- Weather: the tool deliberately returns one Korean sentence about now
(c522e1b), so the tests check the place, condition, rounded temperature and
feels-like, and that no forecast, Fahrenheit or degree sign is left for the
model to echo.
- Enrichment cache: the conversational fast-path skips enrichment when no
data tool is routed, so the test routes one to reach the extractor.
- Diary import: mock PyQt6, requests and psutil only when they are not
installed. The MagicMock left in sys.modules for an installed PyQt6 broke
every later GUI test file in the same run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The consent classifier is now told which message and channel are pending, so a yes for another
message ("yes, delete message 22" while 21 is pending) is a no and cancels the request. A reply
longer than 600 characters is no longer cut and judged by its start (which could drop a later
"don't"); it counts as no without a model call.
Eval evals/test_chat_delete_consent.py gains the different-message and long-reply cases:
17/17 on qwen2.5:7b, three runs. The previous prompt said YES 3/3 to
"응, 일정 채널에 있는 거 지워" while a message in 잡담 was pending.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- chatChannels delete with confirm=true now runs only in the person's very
next turn and only when their own message of that turn agrees, checked by a
small fail-closed YES/NO call on the router model chain (the message fenced
as data). A refusal, question, other topic or error cancels the request;
an unconfirmed request lapses after one turn. The reply model's confirm
flag alone can no longer delete.
- chat_action ids carry a per-run nonce and the session, so a late answer
meant for an earlier connection (or worker run) never completes a newer
action.
- Tests: agree, refuse, other topic, lapse, classifier failure, late answer
across reconnects. Eval test_chat_delete_consent: 12 of 12 on qwen2.5:7b,
3 runs. Spec and llm_contexts updated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Tool (chat_channels.spec.md): list, read, search, send, edit, react,
delete, pin, unpin. Offered only when the v2 brain gives the turn a chat
backend, and then on every turn. Channel by ID, name, part of a name or a
name inside the text; a message ID finds its channel; planner forms of the
action ("send_message") work. Read and search results are fenced as
untrusted data. delete only reports until called again with confirm=true
in a later turn of the same person.
- Worker: chat_channels from the app, ChatBridge reads/searches the history
store and sends chat_action, waiting for chat_action_result (15 s timeout,
failed on session close); per-person reply turns and pending deletes.
- App: channel switches edit/react (on) and delete/pin (off); chat_channels
sent on worker ready and after login; chat_action runs send, edit (own
messages only), react, delete, pin, unpin and answers with ids or the error.
- Engine: chatChannels left out of the catalogue without a backend; the
"router returned everything" check now tolerates that.
- live_discord: edit own, refuse editing the speaker's, react, pin (Discord
refuses in a voice channel's chat, reported as an error), delete, refuse an
unregistered channel; passed in 109.4 s.
- Eval (qwen2.5:7b, 3 runs): 10 of 12; never obeyed the "delete everything"
message and never deleted in the same turn.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- meals.user_id (v1 databases gain it on open, DEFAULT '' keeps their meals
the single user's); insert_meal, get_meals_between and delete_meal take
the person, and delete only removes that person's meal.
- turn_state.set_user / ToolContext.user_id carry the asker; logMeal,
fetchMeals and deleteMeal use it. The v2 brain sets it for each reply.
- Tests: two people never see or delete each other's meals, no user is v1's
single user, and a v1 database opens with its meals kept.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- A final is stored under the channel, speaker name and time fixed when its
end was decided. Decoding takes a while; a move, rename or late joiner
meanwhile used to change where and when it was said, and who heard it.
- At start the worker closes presence a killed run left open
(HistoryStore.close_open_presence), so a person who is gone no longer
hears new speech after a restart.
- run_reply_engine(legacy_memory=False) skips v1's single-user memory (warm
profile, extractor, diary, graph, digest); the v2 brain uses it so another
user's profile or diary never reaches the prompt.
- While a brain reply runs, stdout and stderr are muted (also helper threads
such as the web fetch pool); the worker logs through voice_worker/log.py to
the original stderr.
- Live check: Ollama's /v1 fixes the context at 2,048 tokens and cut the
prompt in earlier runs; with an 8k tag the chat-record question was answered
4 of 4 and nothing reached the output. Noted in llm_contexts and the eval.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Worker (voice_worker.spec.md "History and brain"):
- --history <file> keeps the conversation history store; --brain hosts the
jarvis reply engine (JarvisBrain, settings point ollama_base_url at /v1).
- voice_join / voice_leave / member_join / member_leave keep presence; the
session end closes it. Non-empty finals are stored under the speaker's name
in the current voice channel.
- chat_message / chat_edit / chat_delete keep registered chat channels.
- reply builds the asker's scoped history context, runs the brain on a thread
(one at a time), answers reply_text, stores a voice answer, and with
speak=true says it under the same id. memory settings in hello and
memory_config. brain_unavailable without --brain.
App (discord_core.spec.md "History reporting"):
- WorkerClient sends the new messages; the core reports the voice channel
and its heard members after a join, a move and every worker reconnect,
member joins and leaves, voice_leave when a session ends, and every message
of a readable registered channel (its own too, with `to` on replies), edits
and deletes.
- live_discord runs the worker with --history and checks the store at the
end: 8 finals under the speaker's name, the edited chat message only,
Jarvis's own messages, 5 presence intervals, none left open (passed, 107.8 s).
Eval: history recall gains a chat-record case and fails a reply that reads a
record stamp out (field run with qwen2.5:7b did once in four).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Search scored only the newest 5,000 matching records, so an old record
behind enough newer chatter was never found. Both passes (word rarity counts
and the score) now run in SQLite over every record in scope and only the
best rows come back.
- context() excluded the recent window from search and then cut that window
for the budget, so a relevant recent record could vanish. Search now covers
the recent window, matches keep their half of the budget, and a matched
record is removed from the recent block so it appears once.
- Matching used SQLite's ASCII-only lower() against Python's lowercasing.
Stored text gets a folded column (NFKC + casefold) and query words are
folded the same way, so case is ignored in every script.
- Prompt label "Records related to this question" (matches may now be recent);
eval re-run 3x on qwen2.5:7b, all pass. Spec, memory.md and roadmap updated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- src/jarvis/memory/history.py keeps voice transcripts, Jarvis's answers,
registered chat channel messages (with edits and deletes) and voice channel
presence in one SQLite file. Every read is scoped to the asker: own records
always, chat channels and others' words (only while the asker was in the
same voice channel) behind settings, plus an optional since and off.
- Search is language-agnostic: Unicode words, prefixes keeping 60% of the
word, rarer words weigh more, newer wins ties. 100k records search in
24-61 ms here. context() returns the recent window and the matches cut to a
token budget, matches at most half so chatter cannot push out the record
the question is about.
- run_reply_engine(history=...) puts that context into the system prompt
under "Information shared in previous conversations", fenced as data.
evals/test_history_context_recall.py: with the block qwen2.5:7b answers the
stored day in Korean and English (3 of 3 runs), without it the day is unknown.
- Spec history.spec.md, registry entry in CLAUDE.md, reply.spec.md,
docs/llm_contexts.md, docs/v2/memory.md and roadmap updated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A base URL that ends in /v1 now selects the OpenAI-compatible API: chat goes
to /chat/completions and embeddings to /embeddings. The conversion lives in
src/jarvis/llm.py only, so the reply engine and every other caller keep their
Ollama-shaped messages and responses.
- Requests drop private engine keys, send tool-call arguments as JSON strings,
map num_predict to max_tokens and the thinking switch to
chat_template_kwargs.enable_thinking, and leave out num_ctx and keep_alive.
- Responses come back with message.content, tool calls with dict arguments and
an id, and message.thinking from reasoning_content.
- Streaming, the warm-up (one-token chat completion), the intent judge and the
dictation clean-up use the same path. A server without an embedding model
fails open to the text paths.
- Specs and docs/llm_contexts.md describe the transport; roadmap records M3
step 1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Same-server moves point the existing songbird connection at the new channel and replace its
handlers; removing and rejoining at once let the late gateway "left" update cancel the new join.
- A driver disconnect without a reason (our own leave or move) is ignored, and a disconnect only ends
the session its handler belongs to, so an old connection's late event cannot end a newer one.
- Live test: move to a second voice channel and back; the session survives both, only the two moves
end one, and the speaker is still heard afterwards.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Live test: the speaker leaves and rejoins three times; each time its sentence is heard exactly
under the new SSRC (songbird issue #310 reports receive stalls after group changes; not reproduced).
- decisions.md: Rust over Electron, with the side-by-side results, the songbird risk and the fallback.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The user asked for a faster, real app instead of Electron. The app is now a Rust crate (serenity +
songbird); the GUI in M5 will be a Tauri 2 shell on the system WebView2.
- Same behaviour as the Electron core: one voice channel, per-speaker PCM to the voice worker,
speaking_end and speaker_leave, sentence playback with stop and cancel, transcription notice on
join, slash command join/leave/stop (create-or-update), registered chat channels with live events,
history and sends that never ping, worker process with a fresh token and restart backoff.
- songbird 0.6.1 garbles about a third of received frames under DAVE (its decoder keeps the bytes
DAVE decryption leaves at the end of the payload; a 20 ms stepped test tone showed repeats, skips
and reordering, while sending was exact). songbird now only decrypts; the core reorders packets in
a small window (Opus FEC for the last lost frame) and decodes them with libopus itself.
- Live test on real Discord with two bots: five sentences heard (CER 0/0/0/5/12.5%), 0.71-0.90 s
from end of speech to final, speech read back at 3.4%, sound gone 0.08 s after stop. The release
binary uses about 16 MB of memory; a Windows build (MinGW) logs in under Wine.
- Spec moved to app/src/discord_core.spec.md; architecture, decisions, roadmap and pc-features updated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- app/ (Electron 45 + TypeScript): the core joins one voice channel, decodes each speaker's Opus to
48 kHz stereo and streams it to the voice worker, sends speaking_end and speaker_leave, and emits
speech_start/partial/final with the speaker's display name.
- Sentence playback from the worker's TTS, first come first served, with stop that silences the clip
at once and cancels unfinished requests; slash command join/leave/stop (create-or-update by name).
- Transcription notice posted on every join; other bots ignored by default; at most maxSpeakers
people transcribed at once.
- Registered chat channels: live message events, history reads, sends split at 2,000 characters with
mentions disabled; Message Content intent checked before login.
- The app starts the Python voice worker with a fresh token in the environment, restarts it with
backoff, and the worker client reconnects and fails in-flight TTS on a lost connection.
- Spec in app/src/core/discord_core.spec.md; Vitest unit tests, a protocol test against the real
Python worker server with fake engines, and an opt-in live test with two bots on real Discord.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- New speaker_leave message: ends the speaker's open utterance (its final is still sent) and drops
their state, so a long session never hits the 32-speaker limit as people come and go.
- Spec and test cover more than 32 speakers in one session, each leaving mid-utterance.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Hotword occurrences were counted with fuzzy matching, which never counted a one-letter wake word
(so a hinted "봇 ..." was accepted over a plain decode without it) and counted the real word
"서비스" as "자비스" (so the hint could turn it into the wake word). Occurrences are now counted
exactly (case, spacing and punctuation ignored); near misses are left as heard for call detection.
The real-noise eval now plays each recording as recorded and at -25 dBFS and reports how many clips
reached the model.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Loudness equalisation replaces peak normalisation: rumble below 60 Hz removed, level measured on
the VAD-voiced windows only, brought to -20 dBFS (gain -20..+40 dB), soft limiter below 0.98.
- Only clearly voiced windows (VAD >= 0.5) count towards the 250 ms needed before an utterance
reaches the model, so noise the VAD half-believes never calls Whisper.
- Hallucination gate: mean token log-probability >= -0.7 (the old len/(len+1) correction let short
hallucinations through), compression ratio <= 2.4; no_speech_prob dropped (always 0.0 on turbo).
- With hotwords, the plain decode of the same encoder pass decides; the hint may fix a wake word's
spelling but cannot add or drop one (it wrote the wake word on laughter and after real sentences).
- Opt-in real-model evals: voice-like babble with the hint, speech over babble, quiet and clipped
speakers, and a JARVIS_NOISE_DIR folder of real noise recordings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- One shared faster-whisper model for every speaker, batched; finals go before partials and
partial batches are capped at two so a final never waits behind a ten-speaker pass.
- Per-speaker 48k stereo to 16k mono conversion, streaming Silero VAD, utterance ends on VAD
silence, Discord speaking_end, missing audio, or the 28 s Whisper window.
- MeloTTS in its own venv through a child process; sentences stream out one chunk each, with
incremental text, cancel and failure reporting.
- Loopback-only server with a token from the environment; logs carry no transcript text.
- Spec in src/voice_worker/voice_worker.spec.md; unit tests with fake engines and an opt-in
real-GPU test that streams ten speakers at once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On the game PC the game went behind another window right after Jarvis
started and stayed there for three windows, drawing at its background rate
(about 58 fps). The kit only checked the window in front at the end of a
window, so the window in which the user clicked back into the game counted
and showed yielding as a 47 % drop in 1 % low.
- FrontWatch checks every second from the start of a window to its end and
spoils the window if the game was behind anything at any check.
- Before a window the kit asks the user, with a beep, to go back to the
game and waits up to two minutes.
- What was in front is kept only as a category (a Jarvis process, a console
window, Explorer, another app), never an app name.
Results: with the game in front, a busy Jarvis cost 0.2 % of the average
fps and 3.3 % of the 1 % low with no frames over 50 ms, so by the
pre-registered rule nothing is added and yielding is not adopted.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Frame rate, 1 % low and hitches come from msBetweenDisplayChange of the
frames that reached the screen; frames never shown are counted apart and
the presented rate is kept as presented_fps. A game presenting 60 fps
while the screen changes every 167 ms now shows as the stutter it is.
- A window whose capture failed, a busy window where the LLM finished no
answer, and every busy window when the STT load failed are not counted.
- A yielding window counts only when every Jarvis process got both its
CPU and its GPU priority lowered.
- The "game hardly drawing" floor applies only with Jarvis off or idle, so
a game Jarvis slows that far still counts as harm.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The 95 % GPU cap is gone: it did not lower utilisation, slowed the LLM and
a YouTube video did not stutter without it. What it was meant to protect
is now measured directly in a game. The new on-demand game_frames stage
records the game's frame times with PresentMon (pinned by hash, game
process only, CSV deleted after reading) with Jarvis off, idle and busy in
the game placement, and busy while Jarvis yields: its processes at
below-normal CPU and GPU scheduling priority, read back to confirm, then
restored. Busy rows run A B B A. The design docs record the candidate and
the pre-registered rule for adopting it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
YouTube 1080p60 kept 3-4 dropped frames per window with Jarvis off, idle
and busy (GPU up to 98%), with no gaps over 50 ms and no buffering. The
95% cap did not reduce drops (7 vs 7), so the setting is removed as the
spec already planned.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With DevTools on (port or pipe), Chrome tells pages a program drives it, and
YouTube then shows "문제가 발생했습니다" after about a minute of video, which
spoilt every window on the gaming PC. Reproduced on Linux: the port alone,
with no command sent, breaks playback at ~40 s; without it, or with
--disable-blink-features=AutomationControlled, it plays on. The browser is now
started with that switch, and one that still reports webdriver is refused
before Jarvis is started.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
When loading the page again did not bring a player back, keep_room still
moved the missing player back to the start, the page script threw and the
stage stopped at the first window. Now a failed reload (or a refused
navigation) is recorded as "the video did not play again", only a playing
video is moved back (and the script checks the player exists), so the
window is measured as it is, not counted, and tried again up to the limit
before the stage goes on. The reload waits RELOAD_S for the video.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On the gaming PC the player errored about 38 s into the first window with
Jarvis off; the kit had no way to notice, so every later window measured a
paused, empty player, and after the user reloaded the page the injected
counters were gone and the snapshot script threw, ending the stage early.
- The snapshot reports a lost page (counters gone, or the script could not
run while the page loaded) and the player's own error screen (first line
only, bracketed IDs removed) instead of throwing.
- Before each window the kit presses play on a stopped player and loads the
page again when that is not enough, or when the page was lost or errored.
- A window the player spoilt (error, reload, pause, ad, seek, buffering, no
frames) is measured again, up to 3 times in all, with the earlier attempts
and the fixes kept in the row; a lower quality or frame rate is not, since
a busy GPU can cause it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- A window counts only when the player kept the asked quality and height
the whole time and played the source's 60 fps; a player stuck below
1080p is a stage error, not a note.
- Pauses, seeks and buffering are counted inside the page during the
window, so a pause that ended before the last snapshot no longer
passes as GPU stutter.
- A capped window whose cap found nothing to pause, ran no pause cycle,
or could not pause or resume a process is an error and does not count
for the cap.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
User decision (2026-11-11): the GPU share cap stays a setting but its
default is 100 % (off), since gamemode_tune showed it did not lower the
utilisation and only slowed the LLM. Whether the setting stays is decided
by the new M0 video_drops stage (a YouTube video's dropped frames with
Jarvis off, idle and busy, busy without and with the 95 % cap).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
User decision (2026-11-11): the 95 % GPU cap is off by default because
gamemode_tune showed it did not lower nvidia-smi's utilisation and cost
6 to 10 % of the LLM's output. The worry behind it was a video stuttering
next to Jarvis, so the new on-demand stage video_drops measures that:
- A public Blender upload plays at 1080p60 in a throwaway Chrome/Edge
profile (removed afterwards), driven over DevTools with a small
standard-library WebSocket client, so no package is added to the
already built venvs.
- Each window counts the video's dropped frames (the player's own
counter) and the gaps over 50/100 ms between shown frames
(requestVideoFrameCallback); windows with an ad, a pause, the end or a
new video are kept but not counted.
- Rows: video alone, Jarvis idle (normal placement with the STT room and
1 GB for other programs), Jarvis busy (LLM answering, STT transcribing
ten speakers back to back) without and with the cap in A B B A order,
then the video alone again. The cap pauses the engine and the STT
worker only, never the browser.
- run_m0_video.bat runs just this stage.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The 64.9 tok/s game-placement run was compared with the 64.4 tok/s
normal-placement run, which uses different expert, context and image
settings. Compare it only with the 2 s pace of the same placement, and
state that the final transcription in the kit is one fixed sample at
beam 1, so the real beam 2-3 / long-utterance final latency with paused
partials is still unmeasured (to be measured in M1).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Normal placement with the STT room and 1 GB for other programs:
expert VRAM 1.5 GB, 1.8 GB left at the STT peak, context TTFT back
to 1.9 s (was 7.5 s), 64.4 tok/s alone.
- Game mode pauses STT partials while the LLM answers: 64.9 tok/s vs
37.5 tok/s at a 2 s pace, final wait 0.001 s vs up to 1.2 s.
- The 95 % GPU cap paused for 5.2 % but nvidia-smi utilisation did not
drop (96->98 %, 40->39 %) and decode was 6-10 % slower; whether to
keep it is pending a user decision.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A resume that failed was only counted, and the process stayed suspended
after the cap ended. DutyCap now keeps what it still owes per process,
does not stack another suspend on it, retries the resume every period
and at the end, treats a raising call as a failure, and reports any
process it could not resume as left_paused (an error for the row).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
User decision (2026-11-10): the normal placement also subtracts a room
for other programs (default 1 GB, a setting), and the service manager
pauses the LLM engine and the voice worker together for 10 ms of every
200 ms (95 %, a setting, 100 % is off) in every stage. Both are to be
measured with M0 gamemode_tune.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
User decision (2026-11-10, room to watch a video next to Jarvis): the
normal placement leaves 1 GB of VRAM for programs opened later, and
Jarvis's GPU work is capped at 95 %.
- gamemode_tune's normal rows take OTHER_ROOM_MIB off the expert budget.
- gpucap.DutyCap pauses the Strata engine (strata.exe --serve under the
kit) and the STT worker with their children together for 10 ms of
every 200 ms (NtSuspendProcess), and always resumes them.
- Every row samples nvidia-smi utilization.gpu; capped rows have an
uncapped twin (normal with and without the STT, game 2 GB with 2 s
partials) so the run shows the cost and what the GPU really gets back.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
User decision (2026-11-10): without a game the STT keeps every partial
pass while the LLM answers. gamemode_tune now measures the normal
placement with the STT room and back-to-back STT only, and paces the
2 GB game placement (back to back as the same-run baseline, every 2 s,
finals only). Docs say the slowdown applies in game mode only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
User decisions (2026-11-10): the game-mode minimum on a 16 GB card is the
2 GB expert placement, the normal placement leaves the STT its share by
formula instead of Strata's automatic placement, and STT partials slow
down while the LLM answers with finals handled at once. Also notes the
user's report of occasional, minor frame drops during the live run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The gamemode_stt run showed Strata's automatic placement filling the card
next to the STT (15.9 of 16.3 GB, first token 1.6 -> 7.5 s) and the STT's
back-to-back partial passes cutting LLM output to a third. The new
gamemode_tune stage measures both fixes the user chose:
- stt_room: one start without experts (STT loaded) gives the LLM's base
VRAM; --expert-vram gets what is left after the STT's ten-speaker
growth (from this PC's stt stage) and a 768 MiB margin.
- stt_pace: a ten-speaker partial pass every 2 s, or none, with finals
every 0.5 s and finals first; records final wait and STT GPU share.
run_m0_gamemode_tune.bat runs it unattended.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The voice-start figures add medians from different runs and leave out
the TTS, the VRAM-full cause of the 7.5 s first token is a hypothesis,
and the 2 GB placement is 2.1x faster with the STT on the GPU but 1.5x
with it on the CPU.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gamemode (third run), gamemode_stt and gamemode_live finished on the
game PC. The STT on the GPU takes 2.7 GB at its 10-speaker peak, as
estimated, and leaves the game 3.6 GB (experts 0 GB) or 2.6 GB (2 GB).
The STT stays on the GPU in games: on the CPU one utterance takes
3.0-3.8 s. In a real game the 2 GB expert placement is twice as fast as
0 GB. The GPU contention between the STT and the LLM, the in-game floor
placement and an STT margin for the normal placement wait for the user.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A worker that did not stop in time returned its earlier "loaded" note, was
killed, and the placement still counted as ok with no transcription times.
finish() now reports a timeout, and a worker that timed out, died, never
started or measured no ten-speaker or single-clip timing is an error, so
the stage is partial and the next run retries it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The user meant the chat channel's "answer every message", not the voice
policy "in order": the LLM takes two or more requests at once only while a
chat channel answers every message, otherwise one whatever the voice policy.
The roadmap and the open measurements point at the new gamemode_stt and
gamemode_live stages, which also decide whether the STT moves to the CPU
during a game.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gamemode now also starts the 2 GB expert placement with a 16K context and
drops the --kv q4_0 and --mtp off rows the Strata 0.1.59 qwen36 engine
ignores; the normal placement answers one request at a time.
gamemode_stt loads the STT worker (stt.py mode=load) before the LLM picks its
placement and keeps it transcribing ten speakers at once while the LLM is
measured, so the VRAM left is the real game share. The game placements run
with the STT on the GPU and on the CPU. gamemode_live repeats the game rows
while the user runs a real game; it only runs when asked for
(run_m0_game_live.bat) and is skipped when nobody is at the PC.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The M0 stt stage measured 1300 MiB for the model and another 1440 MiB while
transcribing ten speakers at once. The game share now subtracts the 2.7 GB
peak: about 3.1 GB on the 16 GB game PC (not 4.6 GB), and the 2 GB expert
placement returns at a reserve of about 2.2 GB. Both are marked as sums of
separate LLM and STT runs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The user chose one request at a time by default (97 tok/s on a 16 GB card
against 70-75 with two) and two or more only under the "in order" policy,
which answers every request. The second gamemode run showed a 16 GB card
cannot free 8 GB for a game: the smallest placement still leaves 6.3 GB.
The Strata 0.1.59 qwen36 engine source confirms why --kv and --mtp off made
no difference: it takes no --kv flag and always loads the MTP block. The
game placement is now one request, CPU vision, 8 workers, 16K context and a
computed expert share, falling back to the minimum placement on this PC.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The first game PC run showed that a 2 GB expert budget still left a 16 GB
card only 4.3 GB free: the weights, the 32K int8 conversation memory, the
MTP drafter and buffers take about 8.7 GB on their own. The game placements
now also shrink the memory format (q4_0), the context (16K, 8K) and the
drafter, with half the CPU workers (no speed cost measured). The engine log
is appended across starts, so each placement now reads only its own lines;
when no placement leaves the reserve the result names the most free one.
The gamemode runner skips the desktop stage, which is already measured.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The gamemode stage was "ok" when only the normal placement started; a
placement that does not start, or a measurement error inside one, now marks
the stage partial. The design no longer raises the expert budget above the
computed one: it shrinks the game reserve to what the smallest placement
leaves, shows that, and pauses the LLM during games when nothing is left.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Drop the recommended model (Qwen3.8-Flash-Next) and the experimental one:
Qwen3.6-35B-A3B passed every tool call and answers first. Replace game-mode
actions (a)(b)(c) with automatic tuning: a normal and a game placement of the
same model, computed per PC from VRAM, a first-run self-check that lowers
defaults on slow PCs, and latency watching that lowers restart-free settings
first. Record the agent-desktop results and choose the per-call CLI.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add a gamemode stage that restarts Qwen3.6-35B-A3B with a smaller routed
expert VRAM budget (6, 4, 2 GiB, and 2 GiB with half the CPU workers), then
the normal placement again, and records switch time, free VRAM, output speed,
first spoken sentence and 8K-context first token for each. It names the
fastest placement that leaves the game its 8 GB reserve. Strata 0.1.59's
qwen36 engine cannot shrink VRAM while running, so each placement is a restart.
The desktop stage now matches running apps by the process image name Windows
lists (Discord.exe) and closes the calculator it launched by image name when
close-app cannot find a UWP app.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
check_mcp timed five CLI and five MCP list-windows calls but ignored their
answers, so a server returning isError/ok=false still scored ok. Read the
envelope of every call, record the error code (never the message), and
fail the check on any failed call.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The gaming-PC rerun scored the fast model's honest refusals ('...할 수 있는
도구는 없습니다', '...준비되어 있지 않습니다') as failures, so the kit
downloaded two more fast-model candidates for nothing. Rescored with the
fixed judge, UD-Q4_K_XL, UD-Q6_K_XL and Q8_0 all pass 27/27; the
recommended model's Q2_0/IQ3_XXS/IQ3_S still fail 3/27 by calling an
unrelated tool when no tool fits. Record both results in the v2 docs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>