lla.ma base

The self-hosted, AI-native database console — Supabase Studio's polish, for any Postgres, on your own box.

Browse schema, edit data, run SQL, and read an at-a-glance Overview dashboard for any PostgreSQL database — RDS, Neon, a bare VPS, or your laptop. No vendor lock-in. Runs as a single container you own.

---

---

Why

Supabase Studio is lovely — but it only manages a Supabase database. The moment your Postgres lives anywhere else, you're back to psql or a heavyweight desktop client. lla.ma base gives you that same polish for any Postgres, self-hosted.

• Overview dashboard — size, table/row/index counts, largest tables, advisors (unused indexes, vacuum, missing PKs, cache hit) and a live schema map (foreign-key graph). All from instant catalog queries.
• Browse & edit — paginated, sortable, PK-aware data grid with inline edit / insert / delete. Tables without a primary key are edited safely via ctid.
• SQL editor — query history, multi-statement support, and a result grid.
• Safety first — read-only by default; mutations are blocked until you flip to write mode, and destructive statements (DROP / TRUNCATE / ALTER, unscoped DELETE/UPDATE) require an explicit confirm. A console that won't let you TRUNCATE production by accident.
• Your credentials never reach the browser — all database access is server-side.
• AI-native, bring your own key (on the roadmap) — natural-language → SQL that runs against your own model key, never a managed middleman.

---

Quick start (Docker)

docker build -t llama-base .

docker run --rm -p 3939:3939 \
  -e PG_HOST=host.docker.internal \
  -e PG_PORT=5432 \
  -e PG_USER=postgres \
  -e PG_PASSWORD=your_password \
  llama-base

Open http://localhost:3939. Pick a database on the left — you land on its Overview.

host.docker.internal reaches a Postgres running on your host machine. For a remote database, set PG_HOST to its address (and PG_SSL=require).

Quick start (local dev)

npm install
cp .env.example .env   # then edit the PG_* values
npm run dev            # http://127.0.0.1:3939

---

Configuration

All config is environment variables (no UI to clutter — point it at a database and go):

Variable	Default	Description
PG_HOST	127.0.0.1	Postgres host
PG_PORT	5432	Postgres port
PG_USER	postgres	Connection user
PG_PASSWORD	—	Connection password
PG_ADMIN_DB	postgres	Admin database for server-level ops (list/create database)
PG_SSL	off	off | require (self-signed) | verify (validate cert)
LLAMA_DB_ENGINE	postgres	Database engine (Postgres today; MySQL & Mongo planned)

Connecting to a remote box over SSH

If your Postgres isn't exposed publicly (it shouldn't be), lla.ma base can open the SSH tunnel for you — no second terminal, no manual ssh command. Add an SSH_* block to .env:

SSH_TUNNEL=on
SSH_HOST=your-box.example.com
SSH_USER=ubuntu
SSH_KEY=~/.ssh/id_rsa        # omit to use your ssh-agent
SSH_REMOTE_PORT=5432         # the DB port on that box
# point the app at the tunnel's local end:
PG_HOST=127.0.0.1
PG_PORT=5433

Then npm run dev opens the tunnel automatically before starting the app (and closes it on exit). Need just the tunnel? npm run tunnel. Connecting to a local/RDS/Neon database instead? Leave SSH_TUNNEL unset and npm run dev is a plain dev server.

---

How it's built

Next.js (App Router) + React + Tailwind, TypeScript, the pg driver. The shared @lla-ma/ui design system supplies the header/footer shell and theme tokens. Every database operation goes through a single engine-agnostic adapter (lib/adapters/), so the API and UI never touch a driver directly — which is how MySQL and Mongo will slot in later as one adapter file each.

app/api/*        route handlers — call getAdapter(), never pg directly
lib/adapters/    DatabaseAdapter contract + the Postgres adapter
lib/db.ts        pg connection pools (one per database, lazy)
lib/introspect   catalog queries (schemas, tables, columns, keys)
components/      Overview, SchemaMap, DataGrid, SqlEditor, Sidebar

---

Roadmap

• Connect to any Postgres · schema browser · SQL editor
• PK-aware data grid with inline edit / insert / delete
• Overview dashboard + foreign-key schema map
• MySQL adapter, then MongoDB
• Natural-language → SQL (bring your own API key)

lla.ma base is free and self-hostable forever. A hosted edition (LLA.MA PRO) will add the things that need a server recording history — trend charts over time, query-latency monitoring, alerts, team collaboration, managed AI — and fuse the database console with app hosting under one login. The console you self-host stays complete on its own; PRO is convenience and the timeline, not a paywall on the basics.

---

License

Apache-2.0 — see LICENSE.