asepharyana
ef7708bf7d
GMW moves off omniroute (100.121.180.82:20128) and off the direct NVIDIA vision endpoint onto 9router, which runs on the same host as both services (127.0.0.1:4014) — loopback avoids the TLS/proxy hop and localhost calls bypass 9router's remote-key guard. - gateway + backend: AI_LLM_BASE_URL default -> http://127.0.0.1:4014/v1 - drop stale 'omniroute' router references from comments/docs now that the active router is 9router (llmClient, llmCaller, ARCHITECTURE, AGENTS) Verified against 9router before wiring: model 'text' -> gemini-3.5-flash-lite (SSE, as the pipeline expects), 'multimodal' -> nemotron-3-nano-omni answers image input, and gemini/gemini-embedding-001 returns 3072 dims — matching the existing Qdrant collections (no reindex needed). The GMW key is already registered in 9router's apiKeys table. typecheck + lint + tests green (gateway 138, backend 37 excluding e2e).
Discord Gateway
Event-driven selfbot service: captures Discord events, runs LLM moderation, publishes everything to Redis for the backend to consume.
Architecture, invariants and the AI pipeline are documented in
ARCHITECTURE.md— that file is the source of truth. This README only covers how to run it.
Commands
pnpm install
pnpm typecheck # tsc --noEmit
pnpm lint # biome check --diagnostic-level=error .
pnpm test # vitest run (138 tests)
pnpm build # tsc — CI/prod builds run this inside nix, which also
# runs scripts/fix-imports.mjs to rewrite @/ aliases and
# extensionless imports for Node ESM
pnpm dev # tsx watch src/index.ts
pnpm start # node dist/index.js
Deployment is CI-only: nix build .#discord-gateway → Attic cache → systemd
restart on the VPS. Do not build/hand-copy the artifact.
Layout
src/
├── index.ts # entry → initializeDiscordGateway()
├── app/ # process lifecycle
│ ├── bootstrap.ts # startup order: config → DB → services → metrics → login
│ ├── lifecycle.ts # everything wired on the Discord 'ready' hook
│ ├── process-guards.ts # SIGINT/SIGTERM + uncaught error policy
│ ├── metrics-collector.ts # AI pipeline Prometheus gauges
│ ├── shutdown.ts # graceful shutdown sequence
│ └── retention.ts # expired-record cleanup scheduler
├── shared/ # infrastructure — never imports from modules/
│ ├── config/ database/ logger/ errors/ utils/
│ ├── discord/clientOptions.ts
│ ├── redis-channels.ts # canonical Redis channel + command constants
│ └── moderation-types.ts # domain types shared across services
└── modules/ # feature modules (each exposes an index.ts facade)
├── ai-moderation/ # LLM moderation pipeline (largest module)
├── message-capture/ # Discord listeners + message/attachment DB
├── attachment-upload/ # download → resize → upload
├── event-broadcaster/ # Redis pub/sub publisher
├── command-handler/ # backend → gateway commands over Redis
├── gateway-metrics/ # Prometheus /metrics (METRICS_PORT)
├── monitor/ # weekly digest scheduler
└── reaction-tracking/ thread-tracking/ user-presence/
channel-topic/ guild-member-events/
Dependency direction is one-way: index.ts → app/ → modules/ → shared/.
Callers outside a module import its index.ts facade, never an internal file.
Testing
Vitest, tests in tests/. Config supplies dummy env vars so the suite runs
without live Postgres/Redis/Qdrant; external services are mocked. llmE2e.test.ts
is skipped by default and needs real credentials (pnpm test:e2e:live).