Agent-first, self-hosted PaaS with native builds — no Docker, no Nixpacks, no
runtime VM. Deploy uv (Python), bun/npm (JS/TS), elm, and static sites
directly as OS processes on a VPS you own.
Designed as a lighter alternative to better-paas: same control-plane ideas (Go + SQLite + scoped agent tokens + audit), but the deployment engine runs native toolchains instead of Docker.
- One static Go binary (~12 MB) — no runtime, no cgo. SQLite (
modernc.org/sqlite) for state. - Native build pipeline —
uv sync --frozen,bun install --frozen-lockfile,npm ci,elm make … --optimize; frozen-first with lockfile gating. Seedocs/build-matrix.md. - Built-in process supervisor — spawns apps as child OS processes, streams logs, health-gates before cutover, restarts with exponential backoff. No systemd, no Docker.
- Built-in reverse proxy + Let's Encrypt —
autocert(ACME) in the same binary; route{app}.yourdomain.com→ app port. No Caddy, no nginx. - Scoped agent tokens — separate revocable tokens for AI tools/CI with
apps:read,deploy:trigger,logs:read, etc. (the better-paas idea, simplified). - Audit log — every mutating action recorded with actor + IP.
- CLI
lp— deploy, logs, stop/start/redeploy, agents from the shell (--jsonfor agents/scripts). - Agent skill —
skills/lp-usage/SKILL.md(Agent Skills standard, also installed at.prime/agent/skills/lp-usage/). - Elm dashboard — tiny SPA compiled by
elm makeand embedded into the binary.
make build # builds ./lp (requires go 1.25+, elm for the dashboard)
./lp # API on :8080, proxy on :80
cat data/admin_token.txt # first-boot admin tokenDeploy from the CLI:
./lp login http://localhost:8080 "$(cat data/admin_token.txt)"
./lp deploy https://github.com/you/fastapi-app --name api
./lp logs apiOr the dashboard: open http://localhost:8080, paste the token.
| Variable | Default | Purpose |
|---|---|---|
LISTEN_ADDR |
:8080 |
API + dashboard |
PROXY_ADDR |
:80 |
app reverse proxy (+ ACME HTTP-01) |
TLS_ADDR |
:443 |
TLS listener |
APP_DOMAIN |
— | base domain for {app}.{domain} routing |
PUBLIC_IP |
— | for {app}.{ip}.sslip.io URLs |
ACME_EMAIL |
— | enable autocert (Let's Encrypt) when set |
PORT_MIN/PORT_MAX |
9000/9999 |
app port pool |
LP_DATA_DIR/LP_BUILDS_DIR |
data/builds |
state + build artifacts |
ADMIN_TOKEN |
generated | override admin token |
TRUST_PROXY |
false |
honor X-Forwarded-For |
See docs/architecture.md for the full design and the decisions taken from
better-paas (and the ones deliberately changed).
detect → install → build → start (supervised process)
Runtimes auto-detected in this order (docs/build-matrix.md is authoritative):
- elm —
elm.json("type": "application") - uv —
uv.lock>pyproject.toml>requirements.txt - bun —
package.jsonpackageManager/bun.lock/bun.lockb - npm —
package-lock.json/ barepackage.json - static —
index.htmlat root/public/dist/out/build/_site
Ambiguous roots (e.g. pyproject.toml + package.json) are rejected with a clear
error — set buildMethod or rootDir explicitly. Every default can be overridden
per app (installCommand/buildCommand/startCommand/healthPath).
Bearer-token auth. GET /api/health and POST /api/auth/login are public.
GET /api/apps,POST /api/apps,GET/PUT/DELETE /api/apps/{id}POST /api/apps/{id}/redeploy|stop|startGET /api/apps/{id}/logs?lines=200GET /api/deployments?appId=GET/POST /api/agents,DELETE /api/agents/{id}(admin)GET /api/audit-logs(admin),GET /api/server/info
Each app is an OS process. Crash isolation is provided by the kernel; restart
policy + health gating is ~200 lines of Go (see internal/paas/supervisor.go).
Builds use the exact native toolchains (uv, bun, npm, elm) already on
the VPS — no image build, no registry, no 500 MB image pull per deploy.
MVP: static/bun/elm/uv deploy, stop/start/delete/redeploy, logs, env vars, domains, auto-HTTPS, scoped tokens, audit, CLI, Elm dashboard (in progress). Not yet: cron, backups, rollback UI, multi-server.
AGPL-3.0 (same as better-paas).