feat: Phase 0 bootstrap — Qdrant deploy e schema PostgreSQL

- README.md: contesto ALPHA_PROJECT, architettura multi-agent, stack infrastrutturale - CHANGELOG.md: documenta deploy Qdrant v1.17.0 e creazione database pompeo - db/postgres.sql: schema DDL database pompeo (user_profile, memory_facts, finance_documents, behavioral_context, agent_messages) con multi-tenancy user_id - db/qdrant.sh: script per creazione/ripristino collections Qdrant (episodes, knowledge, preferences) con payload indexes Design decisions: - Multi-tenancy via user_id su Qdrant e PostgreSQL (estendibile a nuovi utenti senza modifiche infrastrutturali) - agent_messages come blackboard persistente per il Proactive Arbiter Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-03-21 10:53:11 +00:00
parent 266d04f06d
commit adb36c0c48
4 changed files with 437 additions and 19 deletions
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -0,0 +1,154 @@
+# ALPHA_PROJECT — Changelog
+
+Tutte le modifiche significative al progetto ALPHA_PROJECT sono documentate qui.
+
+---
+
+## [2026-03-21] PostgreSQL — Database "pompeo" e schema ALPHA_PROJECT
+
+### Overview
+
+Creato il database `pompeo` sul cluster Patroni (namespace `persistence`) e applicato lo schema iniziale per la memoria strutturata di Pompeo. Seconda milestone della Phase 0 — Infrastructure Bootstrap.
+
+---
+
+### Modifica manifest Patroni
+
+Aggiunto `pompeo: martin` nella sezione `databases` di `infra/cluster/persistence/patroni/postgres.yaml`. Il database è stato creato automaticamente dallo Zalando Operator senza downtime sugli altri database.
+
+Script DDL idempotente disponibile in: `alpha/db/postgres.sql`
+
+---
+
+### Design decision — Multi-tenancy anche in PostgreSQL
+
+Coerentemente con la scelta adottata per Qdrant, tutte le tabelle includono il campo `user_id TEXT NOT NULL DEFAULT 'martin'`. I valori `'martin'` e `'shared'` sono seedati in `user_profile` come utenti iniziali del sistema.
+
+Aggiungere un nuovo utente in futuro non richiede modifiche allo schema — è sufficiente inserire una riga in `user_profile` e usare il nuovo `user_id` negli INSERT.
+
+---
+
+### Design decision — agent_messages come blackboard persistente
+
+La tabella `agent_messages` implementa il **blackboard pattern** del message broker: ogni agente n8n inserisce le proprie osservazioni con `arbiter_decision = NULL` (pending). Il Proactive Arbiter legge i messaggi in coda, decide (`notify` / `defer` / `discard`) e aggiorna `arbiter_decision`, `arbiter_reason` e `processed_at`.
+
+Rispetto a usare solo NATS/Redis come broker, questo approccio garantisce un **audit log permanente** di tutte le osservazioni e decisioni, interrogabile via SQL per debug, tuning e analisi storiche.
+
+---
+
+### Schema creato
+
+**5 tabelle** nel database `pompeo`:
+
+| Tabella | Ruolo |
+|---|---|
+| `user_profile` | Preferenze statiche per utente (lingua, timezone, stile notifiche, quiet hours). Seed: `martin`, `shared` |
+| `memory_facts` | Fatti episodici prodotti da tutti gli agenti, con TTL (`expires_at`) e riferimento al punto Qdrant (`qdrant_id`) |
+| `finance_documents` | Documenti finanziari strutturati: bollette, fatture, cedolini. Include `raw_text` per embedding |
+| `behavioral_context` | Contesto IoT/comportamentale per l'Arbiter: DND, home presence, tipo evento |
+| `agent_messages` | Blackboard del message broker — osservazioni agenti + decisioni Arbiter |
+
+**15 index** totali:
+
+| Index | Tabella | Tipo |
+|---|---|---|
+| `idx_memory_facts_user_source_cat` | `memory_facts` | `(user_id, source, category)` |
+| `idx_memory_facts_expires` | `memory_facts` | `(expires_at)` WHERE NOT NULL |
+| `idx_memory_facts_action` | `memory_facts` | `(user_id, action_required)` WHERE true |
+| `idx_finance_docs_user_date` | `finance_documents` | `(user_id, doc_date DESC)` |
+| `idx_finance_docs_correspondent` | `finance_documents` | `(user_id, correspondent)` |
+| `idx_behavioral_ctx_user_time` | `behavioral_context` | `(user_id, start_at, end_at)` |
+| `idx_behavioral_ctx_dnd` | `behavioral_context` | `(user_id, do_not_disturb)` WHERE true |
+| `idx_agent_msgs_pending` | `agent_messages` | `(user_id, priority, created_at)` WHERE pending |
+| `idx_agent_msgs_agent_type` | `agent_messages` | `(agent, event_type, created_at)` |
+| `idx_agent_msgs_expires` | `agent_messages` | `(expires_at)` WHERE pending AND NOT NULL |
+
+---
+
+### Phase 0 — Stato aggiornato
+
+- [x] ~~Deploy **Qdrant** sul cluster~~ ✅ 2026-03-21
+- [x] ~~Collections Qdrant con multi-tenancy `user_id`~~ ✅ 2026-03-21
+- [x] ~~Payload indexes Qdrant~~ ✅ 2026-03-21
+- [x] ~~Database `pompeo` + schema PostgreSQL~~ ✅ 2026-03-21
+- [ ] Verify embedding endpoint via Copilot (`text-embedding-3-small`)
+- [ ] Migrazione a Ollama `nomic-embed-text` (quando LLM server è online)
+
+---
+
+## [2026-03-21] Qdrant — Deploy e setup collections (Phase 0)
+
+### Overview
+
+Completato il deploy di **Qdrant v1.17.0** sul cluster Kubernetes (namespace `persistence`) e la creazione delle collections per la memoria semantica di Pompeo. Questa è la prima milestone della Phase 0 — Infrastructure Bootstrap.
+
+---
+
+### Deploy infrastruttura
+
+Qdrant deployato via Helm chart ufficiale (`qdrant/qdrant`) nel namespace `persistence`, coerente con il pattern infrastrutturale esistente (Longhorn storage, Sealed Secrets, ServiceMonitor Prometheus).
+
+**Risorse create:**
+
+| Risorsa | Dettaglio |
+|---|---|
+| StatefulSet `qdrant` | 1/1 pod Running, image `qdrant/qdrant:v1.17.0` |
+| PVC `qdrant-storage-qdrant-0` | 20Gi Longhorn RWO |
+| Service `qdrant` | ClusterIP — porte 6333 (REST), 6334 (gRPC), 6335 (p2p) |
+| SealedSecret `qdrant-api-secret` | API key cifrata, namespace `persistence` |
+| ServiceMonitor `qdrant` | Prometheus scraping su `:6333/metrics`, label `release: monitoring` |
+
+**Endpoint interno:** `qdrant.persistence.svc.cluster.local:6333`
+
+Manifest in: `infra/cluster/persistence/qdrant/`
+
+---
+
+### Design decision — Multi-tenancy collections (Opzione B)
+
+**Problema affrontato**: nominare le collections `martin_episodes`, `martin_knowledge`, `martin_preferences` avrebbe vincolato Pompeo ad essere esclusivamente un assistente personale singolo, rendendo impossibile — senza migration — estendere il sistema ad altri membri della famiglia in futuro.
+
+**Scelta adottata**: architettura multi-tenant con 3 collection condivise e isolamento via campo `user_id` nel payload di ogni punto vettoriale.
+
+```
+episodes      ← user_id: "martin" | "shared" | <futuri utenti>
+knowledge     ← user_id: "martin" | "shared" | <futuri utenti>
+preferences   ← user_id: "martin" | "shared" | <futuri utenti>
+```
+
+Il valore `"shared"` è riservato a dati della casa/famiglia visibili a tutti gli utenti (es. calendario condiviso, documenti di casa, finanze comuni). Le query n8n usano un filtro `should: [user_id=martin, user_id=shared]` per recuperare sia il contesto personale che quello condiviso.
+
+**Vantaggi**: aggiungere un nuovo utente domani non richiede alcuna modifica infrastrutturale — solo includere il nuovo `user_id` negli upsert e nelle query.
+
+---
+
+### Collections create
+
+Tutte e 3 le collections sono operative (status `green`):
+
+| Collection | Contenuto |
+|---|---|
+| `episodes` | Fatti episodici con timestamp (email, IoT, calendario, conversazioni) |
+| `knowledge` | Documenti, note Outline, newsletter, knowledge base |
+| `preferences` | Preferenze, abitudini e pattern comportamentali per utente |
+
+**Payload schema comune** (5 index su ogni collection):
+
+| Campo | Tipo | Scopo |
+|---|---|---|
+| `user_id` | keyword | Filtro multi-tenant (`"martin"`, `"shared"`) |
+| `source` | keyword | Origine del dato (`"email"`, `"calendar"`, `"iot"`, `"paperless"`, …) |
+| `category` | keyword | Dominio semantico (`"finance"`, `"work"`, `"personal"`, …) |
+| `date` | datetime | Timestamp del fatto — filtrabile per range |
+| `action_required` | bool | Flag per il Proactive Arbiter |
+
+**Dimensione vettori**: 1536 (compatibile con `text-embedding-3-small` via GitHub Copilot — bootstrap phase). Da rivedere alla migrazione verso `nomic-embed-text` su Ollama.
+
+---
+
+### Phase 0 — Stato al momento del deploy Qdrant
+
+- [x] ~~Deploy **Qdrant** sul cluster~~
+- [x] ~~Creazione collections con multi-tenancy `user_id`~~
+- [x] ~~Payload indexes: `user_id`, `source`, `category`, `date`, `action_required`~~
+- [x] ~~Run **PostgreSQL migrations** su Patroni~~ ✅ completato nella sessione stessa
--- a/README.md
+++ b/README.md
@@ -42,7 +42,7 @@ Production-grade self-hosted stack. Key components relevant to ALPHA_PROJECT:
 | **n8n** | Primary orchestrator and workflow engine for all agents |
 | **Node-RED** | Event-driven automation, Home Assistant bridge |
 | **Patroni / PostgreSQL** | Persistent structured memory store |
-| **Qdrant** | Vector store for semantic/episodic memory *(to be deployed)* |
+| **Qdrant** | Vector store for semantic/episodic memory — `qdrant.persistence.svc.cluster.local:6333` |
 | **NATS / Redis Streams** | Message broker between agents *(to be chosen and deployed)* |
 | **Authentik** | SSO / IAM (OIDC) |
 | **Home Assistant** | IoT hub — device tracking, automations, sensors |
@@ -160,17 +160,17 @@ CREATE TABLE behavioral_context (
 );
 ```

-**2. Semantic memory — Qdrant**
+**2. Semantic memory — Qdrant** — `qdrant.persistence.svc.cluster.local:6333`

-Vector embeddings for similarity search. Three collections:
+Vector embeddings for similarity search. Three collections with **multi-tenant design**: isolation via `user_id` payload field (`"martin"`, `"shared"`, future users).

 | Collection | Content |
 |---|---|
-| `martin_episodes` | Conversations, episodic facts with timestamp |
-| `martin_knowledge` | Documents, Outline notes, newsletters, knowledge base |
-| `martin_preferences` | Preferences, habits, behavioral patterns |
+| `episodes` | Conversations, episodic facts with timestamp |
+| `knowledge` | Documents, Outline notes, newsletters, knowledge base |
+| `preferences` | Preferences, habits, behavioral patterns |

-Each Qdrant point includes a metadata payload for pre-filtering (source, date, category, action_required) to avoid full-scan similarity searches.
+Each Qdrant point includes a metadata payload for pre-filtering (`user_id`, `source`, `date`, `category`, `action_required`) to avoid full-scan similarity searches.

 **3. Profile memory — PostgreSQL (static table)**

@@ -266,12 +266,15 @@ Notification is sent via **Amazon Echo / Pompeo** (TTS) for voice, and **Telegra

 ### Phase 0 — Infrastructure Bootstrap *(prerequisite for everything)*

- [ ] Deploy **Qdrant** on the Kubernetes cluster
-  - Create collections: `martin_episodes`, `martin_knowledge`, `martin_preferences`
-  - Configure payload indexes on: `source`, `category`, `date`, `action_required`
- [ ] Run **PostgreSQL migrations** on Patroni
-  - Create tables: `memory_facts`, `finance_documents`, `behavioral_context`
-  - Add index on `memory_facts(source, category, expires_at)`
+- [x] ~~Deploy **Qdrant** on the Kubernetes cluster~~ ✅ 2026-03-21
+  - Collections: `episodes`, `knowledge`, `preferences` (multi-tenant via `user_id` payload field)
+  - Payload indexes: `user_id`, `source`, `category`, `date`, `action_required`
+  - Endpoint: `qdrant.persistence.svc.cluster.local:6333`
+- [x] ~~Run **PostgreSQL migrations** on Patroni~~ ✅ 2026-03-21
+  - Database `pompeo` creato (Zalando Operator)
+  - Tabelle: `user_profile`, `memory_facts`, `finance_documents`, `behavioral_context`, `agent_messages`
+  - Multi-tenancy: campo `user_id` su tutte le tabelle, seed `martin` + `shared`
+  - Script DDL: `alpha/db/postgres.sql`
 - [ ] Verify embedding endpoint via Copilot (`text-embedding-3-small`) as bootstrap fallback
 - [ ] Plan migration to local Ollama embedding model once LLM server is online

@@ -281,13 +284,13 @@ Notification is sent via **Amazon Echo / Pompeo** (TTS) for voice, and **Telegra

 - [ ] **Daily Digest**: after `Parse risposta GPT-4.1`, add:
  - Postgres INSERT into `memory_facts` (source=email, category, subject, detail JSONB, action_required, expires_at)
-  - Embedding generation (Copilot endpoint) → Qdrant upsert into `martin_episodes`
+  - Embedding generation (Copilot endpoint) → Qdrant upsert into `episodes` (user_id=martin)
  - Thread dedup: use `thread_id` as logical key, update existing Qdrant point if thread already exists

 - [ ] **Upload Bolletta** + **Upload Documento (Telegram)**: after `Paperless - Patch Metadati`, add:
  - Postgres INSERT into `finance_documents` (correspondent, amount, doc_date, doc_type, tags, paperless_doc_id)
  - Postgres INSERT into `memory_facts` (source=paperless, category=finance, cross-reference)
-  - Embedding of OCR text chunks → Qdrant upsert into `martin_knowledge`
+  - Embedding of OCR text chunks → Qdrant upsert into `knowledge` (user_id=martin)

 ---

@@ -316,7 +319,7 @@ Notification is sent via **Amazon Echo / Pompeo** (TTS) for voice, and **Telegra

 - [ ] **Newsletter Agent**
  - Separate Gmail label for newsletters (excluded from Daily Digest main flow)
-  - Morning cron: summarize + extract relevant articles → `martin_knowledge`
+  - Morning cron: summarize + extract relevant articles → `knowledge`

 ---

@@ -351,9 +354,9 @@ Notification is sent via **Amazon Echo / Pompeo** (TTS) for voice, and **Telegra
  - PDF → FileWizard OCR → GPT-4.1 metadata extraction (month, gross, net, deductions)
  - Paperless upload with tag `Cedolino`
  - Persist structured data to `finance_documents` (custom fields for payslip)
-  - Trend embedding in `martin_knowledge` for finance agent queries
- [ ] Behavioral habit modeling: aggregate `behavioral_context` records over time, generate periodic "habit summary" embeddings in `martin_preferences`
- [ ] Outline → Qdrant pipeline: sync selected Outline documents into `martin_knowledge` on edit/publish event
+  - Trend embedding in `knowledge` for finance agent queries
+- [ ] Behavioral habit modeling: aggregate `behavioral_context` records over time, generate periodic "habit summary" embeddings in `preferences`
+- [ ] Outline → Qdrant pipeline: sync selected Outline documents into `knowledge` on edit/publish event
 - [ ] Chrome browsing history ingestion (privacy-filtered): evaluate browser extension or local export → embedding pipeline for interest/preference modeling
 - [ ] "Posti e persone" graph: structured contact/location model in Postgres, populated from email senders, calendar attendees, Home Assistant presence data
 - [ ] Local embedding model: migrate from Copilot `text-embedding-3-small` to Ollama-served model (e.g. `nomic-embed-text`) once LLM server is stable
--- a/db/postgres.sql
+++ b/db/postgres.sql
@@ -0,0 +1,170 @@
+-- =============================================================================
+-- ALPHA_PROJECT — Database "pompeo" — Schema iniziale
+-- =============================================================================
+-- Applicare su: postgresql://martin@postgres.persistence.svc.cluster.local:5432/pompeo
+--
+-- Esecuzione dal cluster:
+--   sudo microk8s kubectl run psql-pompeo --rm -it \
+--     --image=postgres:17-alpine --namespace=persistence \
+--     --env="PGPASSWORD=<password>" --restart=Never \
+--     -- psql "postgresql://martin@postgres:5432/pompeo" -f /dev/stdin < postgres.sql
+--
+-- Esecuzione via port-forward:
+--   sudo microk8s kubectl port-forward svc/postgres -n persistence 5432:5432
+--   psql "postgresql://martin@localhost:5432/pompeo" -f postgres.sql
+-- =============================================================================
+
+\c pompeo
+
+-- ---------------------------------------------------------------------------
+-- Estensioni
+-- ---------------------------------------------------------------------------
+CREATE EXTENSION IF NOT EXISTS "uuid-ossp";
+CREATE EXTENSION IF NOT EXISTS "pg_trgm";  -- full-text similarity search su subject/detail
+
+
+-- =============================================================================
+-- 1. USER_PROFILE
+--    Preferenze statiche per utente. Aggiornata manualmente o via agent action.
+--    user_id 'shared' = preferenze della casa (visibili a tutti).
+-- =============================================================================
+CREATE TABLE IF NOT EXISTS user_profile (
+  user_id            TEXT PRIMARY KEY,
+  display_name       TEXT,
+  language           TEXT        NOT NULL DEFAULT 'it',
+  timezone           TEXT        NOT NULL DEFAULT 'Europe/Rome',
+  notification_style TEXT        NOT NULL DEFAULT 'concise',  -- 'concise' | 'verbose'
+  quiet_start        TIME        NOT NULL DEFAULT '23:00',
+  quiet_end          TIME        NOT NULL DEFAULT '07:00',
+  preferences        JSONB,      -- freeform: soglie, preferenze extra per agente
+  updated_at         TIMESTAMP   NOT NULL DEFAULT now()
+);
+
+-- Utenti iniziali
+INSERT INTO user_profile (user_id, display_name) VALUES
+  ('martin', 'Martin'),
+  ('shared', 'Shared')
+ON CONFLICT (user_id) DO NOTHING;
+
+
+-- =============================================================================
+-- 2. MEMORY_FACTS
+--    Fatti episodici prodotti da tutti gli agenti. TTL tramite expires_at.
+--    qdrant_id: riferimento al punto vettoriale corrispondente nella collection "episodes".
+-- =============================================================================
+CREATE TABLE IF NOT EXISTS memory_facts (
+  id               UUID      PRIMARY KEY DEFAULT gen_random_uuid(),
+  user_id          TEXT      NOT NULL DEFAULT 'martin',
+  source           TEXT      NOT NULL,   -- 'email' | 'calendar' | 'iot' | 'paperless' | 'n8n' | ...
+  category         TEXT,                 -- 'finance' | 'personal' | 'work' | 'health' | ...
+  subject          TEXT,
+  detail           JSONB,                -- payload flessibile per-source
+  action_required  BOOLEAN   NOT NULL DEFAULT false,
+  action_text      TEXT,
+  created_at       TIMESTAMP NOT NULL DEFAULT now(),
+  expires_at       TIMESTAMP,            -- NULL = permanente
+  qdrant_id        UUID                  -- FK logico → collection "episodes"
+);
+
+CREATE INDEX IF NOT EXISTS idx_memory_facts_user_source_cat
+  ON memory_facts(user_id, source, category);
+
+CREATE INDEX IF NOT EXISTS idx_memory_facts_expires
+  ON memory_facts(expires_at)
+  WHERE expires_at IS NOT NULL;
+
+CREATE INDEX IF NOT EXISTS idx_memory_facts_action
+  ON memory_facts(user_id, action_required)
+  WHERE action_required = true;
+
+
+-- =============================================================================
+-- 3. FINANCE_DOCUMENTS
+--    Documenti finanziari strutturati (bollette, fatture, cedolini).
+--    paperless_doc_id: riferimento al documento in Paperless-ngx.
+-- =============================================================================
+CREATE TABLE IF NOT EXISTS finance_documents (
+  id                UUID        PRIMARY KEY DEFAULT gen_random_uuid(),
+  user_id           TEXT        NOT NULL DEFAULT 'martin',
+  paperless_doc_id  INT,                  -- ID documento in Paperless-ngx
+  correspondent     TEXT,
+  amount            NUMERIC(10,2),
+  currency          TEXT        NOT NULL DEFAULT 'EUR',
+  doc_date          DATE,
+  doc_type          TEXT,                 -- 'bolletta' | 'fattura' | 'cedolino' | ...
+  tags              TEXT[],
+  raw_text          TEXT,                 -- testo OCR grezzo (per embedding)
+  created_at        TIMESTAMP   NOT NULL DEFAULT now()
+);
+
+CREATE INDEX IF NOT EXISTS idx_finance_docs_user_date
+  ON finance_documents(user_id, doc_date DESC);
+
+CREATE INDEX IF NOT EXISTS idx_finance_docs_correspondent
+  ON finance_documents(user_id, correspondent);
+
+
+-- =============================================================================
+-- 4. BEHAVIORAL_CONTEXT
+--    Contesto comportamentale prodotto dall'IoT Agent e dal Calendar Agent.
+--    Usato dal Proactive Arbiter per rispettare DND e stimare presence.
+-- =============================================================================
+CREATE TABLE IF NOT EXISTS behavioral_context (
+  id                        UUID      PRIMARY KEY DEFAULT gen_random_uuid(),
+  user_id                   TEXT      NOT NULL DEFAULT 'martin',
+  event_type                TEXT,     -- 'sport_event' | 'dog_walk' | 'work_session' | 'commute' | ...
+  start_at                  TIMESTAMP,
+  end_at                    TIMESTAMP,
+  do_not_disturb            BOOLEAN   NOT NULL DEFAULT false,
+  home_presence_expected    BOOLEAN,
+  notes                     TEXT,
+  created_at                TIMESTAMP NOT NULL DEFAULT now()
+);
+
+CREATE INDEX IF NOT EXISTS idx_behavioral_ctx_user_time
+  ON behavioral_context(user_id, start_at, end_at);
+
+CREATE INDEX IF NOT EXISTS idx_behavioral_ctx_dnd
+  ON behavioral_context(user_id, do_not_disturb)
+  WHERE do_not_disturb = true;
+
+
+-- =============================================================================
+-- 5. AGENT_MESSAGES
+--    Blackboard: ogni agente pubblica qui le proprie osservazioni.
+--    Il Proactive Arbiter legge, decide (notify/defer/discard) e aggiorna.
+--    Corrisponde al message schema definito in alpha/README.md.
+-- =============================================================================
+CREATE TABLE IF NOT EXISTS agent_messages (
+  id                UUID      PRIMARY KEY DEFAULT gen_random_uuid(),
+  agent             TEXT      NOT NULL,  -- 'mail' | 'calendar' | 'iot' | 'finance' | 'infra' | ...
+  priority          TEXT      NOT NULL,  -- 'low' | 'high'
+  event_type        TEXT      NOT NULL,  -- 'new_fact' | 'reminder' | 'alert' | 'behavioral_observation'
+  user_id           TEXT      NOT NULL DEFAULT 'martin',
+  subject           TEXT,
+  detail            JSONB,
+  source_ref        TEXT,                -- ID record Postgres o ref esterna
+  expires_at        TIMESTAMP,
+  arbiter_decision  TEXT,               -- NULL (pending) | 'notify' | 'defer' | 'discard'
+  arbiter_reason    TEXT,
+  created_at        TIMESTAMP NOT NULL DEFAULT now(),
+  processed_at      TIMESTAMP
+);
+
+CREATE INDEX IF NOT EXISTS idx_agent_msgs_pending
+  ON agent_messages(user_id, priority, created_at)
+  WHERE arbiter_decision IS NULL;
+
+CREATE INDEX IF NOT EXISTS idx_agent_msgs_agent_type
+  ON agent_messages(agent, event_type, created_at);
+
+CREATE INDEX IF NOT EXISTS idx_agent_msgs_expires
+  ON agent_messages(expires_at)
+  WHERE expires_at IS NOT NULL AND arbiter_decision IS NULL;
+
+
+-- =============================================================================
+-- Fine script
+-- =============================================================================
+\echo '✅ Schema pompeo applicato correttamente.'
+\echo '   Tabelle: user_profile, memory_facts, finance_documents, behavioral_context, agent_messages'
--- a/db/qdrant.sh
+++ b/db/qdrant.sh
@@ -0,0 +1,91 @@
+#!/usr/bin/env bash
+# =============================================================================
+# ALPHA_PROJECT — Qdrant — Setup collections e payload indexes
+# =============================================================================
+# Collections già create il 2026-03-21. Script conservato per tracciabilità
+# e disaster recovery (da eseguire su un'istanza Qdrant vuota).
+#
+# Prerequisiti:
+#   sudo microk8s kubectl port-forward svc/qdrant -n persistence 6333:6333
+#
+# Esecuzione:
+#   bash alpha/db/qdrant.sh
+# =============================================================================
+
+set -euo pipefail
+
+QDRANT_URL="${QDRANT_URL:-http://localhost:6333}"
+QDRANT_API_KEY="${QDRANT_API_KEY:-__Montecarlo00!}"
+
+# Dimensione vettori: 1536 = text-embedding-3-small (Copilot, bootstrap phase)
+# Da aggiornare a 768 alla migrazione verso nomic-embed-text su Ollama
+VECTOR_SIZE=1536
+
+header_key="api-key: ${QDRANT_API_KEY}"
+
+echo "==> Connessione a ${QDRANT_URL}"
+curl -sf "${QDRANT_URL}/" -H "${header_key}" | grep -o '"version":"[^"]*"'
+echo ""
+
+# -----------------------------------------------------------------------------
+# Collections
+# Architettura multi-tenant: isolamento via campo user_id nel payload.
+# Valori user_id: "martin" | "shared" | <futuri utenti>
+# -----------------------------------------------------------------------------
+for COL in episodes knowledge preferences; do
+  echo "==> Creazione collection: ${COL}"
+  curl -sf -X PUT "${QDRANT_URL}/collections/${COL}" \
+    -H "${header_key}" \
+    -H "Content-Type: application/json" \
+    -d "{
+      \"vectors\": { \"size\": ${VECTOR_SIZE}, \"distance\": \"Cosine\" },
+      \"optimizers_config\": { \"default_segment_number\": 2 },
+      \"replication_factor\": 1
+    }" | grep -o '"status":"[^"]*"'
+done
+
+echo ""
+
+# -----------------------------------------------------------------------------
+# Payload indexes (per pre-filtering efficiente prima della ricerca vettoriale)
+# -----------------------------------------------------------------------------
+for COL in episodes knowledge preferences; do
+  echo "==> Indexes per collection: ${COL}"
+
+  for FIELD in user_id source category; do
+    printf "   %-20s (keyword) → " "${FIELD}"
+    curl -sf -X PUT "${QDRANT_URL}/collections/${COL}/index" \
+      -H "${header_key}" \
+      -H "Content-Type: application/json" \
+      -d "{\"field_name\": \"${FIELD}\", \"field_schema\": \"keyword\"}" \
+      | grep -o '"status":"[^"]*"'
+  done
+
+  printf "   %-20s (datetime) → " "date"
+  curl -sf -X PUT "${QDRANT_URL}/collections/${COL}/index" \
+    -H "${header_key}" \
+    -H "Content-Type: application/json" \
+    -d '{"field_name": "date", "field_schema": "datetime"}' \
+    | grep -o '"status":"[^"]*"'
+
+  printf "   %-20s (bool) → " "action_required"
+  curl -sf -X PUT "${QDRANT_URL}/collections/${COL}/index" \
+    -H "${header_key}" \
+    -H "Content-Type: application/json" \
+    -d '{"field_name": "action_required", "field_schema": "bool"}' \
+    | grep -o '"status":"[^"]*"'
+done
+
+echo ""
+
+# -----------------------------------------------------------------------------
+# Verifica finale
+# -----------------------------------------------------------------------------
+echo "==> Collections attive:"
+curl -sf "${QDRANT_URL}/collections" -H "${header_key}" \
+  | python3 -c "import sys,json; [print('   -', c['name']) for c in json.load(sys.stdin)['result']['collections']]"
+
+echo ""
+echo "✅ Setup Qdrant completato."
+echo "   Collections: episodes, knowledge, preferences"
+echo "   Payload indexes: user_id, source, category, date, action_required"