History

Luiz Gustavo 504b20fa5c search: gate dense recall by cosine-distance threshold in the RPC Root-cause fix for "search returns garbage for absent terms". The hybrid RPC's dense branch always returned its k nearest vectors regardless of distance, so a query for a term not in the corpus (e.g. "varginha") surfaced unrelated chunks. The cross-encoder reranker would filter these but costs 18-62s on CPU — unusable for interactive search. Add max_dense_dist (default 0.40) to hybrid_search_chunks: dense neighbours beyond that cosine distance are dropped server-side. Calibrated from measured distances — strong semantic match ~0.12-0.20, no real match ~0.46-0.53. BM25 full-text still matches literal terms; the reranker becomes opt-in refinement. Verified live: varginha/abducao → 0, disco voador/roswell → relevant, all <1s. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>		2026-05-21 16:36:56 -03:00
..
coolify	baseline: Disclosure Bureau pipeline + Next.js UI + Supabase stack	2026-05-17 22:44:36 -03:00
disclosure-stack	ship: synthesize 158 entities, AG-UI artifacts, chat persistence, auth flow	2026-05-18 03:52:59 -03:00
embed-service	baseline: Disclosure Bureau pipeline + Next.js UI + Supabase stack	2026-05-17 22:44:36 -03:00
supabase	search: gate dense recall by cosine-distance threshold in the RPC	2026-05-21 16:36:56 -03:00
DEPLOY-CHECKLIST.md	baseline: Disclosure Bureau pipeline + Next.js UI + Supabase stack	2026-05-17 22:44:36 -03:00
README.md	baseline: Disclosure Bureau pipeline + Next.js UI + Supabase stack	2026-05-17 22:44:36 -03:00
RETRIEVAL.md	baseline: Disclosure Bureau pipeline + Next.js UI + Supabase stack	2026-05-17 22:44:36 -03:00

README.md

Infrastructure — Disclosure Bureau

Self-hosted stack on a single VPS (16 GB / 4 CPU / 200 GB NVMe) managed via Coolify.

                   Internet (443/80)
                          │
                ┌─────────▼─────────┐
                │  Caddy (Coolify)  │ ← auto-TLS Let's Encrypt
                └────┬──────────────┘
                     │
       ┌─────────────┼──────────────────────┬──────────────────┐
       ▼             ▼                      ▼                  ▼
  ┌─────────┐   ┌──────────┐         ┌──────────┐       ┌──────────┐
  │ Next.js │   │ Supabase │         │ Supabase │       │  shared  │
  │   web   │   │ disclosure│        │ project-B │       │ services │
  │ :3000   │   │ stack    │         │  stack    │       │ Meili··· │
  └─────────┘   │  ┌─────┐ │         │  ┌─────┐ │       │ Imgproxy │
                │  │PG/GT│ │         │  │PG/GT│ │       │ Dragonfly│
                │  └─────┘ │         │  └─────┘ │       └──────────┘
                └──────────┘         └──────────┘
                disclosure.top       projeto-b.com

Components

Layer	Service	Notes
Orchestration	Coolify v4	Self-hosted PaaS — manages all containers, TLS, backups
Database + Auth + Storage	Supabase self-hosted (one per project)	Each project gets own Postgres + GoTrue + Storage
Frontend	Next.js 15 (this repo's `/web`)	Deployed via Coolify Git integration
Search	Meilisearch (shared)	Full-text search across pages + entities
Cache + Queue	Dragonfly (shared)	Redis-compatible, multi-threaded
Images	Imgproxy (shared)	On-the-fly resize / WebP conversion
Backups	restic + Backblaze B2	Nightly Postgres + Storage dumps

Quick path

coolify/INSTALL.md — install Coolify on the fresh VPS (~10 min)
coolify/SUPABASE.md — create the disclosure Supabase project (~5 min)
Run supabase/migrations/0001_chat_schema.sql via Supabase Studio SQL editor
coolify/NEXTJS.md — deploy the /web app pointing at the Supabase URL
coolify/SHARED.md — bring up Meilisearch, Dragonfly, Imgproxy

Adding more projects later

For each new project, repeat step 2 (new Supabase project in Coolify UI) and step 4 (new Next.js app). They get their own subdomain, own auth, own data. Total isolation.

Local development

For dev on macOS/Linux without the VPS, see ../web/README.md — uses the Supabase CLI to spin up a local stack on localhost:54321.