oroboros. View on GitHub

An experiment in building with AI

An app that
builds itself.

With you in the loop.

Turn a natural-language request into a reviewable change. Oroboros brings Codex runs, isolated previews, and human approval into one self-hosted workspace.

Explore the project See how it works

Source on GitHub Self-hosted Experimental MVP

The project's green and gold dragon curling into an Ouroboros loop.
Build. Review. Evolve.The next iteration starts with you.

A full-stack idea.
A stack you can inspect.

  • Vue 3 + TypeScript
  • FastAPI + Python
  • PostgreSQL
  • Codex CLI
  • Git worktrees

01 / The idea

A good prompt is
just the beginning.

Getting an agent to write code is one step. Understanding the change, trying it, and deciding what lands is the rest of the loop.

Oroboros connects those steps. Requests become tracked runs. Changes get their own Git worktree and preview slot. Checks and artifacts stay with the run, so you can review the result before approving a merge.

Built for developers, self-hosters, and curious builders exploring how agents can help an app evolve—including the app that runs the workflow.

Read the project specification

02 / How it works

One change.
A complete loop.

From “what if” to “let me try it.”
Explore an example of the current workflow.

oroboros / codex
NEW RUNPrompt → change

What shall we build next?

Your prompt

Add a dark mode toggle to the navigation. Keep the existing light theme as the default.

Route /Note Keep it accessible
Context stays with the run.Submit prompt

Workflow illustration with sample content. Not a live run or a screenshot.

Approval reruns merge checks and merges locally. Remote push and deployment depend on your operator configuration.

03 / Inside the workspace

More visibility.
More intentional changes.

Explore the documentation

START WITH CONTEXT

A prompt with a place.

Use the Codex inbox or Cmd/Ctrl+K overlay. Attach a route and note, then find related runs from the page you’re working on.

SPACE TO EXPERIMENT

Isolated by design.

Three default preview slots pair Git worktrees with leases and database reset tooling. New runs wait when the slots are busy.

CHECK, THEN REVIEW

Results you can inspect.

Review diff snippets, configured validation results, migration warnings, and linked artifacts before choosing to approve or reject.

FOLLOW THE THREAD

Every run has a story.

Track states, events, and lifecycle notifications. Filter by status or route and use recovery actions for interrupted work.

Deployment has its own tools. Commit-based release and rollback scripts, health gates, and a release registry are included for host operators.

Deployment guide

04 / A work in progress, in the open

The ambition is big.
The scope is honest.

This is a hands-on experiment in agent-assisted building. The source is the best place to inspect where it stands.

Implemented in the current MVP

  • Prompt inbox, route context, and run tracking
  • Codex worker and isolated preview workflow
  • Checks, artifacts, and approve/reject review
  • Host deployment and rollback scripts

Beyond the current MVP

  • Authentication and role-based access
  • A managed, multi-tenant hosting service
  • Streaming run updates rather than polling
  • One-click rollback in the web interface

Design goals and limitations, not release promises.

Run it in an environment you control. The current app has no built-in login or access control, and validation quality depends on the checks you configure.

05 / Make it yours

Curious?
Start with the source.

Clone the repository, read the setup guide, and bring up your own workspace. This is a developer setup with host services, not a one-click install.

Read installation instructions Repository quick start
YOUR FIRST STEP 
git clone https://github.com/slobodanmargetic988/Oroboros.git
cd Oroboros

For the full runtime

A Linux host with systemd, Git, Node.js 20+, Python 3.11+, PostgreSQL, Redis, Caddy, and a configured Codex CLI.

  1. 01

    Configure your environment. Follow the guide to create the environment files and set your own paths and services.

  2. 02

    Prepare the database and runtime. Bootstrap the schema, install host services, and provision preview slots.

  3. 03

    Run the health checks. Verify the stack, open the Codex inbox, and start with a small change.

A few things to know.

Is this a hosted service I can sign up for?

No. Oroboros is an experimental, self-hosted project. You install and operate the runtime on your own host. This webpage is a public showcase.

Does it merge every AI change automatically?

The workflow includes human approval. Approval reruns merge checks before merging locally; remote push defaults to manual. Complete release deployment uses the separate operator scripts.

Is Oroboros the same as Ouroboros?

Yes. The GitHub repository is named Oroboros. The README, source code, and runtime also use the spelling Ouroboros, after the loop-forming serpent.

Where did the idea come from?

It’s a practical exploration of what agent-driven building can achieve, using workflows from the related Agents project. The goal is to make the process inspectable and repeatable.