# AI Session Summary — geonet-console

```yaml
session: 2026-07-06-ai-chat-endpoint
project: server-connection/geonet-console
version: ai-fase3
status: deployed
next: UI chat Super Apps PWA + user memory + ai.chat permission guard
agent: Cascade
branch: main
```

## TL;DR

- `POST /api/v1/ai/chat` + `GET /api/v1/ai/status` live di production (`console.gisportal.id`)
- RAG pipeline: embed query → pgvector `10.100.1.24:5433` → 5 context chunks → Ollama `10.100.1.14:11434` → jawaban
- Verified: query "route untuk feedback" → jawab `POST /api/v1/feedback` dengan akurat

## Files baru

| File | Fungsi |
|------|--------|
| `app/Services/AiChatService.php` | Embed query, similarity search pgvector, generate via Ollama |
| `app/Http/Controllers/Api/V1/AiChatController.php` | Handler `POST /api/v1/ai/chat` + `GET /api/v1/ai/status` |
| `config/pgvector.php` | Config pgvector — baca env `PGVECTOR_HOST/PORT/DB/USER/PASSWORD` |

## Files dimodifikasi

| File | Perubahan |
|------|-----------|
| `config/ollama.php` | Default URL → `10.100.1.14:11434`, model → `qwen2.5-coder:14b` |
| `routes/api.php` | Tambah 2 route AI Chat, import `AiChatController` |

## Env vars baru di server `.env`

```
OLLAMA_BASE_URL=http://10.100.1.14:11434
OLLAMA_CHAT_MODEL=qwen2.5-coder:14b
OLLAMA_EMBED_MODEL=nomic-embed-text
PGVECTOR_HOST=10.100.1.24
PGVECTOR_PORT=5433
PGVECTOR_DB=geonet_project_search
PGVECTOR_USER=geonet_ai
PGVECTOR_PASSWORD=<lihat .env server>
```

## API endpoint

| Method | Path | Auth | Keterangan |
|--------|------|------|-----------|
| `GET` | `/api/v1/ai/status` | Bearer | Cek ketersediaan + model aktif |
| `POST` | `/api/v1/ai/chat` | Bearer | Chat dengan RAG codebase |

Request body `/ai/chat`:
```json
{
  "message": "pertanyaan...",
  "history": [{"role":"user","content":"..."},{"role":"assistant","content":"..."}]
}
```

## Deploy notes

- Image di-rebuild di `10.100.1.24` via `docker compose build geonet-console`
- **Bug:** jangan gunakan `scp config/*.php routes/api.php root@host:/path/config/` dalam satu perintah — `routes/api.php` akan tertimpa ke `config/api.php`
- Selalu deploy config dan routes terpisah

## Lanjut dari sini

1. **UI chat** di Super Apps PWA — React window, kirim ke `/api/v1/ai/chat`
2. **User memory** — `POST /api/v1/ai/memory`, simpan ke tabel `user_memory` di pgvector
3. **Permission guard** — tambah `ai.chat` ke `group_permissions`, guard di middleware
4. **Streaming** — opsional, tambah `stream: true` ke `OllamaService::chat()`

## Jangan

- Jangan hardcode `PGVECTOR_PASSWORD` di source — baca dari env
- Jangan expose `/api/v1/ai/chat` tanpa auth (`api.auth` middleware sudah aktif)
- Jangan rebuild image tanpa verifikasi `config/api.php` isi yang benar