No browser caller exists, so no page needs to read the daemon. The wildcard let any page read /gpu-permission and /operator-state.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
GET /smoke ran docker run --gpus all and started the resident worker outside the gate, reachable by any page or LAN host while the household's switch was off. Its only callers were scripts/smoke-vili.mjs and the Eve Smoke button, so the route, the script, the button and the advertised endpoints go. refreshBackendStatus lost its includeKimodoHelp GPU branch and every service docker start: a GET now only asks. The worker starts only inside a job that holds the slot.
--host takes the addresses to listen on, one server each (Raven: 127.0.0.1,10.77.0.4); a wildcard refuses to start. Callers are the tray and health (loopback) and WireGuard peers (Starfire, services).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A real daemon on stub wsl.exe and nvidia-smi.exe runs one job held open by the test; requests during
it are 503 source slot, the slot frees at the job's end, and the next job is admitted.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
POST /gpu-permission and POST /motion/generate now require Content-Type application/json and
refuse a request whose Origin is present and not the daemon's own loopback origin, 403 cross-site,
before the body is read. The blanket OPTIONS handler that approved any preflight is deleted.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
String(null) and String(undefined) already fail the integer test, so the
?? "" guard decided nothing; the Stryker mutant of it survived.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
vili-gpu-tray.ps1 is one NotifyIcon in the interactive session: it lowers
GET /gpu-permission to an icon and tooltip and POSTs an absolute value on click.
It holds no state, reads no process list and writes no file.
install-vili-tray-task.ps1 registers only GameCult\ViliTray at logon.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gpu-tenancy.mjs judges permission, then the one slot, then one host free-VRAM
read per admission. The permission is a gamecult.vili.gpu_permission.v1
document in vili.service.cc, set only from loopback at POST /gpu-permission;
absent means not allowed. The daemon holds one CultCache instance. Tests pin
the gate with fake ports and the daemon edge over HTTP, including the slot
wiring through acquire() (finding s2 slot-wiring-unpinned).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The household's use is no longer detected. gpu-probe.mjs, the PowerShell probe
child, the hysteresis/3D/full-screen/own-VRAM rules, three env variables and the
README paragraph go; the gate keeps only the single slot and the placement
refusal until the permission lands in the next commit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
acquire() judges at the slot: household signals first (admission), then a held slot (slot);
nothing queues, so a request can no longer wait on the worker lock past a changed gate. While
Vili's job holds the slot, and until free VRAM recovers to its pre-job level (capped), low free
VRAM is not the household's; full-screen and 3D always count. A sample with no resolved 3D
counters is a probe fault and the gate fails closed. The placement 503 carries no gate status.
A malformed VILI_GPU_* override refuses to start, naming the variable.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The probe's sample plumbing leaves the gate module (the gate never kills; the probe may kill
its own hung child). The counter path is one exported constant passed to the script as
-Util3dPath, the script reports how many 3D engine counters resolved, and a child that is
alive but silent for 3 intervals is killed and respawned.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Select-Object -First 1 closes the pipe early and leaves LASTEXITCODE -1, so
every iteration threw and the probe wrote no samples (found on Raven).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
worker_error_body is the worker's one error wire shape; GpuBusy.code
reaches the daemon as a field. GpuBusy is documented as placement only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Refuse a new job with 503 + retry-after 60 while the GPU is in use (source
admission); a placement refusal from the encoder child also answers 503
(source placement). Publish the gate as operator_state.gpu. A running job
is never touched.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
One owner for "may heavy GPU work start": numbers-only samples (full-screen
state, sustained 3D use, host free VRAM) in, { inUse, since, checkedAt,
probeOk } out; fail closed at start and when the probe goes stale.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Replaces the 2026-06-12 timing paragraph with the cache, the child, and the numbers
measured on Raven on 2026-10-22. Adds the tests hand mutation showed missing: the
exact 2 GiB boundary, a child answer with too few vectors, and the 4096 dimension.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The worker builds CachedTextEncoder and passes it to load_model, so Kimodo's own
encoder selection (and the CPU encoder behind TEXT_ENCODER_DEVICE) is never built.
generate() returns the per-request encoder record (identity, hits, misses, seconds,
vector hashes) and the worker identity; /health adds the identity and stored count.
startKimodoWorker drops TEXT_ENCODER_DEVICE and mounts the scripts directory, the
cultcache-py source and .vili/embeddings, with the Kimodo commit and image id read
once per worker start. The encode clock is perf_counter: time.time went backwards
under WSL during a probe.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
text_encoding.py holds the CultCache embedding store (gamecult.vili.text_embedding.v0
through cultcache-py), the encoder identity and key, plan_placement with its typed
gpu-busy refusal, and CachedTextEncoder, which sends a request's misses to one child
call. kimodo-text-encode.py is that child: it assembles Kimodo's LLM2Vec encoder in
bf16 with its own loader (operator ruling encoder-own-loader; peft's offload path
raises TypeError in Kimodo's loader), encodes each text alone, writes a MessagePack
answer and exits. GPU reserve 3 GiB: four cold runs on Raven's 16 GB card produced
identical vectors with no CUDA error. The worker and daemon are not wired yet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
It published idunn.daemon_health to 10.77.0.2:17870 (the retired Windows Idunn)
every 15 s and failed each time; Idunn on Yggdrasil binds no RUDP health ingress.
Deletes idunn-rudp.cjs, the publisher and its status, the --idunn-* args in the
daemon and the three launcher scripts, operator_state slot 4, and the
cultnet-ts and direct @msgpack/msgpack dependencies. transportProfile now states
local-cultcache-only. All commits build (node --check passes); not yet run on Raven.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
REC-120:
- Removed the unused `readFile` import (`readFile(` had zero call
sites in the file).
- Removed the never-read `backend-status.json` initializer block; the
`backendStatus` in-memory object it snapshotted is still very much
alive (read at lines 287, 378, 489, 504, 509, 515, 768) and untouched
here.
- Also removed the `existsSync` import, which the initializer block
was its only caller for; left in place it would have become a second
dead import introduced by this same cut.
Kill attempts that failed: none.
- `rg -n "readFile\(" scripts/vili-daemon.mjs` -> no matches (repeated
the census command against current HEAD).
- `rg -n -F --glob '!scripts/vili-daemon.mjs' 'backend-status.json' .`
-> no matches in this repo; cross-repo `rg -n 'backend-status\.json'`
over F:/Projects only turns up Odin's
docs/transport-shortcut-inventory.md, which documents the shortcut
rather than reading the file.
Checks run: `node --check scripts/vili-daemon.mjs` -> syntax OK. No
package.json/test suite exists in this repo to run beyond that.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
