- Change AI_LLM_BASE_URL default from omniroute.imrnes.team to 9router.asepharyana.my.id
- Update AI_LLM_MODEL default from 'text' to 'claude-opus-5' (bare model name
compatible with 9router/OpenAI-compatible router)
- Update .env.example and inline comments to reflect 9router
- discord-gateway config now matches backend (which already uses 9router)
Batch race guard balikin {ok:true, rows:[]} tanpa sinyal saat semua target
masih upload-pending -> processor klasifikasi semua incomplete -> fanout ke
individual queue -> di situ requeue + reschedule 250ms -> balik ke batch:
hot loop ~300ms sepanjang upload (10 siklus/3 dtk di log prod 08:13).
Fix: worker batch kini return uploadPendingIds eksplisit; classifier pure
baru (partitionBatchOutcome) partisi completed/upload_pending/incomplete/
parse_failed/api_failed; target upload-pending DEFERRED dengan poll backoff
linear (AI_ANALYSIS_UPLOAD_POLL_MS 1500 base, cap AI_ANALYSIS_MAX_UPLOAD_POLL_MS
8000), tidak pernah masuk fanout; tail shouldScheduleNext tak menimpa defer.
Test: tests/batchOutcomeClassifier.test.ts (8 kasus, pure tanpa DB/Piscina).
Analisis pertama tetap berkonteks (chat history) demi akurasi, tapi
verdict clean non-actionable (conf>=0.85) juga ditulis di bare key
tanpa konteks. Repeat teks sama di channel lain -> exact cache HIT,
bukan LLM call baru. Guard sama dgn read path; dedupe LRU per proses;
bare row tanpa embedding (tier semantic sudah global).
- Fase-1 exact-cache lookup: N query serial -> SATU query ANY($1::text[])
- Global reuse utk bare key legacy, HANYA verdict non-actionable
(clean/flagless/action=none, conf>=0.85, umur<=72h) — flagged/warn
tetap context-scoped
- Semantic cache dua-band: clean band 0.92 default, actionable tetap
0.97; di antara band -> LLM (fail-open ke akurasi)
- hit_count kini di-increment (bulk UPDATE per batch) -> hit-rate terukur
- Cache hasil wikipediaSearch di Redis (6h, hanya hasil non-kosong)
- Memoize fetchUrlSafely utk type=text (LRU 30m + in-flight dedupe)
- makeImageCacheKey strip query CDN Discord (?ex/is/hm, format/width)
-> attachment sama = satu key vision, skip re-download+re-vision
Spec: .hermes/plans/2026-08-24-ai-analysis-cache-optimization.md
Tests: +33 (cacheGuards, discordImageKeyNormalize, cacheBatchLookup)
- pickBatchWithinBudget: stop di overflow pertama (break), bukan skip —
batch tetap prefix kronologis tanpa gap analisis di tengah timeline.
Diekstrak ke batchBudget.ts (pure, estimator di-inject) + regression test.
- callModerationLLM: param opsional maxTokens; text/media caller menghitung
ceiling dari estimasi prompt (floor 2048, cap 16384) — batch kecil tak
lagi reserve window completion 16k.
- getPending/IncompleteMessagesByConversation: sort hasil UPDATE..RETURNING
by created_at ASC — Postgres tak menjamin urutan, konsumen (anchor konteks
messages[0], prefix batch) bergantung pada urutan kronologis.
- normalizeStoredStatus(): exact-hash & semantic (Qdrant/PG) cache reader
sebelumnya menipiskan 'warn' jadi 'flagged'/'clean' (type narrowing
legacy clean|flagged) — merusak gating auto-delete & label dashboard.
Kini status tersimpan dipertahankan penuh (clean/warn/flagged).
- prompts: hapus referensi <user_history> yang tak pernah di-inject,
SearXNG -> Wikipedia (sudah migrasi), referensi section yang tak ada,
typo 'secifik', dan baris list rusak '|-'.
- moderationBuilders: buang dead code buildUserProfilesBlock/
buildUserProfileRef/UserProfileEntry/buildUserHistoryXml (tanpa caller
produksi sejak context minimization) + test-nya.
- test baru: tests/storedStatusNormalization.test.ts (regresi warn).
upload.asepharyana.my.id redirect ke file mp3 tunggal; generic extractor
yt-dlp expose format ID '0' sehingga '-f bestaudio' gagal 'Requested
format is not available'. Chain bestaudio[ext=m4a]/bestaudio/best tetap
dapat m4a di YouTube dan jatuh ke 'best' untuk direct file.
- Recordings: custom RecordingAudioPlayer (play/pause, buffering spinner,
click-to-seek, time label, eq bars, single-playback antar kartu) +
highlight kartu now-playing
- Media: thumbnail di disc hero + queue row, equalizer saat playing,
badge 'up next', label Paused vs Now playing
- MiniPlayer global di AppFrame (fixed bottom-right, hidden on /media)
menggantikan use-media-player.tsx dead provider (dihapus)
- Voice: mic level meter live (AnalyserNode RMS) + slider mic/listen volume
tsc reported 'Property some does not exist on type {}' on doc.tags
because Drizzle jsonb inference returns a generic object. Cast explicitly.
This unblocks the GMW GitHub Actions deploy (test job).
The Drizzle schema used pgText('tags').array() which emits a Postgres
text[] column, but the migration defines tags as jsonb. The mismatch caused
INSERT/LIST on materi_documents to throw INTERNAL_SERVER_ERROR (500) because
Drizzle sent a text[] where the column expected jsonb.
- schema.ts: tags → pgJsonb('tags').notNull().default('[]')
- migration: dropped+recreated to match schema (jsonb, ms epoch defaults,
owner_user_id default 'anonymous')
Two independent WS handlers (useMessagesWsSync for message_created/
updated/analyzed, and useMessagesStream for message_snapshot) both
prepend live messages to the SWR list without enforcing order.
When frames arrive out-of-order (common with batched WS delivery),
the message feed gets scrambled.
Fix: add sortMessages() helper that sorts newest-first by created_at
(the list's stored order before .reverse() for display) and apply it
in every patchLists/mutate updater: message_created, message_updated,
message_analyzed, message_snapshot, and useLoadMore page appends.
Function declaration is hoisted so useLoadMore (defined above the
helper) can use it.
The fix-imports.mjs script blindly appended '.js' to every @/ alias
import, even when the source specifier already carried a .js
extension (e.g. '@/shared/config/index.js'). This produced
'index.js.js' in the emitted dist/, causing ERR_MODULE_NOT_FOUND
at startup.
This was latent: only triggered once digestScheduler.ts (which
uses @/shared/config/index.js with explicit extension) was built.
The user-reputation removal (2a8f6d9) was also blocked by this
bug — stale binary kept crashing with 'user_reputations' query
errors because it was never redeployed.
Fix: only append .js when the @/ specifier has no existing
extension. Applied to both gateway and backend scripts.
SearXNG was already replaced by Wikipedia REST/Action APIs (wikipediaClient.ts).
Update comments to reflect the current implementation: term glossary now
resolves definitions via Wikipedia → Redis → Postgres cache chain, with no
SearXNG dependency.
Discord GoLive sends screen-share audio on a separate SSRC from the
user's microphone. In @discordjs/voice v0.19, VoiceReceiver.onUdpMessage
silently drops packets for SSRCs not in ssrcMap (which is only populated
from VOICE_STATE_UPDATE/VOICE_SERVER_UPDATE). This caused screen-share
audio to never trigger receiver.speaking and never reach the speakingHandler.
Fix: hookScreenShareAudio() wraps onUdpMessage to:
1. Detect incoming RTP packets with unknown SSRCs (OPRUS payload type 120)
2. Infer the owning userId by proximity to known audioSSRC
3. Clone the user's VoiceUserData into ssrcMap under the new SSRC
4. Let the original handler decrypt and forward to the subscription stream
5. Listen on ssrcMap 'create'/'update' events for video SSRC changes
Also removes the broken initial approach (polling ssrcMap which never
contains screen-share SSRCs).
Automated public weekly summary: top categories/domains/channels + coverage rate, posted to configured webhook. Uses getDatabase() direct query (no oRPC HTTP dependency), guards one-fire-per-week on restart.
Gateway @/ alias imports use no .js extension (relative imports
keep .js). The .js suffix on @/ paths caused double-extension
ERR_MODULE_NOT_FOUND (embeddingClient.js.js) at runtime.
- Persist structured verdict (flags/severity/confidence/evidence) on
moderation_actions so the public web can show WHY a message was moderated.
- Add a persistent Qdrant archive collection (gmw_message_archive); embed
every captured message at capture time (fire-and-forget, best-effort).
- Public semantic search over the archive (backend oRPC + FE toggle on the
messages view). Both features are read-only/public and fully automatic.
Migration: 0015_add_moderation_explainability.sql
- Memoize buildSystemPrompt by (mode|channelCulture); identical signatures
now reuse the ~5k-token core instead of rebuilding per sub-batch call
(textBatchProcessor rebuilt it inside the loop; a 200-msg batch re-sent
the full system prompt ~4x). Correction tail stays per-attempt (uncached).
- Hoist URL-image -> vision evidence out of the per-sub-batch loop in
textBatchProcessor: it depends only on fetched images + full target set,
so compute once per whole batch, not per sub-batch.
- Compact system instructions: collapse 3x-duplicated 'evaluate by content
alone' statements into one standalone rule; trim output.ts channel-culture
+ context framing already covered by rules.ts/system.ts; drop duplicate
programming-error-log few-shot (id 17, covered by rules AMAN list).
- Fix misleading config default: AI_LLM_BASE_URL default -> omniroute
(gateway already runs omniroute via BWS; 9router was dead/misleading).
typecheck + lint + build green.
- Add shared skeleton building blocks (SkeletonHero, SkeletonMetricRow,
SkeletonPanel, SkeletonRows) and wire them into every view's initial
loading branch, replacing bare spinners for a cohesive shimmering shell.
- Polish EmptyState with a glow-ring icon chip instead of a flat icon.
- Harden .light theme: color-scheme, tuned scrollbar + selection for pale
canvas. Dark/light theme toggle (next-themes) already persisted in TopBar.
- resetOffensiveNickname: skip when target role sits above bot
(member.manageable) instead of hammering a doomed setNickname PATCH
that Discord rejects with 50013 'Missing Permissions'. Log the
Discord error code on failure for clear diagnosis.
- llmCaller: include contentPreview (first 200 chars) in the parse-
failure warning so non-JSON LLM responses are debuggable.
- Add wikipediaClient.ts: native fetch to Wikipedia REST/Action APIs
(search + summary), no extra npm dependency.
- Extract shared Redis cache into cacheStore.ts (decoupled from search).
- Term glossary now uses wikipediaSummary for direct article lookup.
- Remove searxngSearch.ts entirely; drop SEARXNG_BASE_URL config,
add WIKIPEDIA_LANG / WIKIPEDIA_TIMEOUT_MS.
- Rename backend searxngCalls metric to webSearchCalls.
User: 'jangan ada reputasi juga' — no profile, no reputation in the prompt,
raw messages only.
- textBatchProcessor: drop initializeUserReputation fetch + <user_reputation>
tag injection (kept the minimal <message> tag + reply/reference context).
- visionAnalyzer (prepareMediaMessage): same removal.
- prompts/system.ts + prompts/output.ts: replace <user_reputation>/<user_history>
instructions with an explicit 'no per-user profile/reputation context'
note so the LLM judges purely on message content + conversation/web/location.
- mediaBatchProcessor: fix stale comment.
Trust/infraction state is STILL written to the DB (userReputationsTable) for
enforcement — only the LLM context injection is removed, so moderation
actions (mute/ban via infraction thresholds) keep working.
Net: even smaller prompts (no per-user context at all) → more messages fit
per request, and one fewer DB round-trip per unique user per sub-batch.
tsc, biome, vitest (129) all clean.
User insight: personal profile summaries bloat the prompt (less room per
request) and add a per-user DB/Redis round-trip for little moderation signal.
Only the behavioural <user_reputation> history is kept.
- textBatchProcessor: stop fetching getUserProfile; remove <user_profiles>
block + <user_profile_ref> from message tags. Keep <user_reputation>.
- mediaBatchProcessor + visionAnalyzer: same removal (profile fetch + ref).
- prompts/system.ts + prompts/output.ts: drop stale <user_profiles>/
<user_profile_ref> instructions; point LLM at <user_reputation> instead.
- aiAnalyzer: gate userProfileLearner behind AI_USER_PROFILE_LEARNING_ENABLED
(default false) — generates profiles nobody reads, pure LLM/DB waste.
- Add AI_USER_PROFILE_LEARNING_ENABLED config knob.
Net: smaller prompts (more messages fit per request), fewer DB round-trips
per sub-batch, and no background LLM calls learning unused profiles.
tsc, biome, vitest (129) all clean.
User insight: rather than many small per-batch API requests, pack many
messages into ONE request so a burst is analyzed with far fewer calls.
- AI_LLM_TEXT_BATCH_SIZE 20 -> 60 (one request now carries ~3x more messages).
- AI_ANALYSIS_MAX_TARGET_TOKENS 4000 -> 14000 (the scheduler's token-budget
gate was trimming pending messages to ~20 before they reached the sub-batch
splitter; raising it lets ~60 messages through to a single LLM call).
- AI_LLM_TEXT_ANALYSIS_TIMEOUT_MS 30000 -> 45000 (one larger call needs more
headroom; gemini-flash-lite has a 1M-token context so 14k+8k is trivial).
Net effect when ramai: a 60-message burst = 1-2 API calls instead of 3+,
less semaphore contention, faster throughput.
- Parallelize per-user reputation/profile fetches in textBatchProcessor
(was a serial ~2N DB/Redis round-trip loop per sub-batch; now Promise.all
over unique users). Cuts per-batch latency, biggest win on small/quiet
batches.
- Make the LLM concurrency semaphore dynamic (cached per config value) instead
of frozen at import time, so AI_LLM_MAX_CONCURRENT is tunable without code
change and reflects current config.
- Bump AI_LLM_MAX_CONCURRENT default 5 -> 8 (gemini-flash-lite is cheap; helps
throughput when busy).
- Lower AI_ANALYSIS_DEBOUNCE_MS 500 -> 250 (snappier first-message analysis
when quiet).
- Lower AI_ANALYSIS_RECOVERY_INTERVAL_MS 15000 -> 10000 (stuck/errored
messages re-analyze sooner).
tsc, biome, vitest (129) all clean.
Backend returns messages DESC (newest first); the view previously rendered
that directly, so the feed was inverted vs Discord (old at bottom, new at top)
while the load-older control sat at the top — contradictory.
- Reverse the display list so it reads oldest→newest top→bottom, like DC.
- Load-older (cursor to lower created_at) prepends at the top; scroll position
is preserved by offsetting scrollTop by the height added above.
- Open at the bottom (newest visible) on first load / scope change.
- New live messages append at the bottom and auto-scroll only when the user is
already near the bottom (nearBottomRef), so reading history isn't disrupted.
- Scroll container now tracked via ref; onScroll updates nearBottom + triggers
load-older when scrolled to the top.
tsc, biome, next build all clean.
- Set stream:false on the /chat/completions request so the bot gets one
complete response instead of an SSE token stream.
- Add reasoning_effort:"none" to suppress extended-thinking/reasoning tokens
(ignored by non-reasoning models like gemini-flash-lite).
- Add parseResponse(): handles both the JSON object 9router returns for
stream:false and the SSE text it may still emit, delegating SSE to parseSse.
Verified live: omniroute returns 200 application/json with message.content.
- Add viewport export with viewportFit: "cover" so iOS exposes
env(safe-area-inset-*) (required for the insets to take effect).
- NavRail / TopBar / main / Toaster now respect safe-area insets so content
clears the iPhone notch and home indicator in both portrait and landscape.
- prefers-reduced-motion: the media query already disabled declared animation
classes; harden it with a global transition/animation duration override and
kill the scan-line shimmer so motion-sensitive users get a fully static UI.
Verified tsc --noEmit + next build clean.
- SectionHeader: action (filters/legends) now wraps below the title on narrow
screens instead of overflowing beside it (flex-wrap, gap-2 sm:gap-3).
- GuildChannelPicker: selects go full-width and stack on mobile (w-full
sm:w-44 / sm:w-52) instead of fixed widths that exceeded a 375px viewport.
- Messages search: w-full sm:w-64 so it doesn't crowd the picker on mobile.
- TopBar: tighter padding (px-4 sm:px-5), smaller title on mobile, connection
status uses compact (dot only) on mobile, ambient pill hidden < sm.
- Shell main + dashboard channel label: responsive padding / shrink-0 widths.
Verified tsc --noEmit + next build clean; targets breakpoints 375/768/1024/1440.
- Show an explicit Loader2 spinner row ("Loading older…") while the next page
fetches, instead of a disabled button.
- Cap appended older pages at MAX_OLDER_PAGES=10 (500 messages) so a long
scroll-up never pulls the entire history; show a "capped" hint pointing to
search. Reset the counter when guild/channel changes.
Wire the existing useLoadMore + useMessagesHasMore pagination hooks into the
Messages view: add a "↑ Load older messages" button at the top of the list and
auto-load the next (older) page when the user scrolls to the top. Backend
messages.list already returns a created_at-based nextCursor (DESC order), so
older pages are just subsequent cursors. Newest-first live feed is preserved;
the load-older control is hidden during search.
Browser connects oRPC over wss://…/trpc (partysocket). The gmw-proxy nginx
only forwarded /api and /ws to the backend, so /trpc upgrades fell through to
Next.js SSR and the socket never opened ("WebSocket is not open"). Add a
/trpc location (WS upgrade headers) mirroring /ws. Backend already serves
oRPC on /trpc (HTTP RPCHandler + WS ORPCWebSocketServer on :4001).
Verified: ws://127.0.0.1:4001/trpc upgrade OPEN; SSR + server-side fetch RPCLink
also use /trpc directly so only the browser path was broken.
flake.nix only rewrote @/ aliases but left extensionless relative imports
(./router) in compiled dist/. node dist/index.js (how prod runs) cannot
resolve extensionless ESM specifiers -> ERR_MODULE_NOT_FOUND -> backend
crashlooped (444 restarts, port 4001 dead). Extract the fixer into a shared
scripts/fix-imports.mjs that appends .js to extensionless relative imports and
rewrites @/ aliases, and wire it into backend + discord-gateway build phases.
Verified: fresh tsc + fixer -> node dist/index.js boots; oRPC over /trpc
serves both HTTP POST and WebSocket (config/dashboard/voice/moderation/
media/chatbot/analysis) end-to-end against Postgres + Redis. next build
passes with the oRPC client + partysocket.
Replace REST module routers with a single typed tRPC appRouter served over
/trpc (HTTP + WebSocket), and rewire the frontend to call it via
@trpc/client wsLink (browser) and httpLink (RSC data layer). Existing
/api/health + /api/metrics stay as plain Express for infra scraping.
Notable fixes surfaced by the live smoke test:
- Express 5 / path-to-regexp v8 rejects the /trpc/* wildcard route; use a
prefix middleware that computes opts.path from the URL instead.
- nodeHTTPRequestHandler treats opts.path as the literal procedure path, so
it is derived per-request from req.url.
- Two ws servers on one http.Server (the /ws voice socket + /trpc) collided
and returned 400 on upgrade; both now use noServer + a manually routed
server.on('upgrade') keyed by path.
Verified: BE tsc+biome+40 vitest green; FE tsc+biome green; live
HTTP and WebSocket calls returned real prod data.
Co-Authored-By: Claude Opus 4.5 (1M context) <noreply@anthropic.com>
The standalone image analysis path (analyzeSingleMediaImage → llmVision →
llmChat) previously had no request-level timeout of its own — it silently
inherited the shared OpenAI client default (60s), and AI_LLM_MEDIA_ANALYSIS_
TIMEOUT_MS only governed the text+media *batch*, not a single vision call.
- Add AI_LLM_VISION_ANALYSIS_TIMEOUT_MS (default 60000) to config.
- llmChat now accepts an optional per-request `timeout` in LlmCallOpts,
forwarded to the OpenAI request options (falls back to the 60s client
default when omitted).
- llmVision passes config.AI_LLM_VISION_ANALYSIS_TIMEOUT_MS, so a single
image/sticker/emoji analysis gets a guaranteed 1-minute budget and is
independently tunable from the text path.
Verified: tsc + biome green, 129 gateway tests pass.
Co-Authored-By: Claude Opus 5 (Nous Research)
The sidebar rendered /dashboard twice: once as a hardcoded NavItem
(lines 45-50) and again via navItems.map() (navItems[0] is also
/dashboard). Dropped the hardcoded item so the single source of truth
(navItems in lib/navigation.ts) drives the rail. Removed the now-unused
LayoutDashboard import.
tsc + biome green.
Co-Authored-By: Claude Opus 5 (Nous Research)
The chatbot agent now has 14 tools (was 4) so it can answer about ANY
server situation from live data instead of a static snapshot:
- get_server_stats (now also returns clean count)
- get_top_channels, get_recent_activity, get_top_flagged
- search_messages (LIKE keyword search)
- get_user_messages, get_user_profile, get_user_reputation
- get_channel_culture
- get_message_detail (full AI analysis of one message)
- get_message_reviews (human moderation queue by status)
- get_voice_recordings (with transcriptions)
- get_moderation_timeline (daily flagged/warn/clean trend)
- get_corrections (AI false-positive correction history)
Security/quality:
- Every executor now uses parameterized drizzle queries (eq/like/and).
The old code interpolated model-supplied IDs into sql.raw() — a SQL
injection vector. Removed.
- Split static tool *definitions* into chatbot.toolDefs.ts (no DB import)
so the LLM-facing schema can be unit-tested without loading the
database/config layer. chatbot.tools.ts keeps only the executor.
Verified: tsc + biome clean, 40 backend tests pass (4 new covering the
tool-contract: names unique, required args declared, full situation
coverage).
Co-Authored-By: Claude Opus 5 (Nous Research)
The chatbot already had an agentic tool loop (get_server_stats,
get_top_channels, get_recent_activity, get_top_flagged), but processMessage
still baked a serverInsights snapshot into the system prompt and told the
model to "answer from that data". That defeats the tools: the model answered
from a stale snapshot instead of living numbers, and the guild/channel scope
the frontend sends was never forwarded to the tools.
Changes (services/backend/src/modules/chatbot):
- Remove getServerInsights() + ServerInsights (dead after this change).
- buildSystemPrompt(): drop the hardcoded stats block; instruct the model it
has NO memorized server numbers and MUST call a tool for any server-data
question, answering only from tool results.
- processMessage(): stop fetching insights; pass the request guildId/channelId
scope through to callLLM.
- callLLM(): accept scope; auto-fill empty guildId/channelId on tool calls from
the request scope so the model never has to guess IDs and tools always query
the right server.
Behavior: answers now come from live DB data via tools, scoped to the server
the user is chatting in. tsc + biome + 36 backend tests green.
Co-Authored-By: Claude Opus 5 (Nous Research)
buildCorrectedFewShotExamples() (a getRecentCorrectedModerations(5)
DB hit) was called inside the per-sub-batch buildContent closure in
textBatchProcessor.ts — re-queried for every sub-batch (≈10× for a
200-msg burst) AND re-fired on each parse-error retry. mediaBatchProcessor
already hoisted it once. Mirror that: fetch once per runTextOnlyBatch,
reuse the cached string inside the closure.
No behavior change — identical content, fewer identical DB reads.
tsc + 129 tests + biome green.
Co-Authored-By: Claude Opus 5 (Nous Research)
The 32 few-shot examples each re-echoed score/confidence/
recommended_action/categories/policy_version inline (~150 chars ×
32). Those fields carry zero moderation-decision signal — the schema
and their ??-default coercion already live in OUTPUT_INSTRUCTIONS +
moderationResponseParser.ts. Removed 96 redundant key/value pairs.
Kept per-example: message_id, status, flags, severity, evidence,
analysis — the fields that actually teach decisions. Parser derives
the rest via ?? fallback, so real output shape is unchanged.
examples.ts: 21.7K→18.5K chars; FEW_SHOT(mixed) 15.3K→13.4K.
Total mixed system prompt now 33.9K (was 39.3K at audit start,
~14% leaner). tsc + 129 tests + biome green.
Co-Authored-By: Claude Opus 5 (Nous Research)
- prompts/system.ts: merge 3 overlapping framing blocks (Blok Data /
Konteks Pengguna / Framing Konteks vs Target) into 1 tight block —
same coverage, no duplicated "standalone judgment / profile-is-
reference-not-evidence" prose.
- prompts/output.ts: trim duplicated user_history/standalone paragraph
in PERSONALITY & MEMORI (keep concrete per-case lessons).
- prompts/examples.ts: drop 2 exact-duplicate-lesson few-shots (LGBT id=19
dup of id=30; weapons-tech id=33 dup of id=32). All teaching signals
retained via the surviving example of each lesson.
Static system prompt: text 32.7K→29.2K, mixed 39.3K→35.8K chars
(~10% smaller). No moderation rule, zero-tolerance category, or decision
tree altered — accuracy-controlling content untouched. tsc + 129 tests +
biome green.
Co-Authored-By: Claude Opus 5 (Nous Research)
- gateway-metrics: collectors now run per scrape so Prometheus sees real
data (process memory/uptime + live AI-analysis pipeline gauges) instead
of an always-empty stub. bootstrap registers the pipeline collectors.
- systemd: MemoryMax 512M -> 1G (live RSS ~500MiB, peak 508MiB; 512M left
~2% headroom and risked an OOM-kill restart; host has 8GB free).
- config: POSTGRES_POOL_MIN 2 -> 0 so main + 4 Piscina worker threads don't
hold ~10 permanently-open idle pg connections against PgBouncer.
- docs: rewrite stale ARCHITECTURE.md / MODULE_STRUCTURE.md (winston ->
pino, removed mock-crc/indonesianTextNormalizer, renamed
aiAnalysisWorker/llmModerationClient).
Verified: tsc clean, 129 vitest pass, biome clean on changed files.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Adds per-message AI moderation analysis time (ai_analysis_duration_ms)
so operators can see how long the LLM took to moderate each message.
Gateway:
- messagesTable: new ai_analysis_duration_ms (bigint) column.
- AIAnalysisUpdate + buildAIAnalysisSet: carry analysisDurationMs through
both single and bulk update paths.
- ai-analysis-worker: measure wall-clock time around runModerationAnalysis
and attach it to every result in the batch.
Backend:
- Mirror schema column; messageMapper maps ai_analysis_duration_ms;
moderation-types + MappedMessage expose it.
Frontend:
- message.ts type gains ai_analysis_duration_ms.
- AiBadge (messages view) shows 'status · 1.2s' when duration is present;
analysis view badge mirrors the same formatting.
DB:
- scripts/add-ai-analysis-duration.sql (idempotent ADD COLUMN IF NOT EXISTS).
No behavior change for moderation logic; null until new gateway build
records values.
Qdrant upserts were failing with 'This operation was aborted' ~32x/2h,
so semantic moderation cache entries were silently dropped. Root cause:
upsertQdrantPoint ran ensureQdrantCollection() on EVERY call — a GET
(and sometimes DELETE+PUT) round-trip — while the request AbortController
had only a 10s timeout. Under moderation load Qdrant is busy (the
gmw_text_moderation collection is not yet HNSW-indexed, so searches are
full-scans), the extra round-trips pushed the upsert past 10s, and the
client aborted it.
- Memoise ensureQdrantCollection() at module scope so the collection is
verified exactly once per process (resetQdrantCollectionCache() for
tests / config reload).
- Bump the upsert request timeout 10s -> 30s so a transiently busy
Qdrant no longer aborts the write.
Qdrant server itself is healthy (<100ms for direct upsert; collection is
green), so no server-side change is needed. Semantic cache should now
populate reliably.
Address every remaining biome lint/format warning across both services
so the codebase ships warning-free:
- textCacheStore: drop unused deleteExpiredQdrantPoints import; hash
image cache key (sha256[:32]) so long/base64 URLs no longer blow the
text_analysis_cache PK B-tree 8191-byte index (was aborting the media
analysis lock INSERT).
- bootstrap: drop unused unhandledRejection promise param.
- moderationOrchestrator: drop unused destructure at L197.
- mediaDownloader / textBatchProcessor / transmitter: replace non-null
assertions with proper null guards (stickerName ?? '', urlImages.get
guard, backpressureQueue.shift guard).
- backend utils: throw lastError ?? fallback instead of lastError!.
- message-capture: remove unused (retentionDb), (moderationActionsDb,
reviewsDb); simplify renderDiscordMentions guard to optional chain.
- transmitter: remove dead write-only field + its assignments.
No behavior change beyond the cache-key hashing (now deterministic
fixed-length) and the intentional null-safety guards.
imageResizer.ts had a line exceeding the print width that biome flagged
as a formatter error, failing the Build & Deploy biome check. Re-format
the file. No logic change.
Two root causes behind 'all image analysis failing':
1. imageResizer still emitted lossless PNG for vision input. A 1024px
Facebook photo balloons to multi-MB PNG base64 that the vision model
silently rejects ('Vision API null response'). Switch to JPEG q85
(no upscaling) — same photo drops to ~100-400KB, model processes fine.
Re-encodes even already-small images so raw originals never bloat the
data URL. Added tests/imageResizer.test.ts covering both cases.
2. acquireMediaAnalysisLock INSERT aborted with 'index row requires N
bytes, maximum size is 8191'. text_analysis_cache.text is the PK in a
B-tree index (8191-byte/row cap); callers pass the raw image URL as the
key, and base64 data URLs / very long URLs blow past the limit, so the
lock INSERT fails and every media analysis is skipped. Hash the URL in
makeImageCacheKey (image:<sha256[:32]>) — fixed-length, deterministic,
well under the limit. All store/get/lock/delete callers already route
through this function so lookup stays consistent.
dontPatchELF only disabled the patchELF sub-phase; fixupPhase's
shrinkELF step still emits the same error on the prebuilt .node addons
and .o/.a object files in node_modules. Skip the entire fixupPhase
(dontFixup = true) for the gateway — node is the external interpreter
and .node addons are self-contained dlopen prebuilts, so Nix RPATH
patching/stripping is neither needed nor wanted.
Add dontPatchELF = true to the discord-gateway derivation. Nix's
fixupPhase runs patchELF over $out/node_modules and chokes on the
non-ET_DYN ELF files (.o/.a objects + prebuilt .node addons), emitting
hundreds of non-fatal 'patchelf: wrong ELF type' lines per build. The
real binary is node (external, RPATH-fixed) and the .node addons are
self-contained prebuilts loaded via dlopen, so Nix RPATH patching is
neither needed nor wanted. Shebang patching still runs.
Drop npm_config_build_from_source=true so node-pre-gyp downloads the
published prebuilt .node for Node 22 (ABI node-v127, linux-x64-glibc-2.35)
instead of compiling libopus C++ every build. Replace the hardcoded
'npm run install' (node-gyp compile) loop with 'pnpm rebuild @discordjs/opus'
which runs the package's own install script (prebuilt fetch, source build
only as fallback). sharp already uses @img prebuilt packages (its install
script failure is non-fatal), so only opus was actually compiling.
Router.push was a no-op in the standalone build (Next trailingSlash
interaction), so the sidebar buttons and command palette silently failed
to navigate. Replaced next/link + router.push with plain <a href> anchors
in NavRail and CommandPalette — verified working on all routes.
Biome tightened to zero warnings:
- Disable noArrayIndexKey (positional equalizer bars), noStaticElementInteractions
(intentional dismiss/hover overlays), useMediaCaption (voice clips)
- Avatar uses background-image instead of <img> (noImgElement)
- Command palette list items keyed correctly
- Format pass to satisfy the formatter
- Created nav-debug.cjs to log anchor tags and simulate clicks on the Voice navigation link, capturing click events and page navigation.
- Added nav-test.cjs to test the Voice link click and log the URL at various intervals, capturing any page errors.
- Introduced nav-test2.cjs to check the presence of specific elements on the /voice/ page and log any console errors.
- Implemented nav-test4019.cjs to monitor network requests and responses related to the Voice navigation, verifying button presence and click functionality.
Hapus template dashboard lama (top bar + side rail + main + right panel +
bottom prompt). Ganti dengan layout yang benar-benar beda:
- AmbientField: full-bleed WebGL canvas haze, drift speed + densitas
ngikut load server, warna ngikut signal moderasi terakhir
(clean→lime, warn→amber, flagged→vermilion). Background tanpa container.
- View jadi full-bleed: headline raksasa bottom-left, metric cluster
floating top-right (no box), event ribbon drift di tengah, command
whisper di very bottom.
- AmbientShell di layout.tsx: gak ada TopBar/LeftRail untuk /dashboard
exact. Route lain (messages/voice/media/dll) tetap ClassicShell.
- Tidak ada card, tidak ada grid, tidak ada panel, tidak ada tab.
Verified: tsc clean, next build 11/11 halaman, biome clean.
- config: add AI_LLM_VISION_BASE_URL + AI_LLM_VISION_API_KEY (separate from text router)
- llmClient: llmVision() now calls dedicated vision endpoint when configured
(axios POST to integrate.api.nvidia.com, model nvidia/nemotron-3-nano-omni-30b-a3b-reasoning,
reasoning_budget 16384, non-stream), falls back to router combo otherwise
- keeps text/moderation on omniroute, vision on NVIDIA direct
- VoiceView now reads connected/activeChannelName from useVoiceStatus
(SWR live, invalidated by connect/disconnect) instead of initialStatus
- Seed useSpeakers from live status.activeSpeakers
- Add 4s refreshInterval to useVoiceStatus so state converges
(tsc clean, next build green)
Symptom: video plays ~1s then freezes. BaseMediaStream sync logic:
- video _pts advances 33.3ms/frame (timeBase 1/fps), audio _pts advances
20ms/packet (timeBase 1/48000) — two synthetic frame-index timebases that
never share a clock.
- If audio starts late (ffmpeg audio init / Ogg header), ptsDelta = video-audio
stays positive → isAhead() true → video loops 'await sleep(frametime) while
isAhead()' → video freezes. Downchain: vPipe fills → proc.stdout paused →
demuxer emits ~15fps (log: 30 frames per 2s).
Upstream dank sets syncStream because node-av provides REAL PTS from NUT in a
consistent timebase. Our raw-h264 demuxer has no real PTS; per-stream sleep-PTS
pacing alone keeps both at 1000ms/s, which is correct without a shared clock.
Re-enable sync only if real PTS is added.
Lag root cause: vPipe/aPipe were objectMode PassThrough HWM 128 → the pipe
held up to 128 frames ≈ 4.3s of video before backpressure reached the encoder.
The viewer was watching a 4+ second stale backlog.
Fixes (both faithful to @dank074/discord-video-stream):
1. vPipe/aPipe HWM 2 — at most ~1-2 frames in flight (~66ms @ 30fps), so the
writeFrame() backpressure pauses ffmpeg stdout almost immediately and the
whole chain (encoder → NUT → demuxer → vPipe → BaseMediaStream → WebRTC)
runs at the sender's real pace, exactly like dank's 'resume &&= vPipe.write'.
2. Wire vStream.syncStream = aStream — audio is the master clock; video
sleeps/wakes on ptsDelta like upstream newApi.js. Prevents A/V drift under
variable encoder throughput.
Per user direction ('pakai dank sebagai referensi karena itu yg berhasil'):
drop the custom setInterval/tail-drop emission clock entirely. The demuxer
now writes each access unit straight to vPipe with a monotonic PTS and lets
BaseMediaStream (ported 1:1 from @dank074) handle pacing via sleep-PTS + A/V
sync, exactly like the upstream library. The custom clocks were the source of
the blank tile (IDR delivery race) and the lag (head-drop watching 10s-old
frames).
Adds proper backpressure: pause ffmpeg stdout when vPipe.write() returns
false, resume on drain — mirrors dank's 'resume &&= vPipe.write(packet)' so the
encoder self-throttles to the WebRTC sender's real pace instead of bursting.
The tail-drop rewrite let a P-frame supersede a pending keyframe before the
emit tick fired, so the decoder never received an IDR → blank GoLive tile.
Give keyframes their own slot (pendingKey) that P-frames cannot steal, and
only emit a P-frame once at least one IDR has been shown (haveReference).
IDR is always emitted first when present so the reference re-establishes.
The Node token-bucket pacer used HEAD-drop (emit frames in arrival order,
drop newer ones when over budget). Under the encoder's ~330fps burst (ffmpeg
-re does not reliably throttle YouTube-DASH webm), the viewer was watching
frames ~10s behind live → frozen / 'patah-patah' video while audio (not
rate-limited) played current = desync.
Replace it with a steady setInterval emission clock at videoFps: each tick
emits exactly ONE frame — the NEWEST buffered one — and discards everything
older (tail-drop). At most one frame is ever held, so no backlog and no lag;
the emit clock (not the encoder rate) defines playback speed. Keyframes are
never superseded so the decoder keeps getting IDRs. Audio stays in sync.
yt-dlp 2026.07.04 rewrites the --cookies file on close. Handing it the
root-owned /etc/.../ytcookies.txt (not writable by the gmw service user)
caused PermissionError -> exit 1 on every screen-share download attempt.
- buildCookieArgs on-disk branch now copies the system cookie file into a
per-run temp file (like the env branch) so write-back lands somewhere we
own; unreadable -> anonymous.
- resolveInputWithRetry Invidious fallback regex now also matches
permission|EACCES|cookie, so a cookie failure triggers the link-alternative
(no-auth Invidious mirror) path instead of failing all retries.
- adds regression test asserting the original cookie path is never passed to yt-dlp
The live pipe (yt-dlp -o - -> ffmpeg) delivers data at network speed with
unreliable PTS, which defeats ffmpeg -re and made x264 -r 30 force-duplicate
held frames -> ~1fps video (the patah-patah symptom). Per user suggestion,
download the FULL clip to a temp file first (downloadScreenInput), then feed
that FILE PATH to prepareStream. String inputs already get -re, so the
encoder now paces cleanly at 1x against a monotonic-PTS file — proven
reliable in local tests (vs the live pipe which always bursted). Temp file
is removed on stream end / stop.
- getDirectScreenInput -> downloadScreenInput (returns file path)
- resolveInputWithRetry now awaits a completed file + retries on failure
- screenShareController.stops/cleanup removes the per-run tmpdir
- screenShareInput.test.ts updated to the file-download contract
Previous code only added ffmpeg -re when input was a string URL. Screen
share passes a Readable pipe (yt-dlp merge -> stdout) delivered at network
speed (bursts + stalls). Without -re the encoder slurps it instantly and,
when the merge stalls, x264 -r 30 force-duplicates the last held frame
~30x -> viewer sees ~1fps while WebRTC still paces 30fps. Add -re for all
inputs so the encoder paces at the stream's native PTS rate and emits a
fresh picture every frame.
- Demuxer.ts: deterministic token-bucket video pacing (replace unreliable ffmpeg -re which did not throttle the live multi-stage pipe — demuxer emitted ~240fps vs 30fps sender, 100k+ frame backlog, frozen video). Surplus non-key frames dropped; keyframes forced through; audio on fd3 unaffected.
- biome.json: pin noExplicitAny/noUnused* to off/warn. Biome 2.5.x (drifted via --no-frozen-lockfile) promotes these to errors and was failing the CI gate on pre-existing backend code unrelated to this change. Restores the warn-level behavior the config schema 2.2.0 expects.
Root cause (3rd iteration): ffmpeg '-re' on the demuxer does NOT reliably
throttle a multi-stage live pipe (merge ffmpeg -> encoder x264 -> NUT ->
demuxer). In production the demuxer still emitted ~240fps while the WebRTC
sender consumed 30fps, building a 100k+ frame backlog (observed: frames=197490
vs sent #24600, ~8.4 min in). The sender always emitted the OLDEST buffered
frame -> video frozen ~10 min behind live, while audio (tiny, jitter-buffer
recovered) stayed smooth. Local file/pipe tests showed -re working (30fps)
but the live YouTube/WebM pipeline did not — -re is not trustworthy here.
Fix: enforce 1x video output with a token-bucket limiter in the demuxer
(Node side), independent of ffmpeg. Capacity = 1s of frames, refill 1 token
per 1000/fps ms. Surplus non-key frames are DROPPED (never buffered) so the
sender always emits the newest frame; keyframes are forced through even over
budget so the decoder keeps a fresh IDR. The limiter does NOT stall the ffmpeg
process (unlike the earlier proc.stdout pause), so audio on fd3 keeps flowing.
Verified: tsc --noEmit clean.
Root cause (revisited): the previous gate paused proc.stdout when vPipe was
full. That stalled the SAME ffmpeg process that also writes audio on fd3, so
audio stuttered; and the ~8s backlog already built never drained → permanent
lag. Symptom: 'video still lags bad, now audio also choppy'.
Fix:
- spawn demuxer ffmpeg with -re for stream (pipe) input. Verified locally:
a 5s NUT clip demuxes in 0.088s without -re (57x burst) vs 4.539s with -re
(real-time). -re throttles the input read, which back-pressures the whole
upstream chain (encoder x264 -> merge ffmpeg -> yt-dlp) through OS pipes,
pinning production at 1x. No unbounded backlog.
- drop oldest queued frame when vPipe readableLength >= 30 (transient sender
stall guard) instead of pausing stdout — keeps video fresh and audio intact.
- removed gateSource/sourcePaused entirely.
Audio and video now pace together at 1x; video is the newest frame, not an
8-second-old one.
Root cause: prepareStream's ffmpeg consumed a YouTube VOD at download/CPU
speed (~10x real-time), so the demuxer buffered a huge frame backlog.
The sender paces at 30fps but always emitted the OLDEST buffered frames, so
the viewer saw frozen/laggy video while audio (tiny, jitter-buffer
recoverable) stayed smooth. That is exactly the 'video stuck, voice normal'
symptom reported live.
Fix: propagate vPipe backpressure UP to the demuxer's ffmpeg stdout — when
the sender can't keep up, pause the source, which stalls the demuxer and
back-pressures the encoder, pinning the whole pipeline to 1x. Also add a
realtime (-re) option for file/URL inputs (no-op for the streaming path,
which is what screen share uses).
Verified: 10s test clip encodes in 1.8s without -re vs 9.5s with it; tsc --noEmit clean.
Root cause: BaseMediaConnection.sendOpcode is a silent no-op when
ws.readyState !== OPEN. In GoLive, playStream() calls setVideoAttributes(true)
+ setSpeaking(true) the instant createStream() resolves (right after
SELECT_PROTOCOL_ACK), but the StreamConnection WebSocket can still be in
CONNECTING for a few ms — so op 12 (VIDEO, activating the video SSRC) was
silently DROPPED every session. Empirically verified: 0 ops 12/5 ever logged
across the entire journal, yet 10k+ video frames were sent and audio played
(audio SSRC is activated via the VoiceConnection handshake, independent of
GoLive op 12). Discord's media server thus received video RTP on video_ssrc
but was never told to forward it → black/broken shared-screen video with
working voice.
sendOpcodeWhenOpen retries up to ~2s for ws OPEN instead of dropping. Also
emits a=fmtp:101 packetization-mode=1;profile-level-id=42e01f in the answer
SDP (H264 FU-A fragments require packetization-mode=1 to reassemble).
Also removes pre-existing noNonNullAssertion lint (biome 2.5.8 now errors)
that was blocking the deploy CI.
Root cause: BaseMediaConnection.sendOpcode is a silent no-op when
ws.readyState !== OPEN. In GoLive, playStream() calls
setVideoAttributes(true) + setSpeaking(true) the instant createStream()
resolves (right after SELECT_PROTOCOL_ACK), but the StreamConnection WebSocket
can still be in CONNECTING for a few ms — so op 12 (VIDEO, enabling the video
SSRC) was silently DROPPED every session. Empirically verified: 0 ops 12/5 ever
logged across the entire journal, yet 10k+ video frames were sent and audio
played (audio SSRC is activated via the VoiceConnection handshake, independent
of GoLive op 12). Discord's media server thus received video RTP on video_ssrc
but was never told to forward it → black/broken shared-screen video with
working voice.
sendOpcodeWhenOpen retries up to ~2s for ws OPEN instead of dropping. Also
keeps the H264 packetization-mode=1 answer-SVP (defensive SDP correctness).
Also fix: emit a=fmtp:101 packetization-mode=1;profile-level-id=42e01f in the
answer SDP — H264 FU-A fragments require packetization-mode=1 to reassemble.
- Add examples for technical discussions (kinetic energy, drone weapon
engineering, physics simulations) that should be marked clean
- System rule: physics/engineering topics (kinetik, gravitasi, energi,
drone, senjata, drone warfare, CAD, CNC, 3D printing, robotics, aerospace)
are safe when in technical context — flag only if explicit threat
- Riwayat pengguna dengan pelanggaran sebelumnya tidak memengaruhi
penilaian pesan bersih yang terpisah dan tidak mengandung pelanggaran
- Removed getUserRecentInfractions usage in textBatchProcessor.ts and visionAnalyzer.ts
- Removed buildUserHistoryXml import and calls
- Messages are now evaluated standalone, not influenced by past violations in other channels
- Updated moderation prompts with clearer instructions about user_history usage
- Fixes issue where benign messages like 'tubuh manusia vs gravitasi' were incorrectly flagged due to carryover from previous drone weapons discussion
The user history context was causing the LLM to interpret unrelated current messages
as threats because it conflated them with past violations. Now each message is judged
on its own merit with only channel-specific context.
Verifies that two data URLs sharing the first 128 chars (same MIME prefix
+ identical base64 header — the real-world scenario that caused ALL images
to reuse the same cached vision analysis) produce DIFFERENT cache keys
under the fixed full-dataURL hashing, whereas the old 128-char-prefix
approach would collide. Also includes consistency + prefix tests.
Add debug logging to trace cacheKey + messageId + content length on
every vision cache HIT and MISS, so we can detect if the vision model
returns duplicate analysis for different images (provider issue vs
cache collision). Includes the phash on cache miss (new analysis cached).
Follow-up to 9f7ce7d which fixed makeImageCacheKey to hash full data
URL instead of just first 128 chars (root cause of all images sharing
the same cached 'konten judi' verdict due to hash collision).
Root cause: makeImageCacheKey() only hashed the first 128 chars of the
data URL. Since all resized images use the same MIME prefix
('data:image/png;base64,') + identical base64 header bytes, nearly every
image got the same 16-char hash → 'image:<same-hash>' → all images reused
the first cached vision analysis (often a gambling-detection verdict).
Fix: hash the entire data URL instead of just the prefix. Verified
114 stale 'image:' entries + 745 stale 'phash:' entries purged from prod
DB. tsc --noEmit clean, 133 tests pass.
bws-exec exposes the BWS secret as env GMW_YT_DOWNLOADER_COOKIES.
Materialize to temp Netscape file (yt-dlp --cookies needs a path).
Falls back to /etc/gmw-discord-gateway/ytcookies.txt written by deploy.
YouTube now blocks anonymous embeds (403 'Sign in to confirm you're not a
bot'). Resolve with account cookies via --cookies.
- mediaSource: buildCookieArgs() reads GMW_YT_COOKIES_PATH (default
/etc/gmw-discord-gateway/ytcookies.txt) and injects --cookies into
resolveMediaUrl + getDirectScreenInput + extractMediaInfo. Falls back
to anon if file missing (graceful 403, not crash).
- bws-exec now writes cookies file from BWS secret gmw_yt_downloader_cookies
on service start (systemd ConfigFile).
Root cause of "langsung left": YouTube bot-block/403 on u_c1tRmj7E4 (live
stream, LOGIN_REQUIRED) made yt-dlp timeout in resolveInputWithRetry (12s).
The timeout handler did cleanup() (removing once() listeners) THEN
tee.destroy(new Error(...)) — the PassThrough emitted 'error' with NO
listener left → unhandled stream 'error' event → uncaughtException →
gracefulShutdown → bot left voice.
Fix:
- resolveInputWithRetry: tee.destroy() silently after cleanup (error carried
in the rejection only); add permanent no-op tee.on('error') safety.
- prepareStream: output.on('error') no-op so ffmpeg spawn failure before
playStream attaches a demux listener never crashes the gateway.
- bootstrap: serialize uncaughtException/ClientError/DB errors with
{err, errorMsg, stack} (pino only serializes the 'err' magic key — the old
{error: err} key printed {} so crashes were invisible).
Revert 392bc35: streaming raw h264 video + opus on separate pipes broke
because prepareStream.output (pipe:1) feeds the Demuxer, but the opus
pipe:3 was never attached to the Demuxer's input — so for audio-capable
streams the Demuxer saw format=h264 (video-only) and emitted -an,
dropping audio RTP.
Correct design (from f1aa08c): prepareStream muxes video+audio into NUT
on a SINGLE pipe:1. The Demuxer then spawns a child ffmpeg that
demuxes NUT → -f h264 pipe:1 (pure AnnexB, start-code scan sees real
IDR type 5) + -f opus pipe:3 (Ogg Opus via createOggOpusDemux). The
start-code parser never touches NUT framing — it runs on the child
ffmpeg's clean h264 stdout.
Root cause: prepareStream muxed video+audio into a NUT container on pipe:1.
The Demuxer scans pipe:1 for AnnexB start codes (00 00 01) to split NAL
units into access units and classify keyframes (nal_type 5). NUT container
framing bytes sat in the stream and were scanned as NALs — NAL type 0
(NUT header) instead of 5 (IDR) → every frame classified key=false →
Discord decoder never got a decodable frame → static/black GoLive tile.
Fix: output raw H264 AnnexB on pipe:1 (demuxer target) and Ogg Opus on
fd3/pipe:3 for audio. NUT is only needed for *input* parsing (single
pipe carries both streams); output is demuxed into separate raw streams.
The demuxer resolved as soon as the VIDEO init line arrived on ffmpeg stderr.
With live NUT input the audio init line ('Stream #0:1: Audio: opus') lands in a
LATER stderr chunk (NUT info-stream packets are read incrementally from the
pipe), so `return { audio: aInfo }` captured undefined → playStream skipped
AudioStream → zero audio RTP on the audio SSRC → Discord showed a static
GoLive tile even though the NUT carried opus audio.
Fix:
- wait for BOTH video and audio init lines (when audio is expected) before
resolving demux metadata, with a 3s timeout fallback
- default aInfo to opus/48kHz when withAudio instead of undefined, so the
audio stream is always exposed even if the metadata line races the return
- Add term_glossary_cache table + migration 0014: resolved definitions are
stored permanently (definitions rarely change); misses stay ephemeral in
Redis/LRU with 1h TTL so transient failures get retried
- Lookup flow: LRU -> Redis -> Postgres (permanent) -> live SearXNG; DB hits
re-warm the fast caches; stale Redis miss sentinels no longer shadow DB
- Rate-limit-aware live lookups: concurrency 2 + stagger, retry once on empty
results, strict definition filter (Wikipedia preferred, rejects
disambiguation/ads/translate-homepages)
- Make SEARXNG_BASE_URL configurable via env (default unchanged)
Screen share showed a single frozen frame: the GoLive pipeline sent video
only (-"-an", h264 muxer cannot carry audio) so the audio SSRC never
transmitted and Discord kept the stream in thumbnail state.
- prepareStream: mux NUT when includeAudio (h264 muxer drops audio) and
return the actual container format
- Demuxer: support NUT input with a second output pipe (fd3) carrying
Ogg Opus; parse OGG pages into opus frames (20ms, 48kHz) emitted as
GoLiveFrames; fix metadata parsing that dropped the audio stream line
when it arrived in a later stderr chunk (early parsedMeta return)
- playStream: pipe audio.stream into AudioStream → RTP on the audio SSRC
- Encoders: -x264-params repeat-headers=1 → SPS/PPS inline before EVERY
IDR (NUT remux drops container extradata; also enables PLI recovery)
- screenShareController: includeAudio true
- tests: demuxerNut.test.ts — OGG parser unit test + real ffmpeg NUT
integration (video access units + parsed opus frames)
Second root cause (2026-08-12): even with yt-dlp http_headers forwarded,
YouTube still returns 403 when a signed DASH URL from --dump-single-json is
fetched raw by ffmpeg/curl on some videos (verified on fONoh7Pc6VU: curl
with the EXACT headers got 403; yt-dlp's own downloader succeeded). The
signature is tied to the extracting client context (po_token/visitor), not
just UA/IP.
Fix: getDirectScreenInput now spawns 'yt-dlp -o -' and returns its stdout
as a Readable — the same mechanism resolveMediaUrl already uses for music.
yt-dlp handles auth, cookies and transient retries internally. Merge
fragments go to /tmp/gmw-ytdlp-tmp (Nix store CWD is read-only → EACCES).
Removed resolveScreenInput + mergeScreenStreams (dead code).
Controller resolveInputWithRetry unchanged: tees the stream, waits for the
first byte (12s), retries with a fresh yt-dlp run up to 3x on error/EOF/
timeout, and destroys stuck inputs (EPIPE) so no process leaks.
Tests: rewritten for streaming (yt-dlp emits bytes; fail mode = exit 8
without stdout → stream must terminate with zero bytes).
Root cause (2026-08-12 11:50 test): merge ffmpeg hit a transient YouTube
403 and exited code 8 BEFORE prepareStream attached its input listeners
(voice release+join takes ~10s). The input's end/error events fired into
the void, the encoder stdin never received EOF, demux resolved with
fallback 0x0 metadata, setSpeaking fired anyway → stream 'started' with
zero frames for 8+ minutes (black tile, both ffmpeg processes hung).
Fixes:
- mediaSource: pass yt-dlp http_headers (UA/referer) to the merge ffmpeg
via -headers to suppress transient 403s; destroy the returned stream
with an error when the merge exits non-zero before producing bytes.
- screenShareController: resolveInputWithRetry — tee the merge stream and
wait for the first readable byte (12s timeout) before proceeding; on
error/EOF/timeout retry the whole resolution with a FRESH yt-dlp run
(signed DASH URLs expire fast) up to 3 attempts. Stuck merges get
EPIPE via input.destroy() so no process leaks per attempt.
- prepareStream: race guard — if the input already ended/destroyed before
listeners attach, EOF the encoder stdin immediately; first-frame
watchdog in playStream rejects 'started but nothing flowing' after 10s
instead of resolving with a silent black stream.
Tests: +2 (merge-fail zero-byte terminal state, -headers forwarding).
SDP offer advertises profile-level-id=42e01f (constrained baseline) but
x264 encoded the default High profile — Discord's receiver configures its
decoder from the negotiated profile, so the High-profile bitstream failed
to decode → black GoLive tile despite valid access units + correct RTP
timestamps (fixed in 42a503c).
- Add -profile:v baseline to H264 encoder options (matches @dank074's
proven config; SPS now 6742c01e → profile_idc=66 baseline, aligns with
the 42e01f fmtp).
- Default x264 tune film → zerolatency (no lookahead — correct for live
GoLive; @dank074 uses it).
- Update goLive-port test to assert baseline + zerolatency.
Demuxer emitted each AnnexB NAL as its own WebRTC frame (SPS/PPS/SEI
separate from slices) with a near-zero timestamp delta (duration=1 in a
1/90000 timebase → RTP +1/frame instead of +3000 @30fps). Discord's H264
receiver never receives a complete decodable access unit → black GoLive
tile despite frames flowing.
- Group NALs into access units: buffer param-set/SEI NALs, flush one
frame per slice with preceding parameter sets (AnnexB start codes kept
so the H264RtpPacketizer finds NAL boundaries).
- Timestamp each frame at the video frame rate: duration=1, timeBase
1/fps → BaseMediaStream frametime=1000/fps ms → RTP +clockRate/fps
(3000 @ 30fps/90kHz) and correct pacing.
- Thread explicit frameRate from playStream options (raw H264 has no
timing info; ffmpeg guesses 25fps on stderr).
- Strengthen golive-demux-live-e2e: validates every frame has a slice,
no bare param-set frames, keyframes carry SPS/PPS, timeBase 1/30.
Test 00:32 confirmed the video pipeline WORKS (1410 frames @ 1280x720 sent,
ready=true, camera off) but the gateway crashed at stream stop:
unhandledRejection → graceful shutdown → systemd restart (bot offline).
Root cause candidates (both were fire-and-forget promises without .catch):
- BaseMediaConnection.setProtocols().then(...) — rejects when the PC is
closed while setProtocols is in flight (stream teardown)
- void webRtcConn.createOffer().then(...) — rejects when the PC closes
while the offer is still gathering
Fixes:
- .catch on both promise chains (log + continue; teardown is expected)
- unhandledRejection handler now treats transient stream errors (EPIPE,
ERR_STREAM_DESTROYED, ERR_STREAM_WRITE_AFTER_END, ECONNRESET) like
uncaughtException already does — warn + continue instead of shutting
down the whole gateway. Non-transient rejections still log + shutdown
(with String(reason) so the detail actually shows).
ROOT CAUSE of empty GoLive tile (finally): prepareStream ran with
includeAudio: true + output -f h264. The h264 muxer cannot mux audio
('h264 muxer does not support any stream of type audio') → header write
fails -22 → stdout empty → Demuxer ffmpeg 'Invalid data found when
processing input' → 0 frames → black tile. Reproduced locally end-to-end
(13s backpressure delay + prepareStream + demux).
Fixes:
- screenShareController: includeAudio: false (video-only GoLive; demux
path never delivers audio anyway)
- Demuxer: pin input format -f h264 for stream inputs (raw AnnexB H264
has no magic header → auto-detect unreliable on delayed pipes)
- Streamer.signalStream: self_video: false — stop flipping on the bot's
camera in Discord (user request; screen share ≠ camera)
Verified: local repro now emits 644 frames 1280x720 (was 0); tsc/biome/
vitest all green.
Root cause of 'tile appears but content empty': demux() spooled the live
NUT/H264 input to a temp file and awaited stream 'finish' — but the merge
ffmpeg output never ends during playback, so demux deadlocked, no probe,
no transcode, 0 frames sent.
- Demuxer: pipe input straight into ffmpeg stdin (-i pipe:0), parse NAL
frames live from stdout; parse video metadata from ffmpeg stderr with a
1.5s race (fall back to H264 defaults). No spool, no await-end.
- screenShareController: pass width/height/frameRate (1280x720@30) to
playStream — matches the prepareStream encode settings, so setVideoAttributes
gets real dimensions even when ffmpeg can't report metadata on an open pipe.
- Add tests/golive-demux-live-e2e.ts: proves frames flow while input is
still open (regression test for the deadlock).
- signalStream: flip voice state to self_video:true/self_deaf:false before
STREAM_CREATE (Discord silently ignores the request while video disabled)
- createStream: attach dispatch listeners before first signal (race), clean
up listeners on timeout, retry STREAM_CREATE every 3s up to 4 attempts
(upstream issue #217/#219 — Discord randomly drops the request)
- sendOpcode: direct [goLive:Streamer] log bypassing bootstrap debug filter
(proves op 18 is actually broadcast)
| F1 | Exact-hash cache key menyertakan context (channel/thread) → teks sama di channel lain selalu miss | Killer hit-rate #1 |
| F2 | Semantic tier TIDAK memfilter context (Qdrant payload tak punya context) — sudah global tapi hanya aman krn sim 0.97 ketat | Inkonsisten dgn exact tier |
| F3 | Phase-1 lookup loop `await getCachedTextModeration(key)` per pesan → N round-trip PgBouncer per batch (60 msg = 60 query serial) | Latensi + beban DB |
| F4 | Verdict actionable (flagged/warn) dan clean sama-sama boleh di-serve semantic; toleransi akurasi beda | Risiko akurasi |
| F5 | `hit_count` tidak pernah di-increment oleh reader manapun | Hit-rate tak terukur |
| F6 | `wikipediaSearch()` (blok `<web_searches>`) tanpa cache — re-fetch tiap batch utk query sama | Latensi + spam ke WP |
| F7 | `fetchUrlSafely()` tanpa cache — link sama di batch berikutnya di-download lagi penuh | Latensi + bandwidth |
| F8 | Vision cache key dari data-URL base64 hasil resize → attachment sama via jalur berbeda (URL vs embed) = key beda → re-download + re-vision | Duplikasi kerja vision |
Non-goals: mengubah pipeline enforcement (auto-mute/ban trust-store writes), mengubah prompt
kebijakan moderasi, mengubah model/embedding provider.
## Desain
Semua perubahan degrade gracefully — cache gagal → perilaku lama (LLM). Akurasi dilindungi
asimetris: **hemat boleh untuk verdict non-actionable, konservatif untuk yang memicu aksi.**
### D1 — Cache metrics (F5)
-`textCacheStore.getCachedTextModeration()`: saat hit valid, increment `hit_count`
return`Kamu lagi ngobrol sama chatbot Discord Watcher — temen ngobrol yang tau keadaan server.
constscopeLine=scope.guildId
?`- Scope: kamu menjawab soal server/guild id="${scope.guildId}"${scope.channelId?`, channel id="${scope.channelId}"`:""}.`
:"- Scope: tidak ada guild spesifik — jawab umum soal server ini.";
return`Kamu adalah chatbot Discord Watcher — temen ngobrol yang tau keadaan server, dan kamu PUNYA AKSES ke data server lewat tools.
Data server saat ini:
- Pesan: ${insights.total_messages}
- User aktif: ${insights.active_users}
- Flagged: ${insights.flagged}
- Warning: ${insights.warned}
${scopeLine}
ATURAN PENTING — JANGAN PAKAI KONTEKS STATIS:
- Kamu TIDAK punya hafalan soal angka server (jumlah pesan, user aktif, flagged, dll). JANGAN tebak atau karang angka.
- Untuk SEMUA pertanyaan soal data server (jumlah pesan, user aktif, channel ramai, aktivitas terbaru, pesan di-flag), WAJIB panggil tool yang sesuai (get_server_stats, get_top_channels, get_recent_activity, get_top_flagged). Jawab HANYA dari hasil tool.
- Tool otomatis di-scope ke guild/channel di atas — kalau argumen guildId/channelId kosong, biarkan kosong (sudah otomatis ter-isi). Jangan isi ID yang kamu tebak.
- Kalau tool balas error atau kosong, bilang aja data lagi ga ketemu, jangan karang.
Gaya ngobrol:
- Santai, hangat, kayak ngobrol sama temen
- Pake Bahasa Indonesia sehari-hari, ga perlu kaku
- Sesekali pake emoji wajar aja, ga berlebihan
- Kalo ditanya sesuatu yang kamu tau dari data server, jawab pake data itu
- Kalo ga tau atau ga nyambung, bilang aja terus tanya balik biar ngobrolnya jalan
- Jangan sebut "rule", "instruksi", "prompt" atau apapun soal cara kamu berpikir
- Kalo ditanya di luar data server dan kamu ga tau, bilang aja terus tanya balik biar ngobrolnya jalan
- Jangan sebut "rule", "instruksi", "prompt", "tool", atau apapun soal cara kamu berpikir
* Static tool *definitions* for the chatbot LLM (OpenAI function-calling
* format). Kept separate from the executor (chatbot.tools.ts) so the schema
* the model depends on can be imported without pulling in the database /
* config layer.
*
* The chatbot is a server-watcher agent: it can answer about ANY server
* situation — activity, moderation queue, specific users, channels, voice
* recordings, AI correction history, and trends over time — by calling these
* tools, which the executor implements against real tables.
*/
exportinterfaceToolDef{
type:"function";
function:{
name: string;
description: string;
parameters:{
type:"object";
properties: Record<string,unknown>;
required?: string[];
};
};
}
exportconsttools: ToolDef[]=[
{
type:"function",
function:{
name:"get_server_stats",
description:
"Ambil statistik ringkas server/guild: total pesan, user aktif, jumlah pesan flagged, warn, dan clean. Panggil untuk jawab pertanyaan umum soal kondisi server. guildId/channelId otomatis ter-isi dari scope; kosongkan untuk semua data.",
parameters:{
type:"object",
properties:{
guildId:{type:"string",description:"ID server (opsional)."},
"Cari pesan berdasarkan kata kunci di isi pesan (case-insensitive, LIKE). Untuk 'ada yang bahas X gak?' / temukan topik tertentu. Hindari kata terlalu umum.",
parameters:{
type:"object",
properties:{
query:{
type:"string",
description:"Kata kunci pencarian (wajib).",
},
guildId:{type:"string",description:"ID server (opsional)."},
limit:{type:"number",description:"Jumlah hasil (default 5)."},
},
required:["query"],
},
},
},
{
type:"function",
function:{
name:"get_user_messages",
description:
"Ambil pesan terbaru dari satu user tertentu (user_id), opsional di-scope ke guild/channel. Untuk 'chat si A gimana akhir-akhir ini?' — butuh user_id.",
parameters:{
type:"object",
properties:{
userId:{type:"string",description:"ID user (wajib)."},
guildId:{type:"string",description:"ID server (opsional)."},
"Ambil ringkasan profil AI dari seorang user (pola perilaku, gaya bicara) dari tabel user_profiles. Untuk 'siapa si A?' / konteks perilaku. Butuh user_id.",
parameters:{
type:"object",
properties:{
userId:{type:"string",description:"ID user (wajib)."},
guildId:{type:"string",description:"ID server (opsional)."},
},
required:["userId"],
},
},
},
{
type:"function",
function:{
name:"get_user_reputation",
description:
"Ambil skor trust, jumlah infraction, dan streak pesan bersih seorang user dari user_reputations. Untuk 'berapa trust score si A?' / riwayat pelanggaran. Butuh user_id.",
parameters:{
type:"object",
properties:{
userId:{type:"string",description:"ID user (wajib)."},
guildId:{type:"string",description:"ID server (opsional)."},
},
required:["userId"],
},
},
},
{
type:"function",
function:{
name:"get_channel_culture",
description:
"Ambil ringkasan norma/slang channel dari tabel channel_cultures (AI-generated). Untuk 'norma channel ini gimana?' / konteks sebelum nge-flag. Butuh channel_id.",
"Ambil rekaman suara terbaru (voice_recordings): user, channel, transkripsi, status upload. Untuk 'ada rekaman suara terbaru?' / cek transkripsi. Bisa di-scope ke user_id atau channel_id.",
parameters:{
type:"object",
properties:{
userId:{type:"string",description:"Filter user (opsional)."},
channelId:{
type:"string",
description:"Filter channel (opsional).",
},
guildId:{type:"string",description:"ID server (opsional)."},
"Ambil tren harian: per hari, jumlah total pesan vs flagged vs warn vs clean. Untuk 'minggu ini pelanggaran naik?' / lihat tren moderasi. guildId otomatis ter-isi.",
parameters:{
type:"object",
properties:{
guildId:{type:"string",description:"ID server (opsional)."},
description:"Jumlah hari ke belakang (default 14, max 60).",
},
},
},
},
},
{
type:"function",
function:{
name:"get_corrections",
description:
"Ambil riwayat koreksi false-positive AI (corrected_moderations): pesan yang awalnya di-flag tapi dikoreksi manusia, beserta alasannya. Untuk 'AI pernah salah nge-flag apa aja?' / audit akurasi moderasi.",
parameters:{
type:"object",
properties:{
guildId:{type:"string",description:"ID server (opsional)."},
* Tools the chatbot LLM can call. Definitions describe the schema to the
* model; the executor implements each one against the real database.
* This turns the chatbot from "blind stats guesser" into an agent that
* pulls real, current server data on demand.
* Executor for the chatbot's server-watcher tools. The tool *definitions*
* live in chatbot.toolDefs.ts (no DB import); this file implements each one
* against the real database.
*
* All queries use parameterized drizzle operators (eq/like/and) — never string
* interpolation into raw SQL — so model-supplied arguments cannot inject SQL.
*/
exporttypeToolResult=string;
/** JSON schema for a tool definition (OpenAI function-calling format). */
exportinterfaceToolDef{
type:"function";
function:{
name: string;
description: string;
parameters:{
type:"object";
properties: Record<string,unknown>;
required?: string[];
};
};
}
exportconsttools: ToolDef[]=[
{
type:"function",
function:{
name:"get_server_stats",
description:
"Ambil statistik ringkas server/guild saat ini: total pesan, user aktif, jumlah pesan flagged, dan jumlah warning. Panggil ini untuk menjawab pertanyaan umum tentang kondisi server. Opsional fill guild_id untuk scope ke guild tertentu, channel_id untuk scope ke channel.",
parameters:{
type:"object",
properties:{
guildId:{
type:"string",
description:"ID guild/server (opsional). Kosongkan = semua data.",
},
channelId:{
type:"string",
description:"ID channel (opsional).",
},
},
},
},
},
{
type:"function",
function:{
name:"get_top_channels",
description:
"Ambil daftar channel paling aktif (jumlah pesan terbanyak) di server. Panggil buat jawab 'channel mana paling ramai' atau aktivitas per-channel.",
parameters:{
type:"object",
properties:{
guildId:{
type:"string",
description:"ID server (opsional).",
},
limit:{
type:"number",
description:"Jumlah channel teratas (default 5, max 10).",
},
},
},
},
},
{
type:"function",
function:{
name:"get_recent_activity",
description:
"Ambil aktivitas/pesan terbaru di server: siapa yang baru ngomong, di channel mana, jam berapa. Panggil buat jawaban soal 'lagi ngapain' / aktivitas terbaru di server.",
parameters:{
type:"object",
properties:{
guildId:{
type:"string",
description:"ID server (opsional).",
},
limit:{
type:"number",
description:"Jumlah pesan terakhir (default 5).",
},
},
},
},
},
{
type:"function",
function:{
name:"get_top_flagged",
description:
"Ambil pesan yang paling sering di-flag atau kena warning. Panggil buat jawab soal pesan bermasalah / moderator.",
parameters:{
type:"object",
properties:{
guildId:{
type:"string",
description:"ID server (opsional).",
},
limit:{
type:"number",
description:"Jumlah pesan (default 5).",
},
},
},
},
},
];
/** Executes a tool call against the real DB and returns a readable result. */
exportasyncfunctionexecuteTool(
name: string,
@@ -122,9 +31,11 @@ export async function executeTool(
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
asepharyana