message). The text goes through System.Speech.Synthesis.
SpeechSynthesizer, with a preference pass that picks the first
installed ru-RU voice if present (Microsoft Irina / Pavel etc.) so
Russian replies sound right out of the box, and falls back to the
default English voice otherwise.
PowerShell is invoked fire-and-forget (no Wait, stdout/stderr null)
so the voice loop never blocks. A long answer keeps speaking while
jarvis returns to wake-word listening; the mic may pick up some of
the speech which is the same compromise voices::play_* already makes.
A cleaner version would pause pv_recorder for the duration of the
utterance — that goes in the backlog as part of the eventual
unified IpcEvent::Speak refactor.
Toggle: set JARVIS_LLM_TTS=false to disable (e.g. on a server where
the GUI is the speaker). Non-Windows builds get a no-op stub.
LLM auto-fallback (jarvis-app/src/app.rs + jarvis-core/src/config.rs):
- When neither intent classifier nor levenshtein finds a command match,
route the utterance straight to the LLM instead of just playing the
"not found" sound. Triggers ("скажи X", "answer Y") still work and
take precedence — they short-circuit before command lookup.
- Two new config knobs:
LLM_AUTO_FALLBACK = true — master toggle
LLM_AUTO_FALLBACK_MIN_CHARS = 4 — suppress for very short utterances
so background noise doesn't burn
Groq quota
- Requires GROQ_TOKEN; if absent, behaviour is unchanged (play_not_found).
codegen/ command pack:
- Phrases: "напиши код X", "сгенерируй скрипт Y", "write code Z", etc.
- Strips the trigger from the recognized phrase, sends what remains to
Groq with a strict system prompt ("return ONLY code, no fences, no
commentary") at temperature 0.2.
- Parses the JSON content, unescapes \n / \" / \\ / \t, strips any
remaining ```lang ... ``` fences, drops the result into the clipboard
via jarvis.system.clipboard.set.
- Notifies a 120-char preview + plays ok-sound. Works for any language
the model handles (Python by default if unspecified).
- GROQ_TOKEN / GROQ_MODEL / GROQ_BASE_URL read from env at call time —
same envvars the voice-loop fallback already uses.
ocr/ command pack:
- Phrases: "прочитай экран", "что на экране", "read screen", etc.
- Captures the primary screen via System.Windows.Forms + System.Drawing
to a temp PNG, then shells to tesseract.exe (-l rus+eng for ru/ua,
-l eng for en).
- Resolves Tesseract by PATH first, then C:\Program Files\Tesseract-OCR
and the x86 install dir; if none works the user gets a friendly
"winget install UB-Mannheim.TesseractOCR" hint in a notification.
- Recognized text goes to clipboard + a 200-char preview notification.
All three features are portable (env-var resolution, no hardcoded user
paths). Command-pack total is now 11. cargo test -p jarvis-core --lib
commands::tests passes 3/3.
new module crates/jarvis-app/src/llm_fallback.rs holds an optional
LlmClient + ConversationHistory built once at startup. if GROQ_TOKEN is
unset the module logs a warning and stays disabled — voice commands keep
working as before.
both the wake-word voice path (recognize_command in app.rs) and the
gui-side text command path (process_text_command) now check whether the
recognized phrase starts with one of the configured trigger phrases
(ru: 'скажи' / 'ответь' / 'произнеси'). when it does, the remainder of
the phrase is sent to Groq and the reply is published as a new
IpcEvent::LlmReply { text } so the gui can speak it.
on api error the trailing user turn is popped from history, the russian
fallback line is sent over ipc and a 'error' voice cue plays. the loop
itself never panics.
