wild-pc/README.md

Wild PC

Standing to author, run, govern, and maintain your own software.

A personal software platform. Wild PC manages independent services, tools, and frontends — and launches your coding agents — from a single CLI, with a unified gateway, systemd integration, and a web dashboard.

Historically, applications have usually been developed by third parties, packaged for distribution, installed through app stores or package managers, and updated through channels controlled by someone other than the user. That model still works for large applications, but it is often too heavy for the small, personal, situational software people increasingly want to create.

AI-assisted development changes the shape of this problem. Users can now create useful software directly: small services, scripts, dashboards, agents, automations, and configurations of existing tools. Many of these programs do not need a public release process, an app store listing, or a conventional distribution channel. They need a reliable place to run.

Wild PC provides that place.

Wild PC gives simple applications a consistent local environment for development, registration, deployment, discovery, and operation. Programs remain independent, but Wild PC provides the surrounding structure: process management, routing, metadata, service lifecycle, logs, and a common interface for running and inspecting the software in your domain.

In this sense, Wild PC is a personal software estate: a practical way to organize the software you create and run yourself, without requiring every tool to become a fully packaged product.

How it works

Wild PC separates what software exists from how it runs:

• Programs — the catalog. A program is a source repo Wild PC knows how to work with (dev verbs, build) and where it lives. One file per program under ~/.wildpc/programs/<name>.yaml.

• Deployments — how a program is realized on this node. One file per deployment under ~/.wildpc/deployments/<name>.yaml, discriminated by its manager:

  manager	what it is	derived kind
  systemd	a supervised process — or, with a schedule, a timer	service / job
  caddy	a static site served by the gateway from a build dir	static
  path	a CLI installed on your PATH (uv tool install)	tool
  none	an external service on another node (reference only)	reference

  For systemd, a nested run.launcher (python / command / container / compose / node) says how the process starts. The kind is always derived from the manager (+ schedule) — never stored.

A program can have several deployments — a CLI that is both a tool on PATH and a scheduled job — so a program has no single kind of its own; it has deployments, each with its own.

Standing everything up is one honest step: wildpc apply renders systemd units and gateway config from your config, then reconciles the runtime to match — activating what's enabled, restarting what changed, deactivating what's disabled. Edit config, wildpc apply; wildpc apply --plan shows the diff first.

Stacks

Wild PC stacks are pre-configured development environments that provide starting points for building Wild PC programs. A stack can define the language, framework, dependencies, tools, conventions, and Wild PC integration needed for a particular kind of application — python-fastapi, python-cli, react-vite, supabase.

Stacks are designed to work well with coding assistants. They give assistants a consistent target when generating Wild PC programs, making it easier to produce applications that are correctly structured, configured, and ready to run under Wild PC. If your coding assistant understands Wild PC, it can help you create, register, manage, and evolve custom applications more efficiently.

A stack seeds a new program's scaffold and default dev-verb commands, but it's optional at runtime: a program stands on its own via its declared commands: and source:, so you can adopt any existing repo with wildpc program add — no stack required.

Agents

Wild PC can launch your coding agents. Declare them in wildpc.yaml under agents: — each entry is just a command Wild PC runs in a terminal:

agents:
  claude:
    command: claude
    description: Anthropic Claude Code
  aider:
    command: aider
    args: ["--no-auto-commits"]
    cwd: /data/repos/my-project

The dashboard has a terminal dock that launches any declared agent in a pseudo-terminal (over a WebSocket) and manages live sessions (list, resume, kill). Wild PC is assistant-agnostic — it only ever runs command args in a pty and never parses the agent's output, so any interactive CLI works. With no agents: block set, a sensible default set (claude, opencode, amplifier, …) is offered.

Quick start

Prerequisites: a Debian/Ubuntu-family Linux with apt and sudo, systemd (user services), and git. install.sh sets up everything else — Docker, Caddy, uv, and the wildpc CLI itself.

git clone <this-repo> ~/wildpc && cd ~/wildpc

# One command: installs uv + the wildpc CLI, sets up infra (Docker, Caddy, MQTT,
# Postgres), creates ~/.wildpc, registers Wild PC's own control plane, builds the UI.
./install.sh

wildpc apply                    # converge the runtime to config (units, routes, run)
wildpc doctor                   # verify — every check should be green
open http://localhost:9000      # the dashboard

wildpc doctor is your friend at every step: it inspects setup and runtime and, for anything not green, prints the exact next command. Run it any time something looks off — after an install, a deploy, or a config change.

Exposure: from localhost to your own HTTPS domain

Localhost is the first rung; you climb only as far as you need. Each rung is a small config change plus wildpc apply, and wildpc doctor tells you what a rung still needs.

Rung	You get	What it takes
localhost (default)	dashboard + services on :9000 / host:port	nothing — this is the quick start above
LAN HTTPS	real https://<name>.<your-domain> on every device on your network, publicly-trusted cert, services stay internal	own a domain; set gateway.tls: acme + domain:; one wildcard DNS record on your router; a Cloudflare token. → docs/dns-and-tls.md
Public	a chosen service reachable from the internet	public: true on the service + a Cloudflare tunnel. → docs/tunnel-setup.md

The jump to LAN HTTPS is the involved one (DNS + a token + binding :443). Set gateway.tls: acme and run wildpc doctor — it enumerates exactly the pieces that are still missing, each with its fix.

Creating programs

wildpc program create scaffolds the source and its deployment from a stack:

# A service — FastAPI app + a systemd service deployment (health, unit, route)
wildpc program create my-api --stack python-fastapi --description "Does something"
wildpc program test my-api
wildpc apply my-api                       # render unit + route, then start it

# A tool — a CLI installed on your PATH
wildpc program create my-tool --stack python-cli --description "Does something"
wildpc apply my-tool                      # installs its path deployment on PATH

# A static frontend — built once, served by the gateway
wildpc program create my-app --stack react-vite --description "Web interface"
wildpc program build my-app && wildpc apply

# Adopt an existing repo (no stack needed — dev verbs detected or declared)
wildpc program add ~/projects/some-rust-tool

CLI

Operations live under the resource they act on. program is the catalog; service, job, and tool are views over the one deployment set. Lifecycle is convergence — wildpc apply — not a pile of per-kind verbs.

# Programs — the software catalog
wildpc program list|info|create|add|clone|delete|run
wildpc program build|test|lint|type-check|check [name]        # dev verbs

# Deployment lenses (service = systemd, job = systemd + schedule, tool = path)
wildpc service  list|info|create|delete|restart|logs
wildpc job      …same verbs; create takes --schedule
wildpc tool     list|info                                      # CLIs on your PATH

# Platform-wide
wildpc apply [name] [--plan]                  # converge runtime to config — the workhorse
wildpc list [--kind K] [--stack S] [--json]   # catalog + every deployment view
wildpc status                                 # unified runtime status
wildpc doctor                                 # diagnose setup + health, with fix hints
wildpc restart [name]                         # imperative bounce (one or all)
wildpc gateway               # status + route table (inspection)

To turn a deployment off, set enabled: false in its config and wildpc apply — there's no start/stop/enable/install verb; the manager decides the mechanism (systemd unit, PATH install, gateway route), the verb is always apply.

wildpc tool list --json is the machine-readable tool catalog assistants use to build context — it surfaces each tool's actual executable (which can differ from the program name, e.g. litellm-intent-router installs intent-router), its description, and whether it's installed.

Configuration

The registry lives under ~/.wildpc/: a global wildpc.yaml plus one file per resource under programs/ and deployments/.

# ~/.wildpc/wildpc.yaml — globals
gateway:
  port: 9000
repo: /data/repos/wildpc     # resolves `source: repo:<name>` for wildpc's own programs

# ~/.wildpc/programs/my-api.yaml — the catalog entry
description: Does something useful
source: /data/repos/my-api
stack: python-fastapi

# ~/.wildpc/deployments/my-api.yaml — a systemd service (derived kind: service)
program: my-api
manager: systemd
run: { launcher: python, program: my-api }
expose: { http: { internal: { port: 9001 }, health_path: /health } }
proxy: true                        # served at my-api.<gateway.domain>
manage: { systemd: {} }
defaults:
  env:
    MY_API_PORT: ${port}           # = expose.http.internal.port
    MY_API_DATA_DIR: ${data_dir}   # = $WILDPC_DATA_DIR/my-api
    API_KEY: ${secret:MY_API_KEY}

defaults.env is the single, explicit source of a deployment's environment — Wild PC injects nothing implicitly. The placeholders ${port}, ${data_dir}, ${name}, ${public_url}, and ${secret:NAME} map your program's own env var names to Wild PC's computed values. Secrets live in ~/.wildpc/secrets/ (never in a repo).

Gateway, DNS & TLS

Every gateway-exposed deployment gets its own subdomain — <name>.<gateway.domain> — routed to it by the Caddy gateway (there are no path-prefix routes). Exposure is a single checkbox: proxy: true on a service, while a static deployment is inherently served. The dashboard is wildpc.<domain> and the API wildpc-api.<domain>; on a node with no domain, :9000 serves the dashboard plus a /api proxy.

gateway.tls is a per-node choice: off (plain HTTP on :9000) or acme (a real Let's Encrypt wildcard *.<domain> via a DNS-01 challenge — publicly trusted, no CA to install, while the services stay internal-only). Reaching a service from the public internet is separately opt-in via a Cloudflare tunnel (public: true). See docs/dns-and-tls.md.

Layout

~/.wildpc/                 # instance: config, artifacts, secrets
  wildpc.yaml              #   globals (gateway, repo, agents)
  programs/  deployments/  #   one file per program / deployment
  secrets/                 #   secret files (mode 700)
  artifacts/specs/         #   generated Caddyfile, registry.yaml
/data/repos/<name>/        # your program source (absolute source:)
/data/wildpc/<name>/       # per-deployment data volume
<repo>/                    # wildpc itself: cli/ core/ wildpc-api/ app/ docs/

Independence principle: your programs never depend on Wild PC. They accept configuration (data dir, port, URLs) via environment variables; only Wild PC's own programs (CLI, API, gateway) know Wild PC internals.

Dashboard & API

The dashboard (app/, served at wildpc.<domain> or http://localhost:9000) lists programs and deployments, edits their config, drives lifecycle, shows the gateway route table and logs, and hosts the agent terminal dock.

wildpc-api (port 9020, proxied at wildpc-api.<domain>) is the control plane: /deployments, /programs, /services, /jobs, /gateway, /status, an SSE /stream, config editing under /config/…, and the agent session endpoints under /agents. The full endpoint reference is in CLAUDE.md.

Mesh (opt-in)

Wild PC nodes can discover each other via MQTT + mDNS to form a personal infrastructure mesh — the gateway can route to services on other nodes and the dashboard shows discovered nodes and cross-node routes. It's all off by default; single-node needs none of it. Enable on wildpc-api via WILDPC_API_MQTT_ENABLED and WILDPC_API_MDNS_ENABLED.

Docs

• docs/registry.md — the registry model, wildpc.yaml, deployment fields, lifecycle
• docs/dns-and-tls.md — gateway routing, DNS, the off / acme TLS modes
• docs/stacks/ — per-stack guides (python-fastapi, python-cli, react-vite, supabase)
• AGENTS.md — the canonical, assistant-agnostic operator guide (recipes, gateway, tunnel, mesh)