Describe what you need — a shop, a booking page, an internal tool — and watch it come together in a live preview. When it looks right, put it in front of your customers with one click. It runs on your own server, with the AI subscription you already have.
One command sets up Docker, Caddy and TLS on a fresh server, then prints your access token.
Each task gets its own branch, dev server and database. Chat with Claude Code or Codex; every turn is a checkpoint.
Blunc merges, builds, checks, backs up, migrates and switches traffic. If anything fails, it rolls back.
Sign in once on the server and every agent runs on your plan. Blunc never resells tokens or charges for usage.
Start from a fresh server. Blunc takes over Docker and ports 80 and 443, so it shouldn’t share the box with anything else.
Point the domain and a wildcard for previews at the server. TLS certificates are issued automatically.
Claude Code and Codex share one composer. Pick how much the agent may do with ⇧Tab, switch models in the same menu, and Blunc passes the right flags to each CLI.
Each task has its own environment variables, written to .env.local in its worktree. Edit a value and the dev server restarts with it.
Every file the agent touches, live, with its diff.
A checkpoint per agent turn. Step back to any of them.
Build errors are caught and handed to the agent in one click.
Browse tables and run SQL against the task’s own Postgres.
Start a task and Blunc creates the branch, a git worktree, a dev server with its own URL and a database to match. Chats in the task share all of it.
When the agent fans out, each sub-agent gets a row with its task, its current step and the result. Open one to read its full log; the main chat stays readable.
Install Agent Skills from a marketplace or a git repo, connect MCP servers with a sign-in, and turn them on per project. The agent picks them up on its own; you see every call in the chat.
Deploy to Production runs seven steps and stops at the first failure. Only a person can press it: the agent can ask for a deploy, never start one.
Prod and main are untouched. The log goes to the agent with Resolve with AI.
Traffic goes back to the old image automatically. Additive migrations keep your data.
Each project has a dashboard over live Production data. Describe what you want to see; the agent writes the SQL or MetricsQL and the chart, and you add it with one click.
Blunc is one Docker stack on one Linux VPS, and 4 vCPU with 8 GB of RAM is enough to start. Use a clean server: Blunc takes over Docker and ports 80 and 443, so it shouldn’t share the box with anything else. Agents and dev servers are treated as untrusted from day one.