Hi HN, I reflected on how to scale my one-person freelancer activity, so I built Openvisor to make myself available 24/7. Customers deposit a project, sandboxed coding agents (OpenHands) build it enriched by my knowledge (10y of eng. practices), a demo goes live on a subdomain, and billed per-token. It runs in production (berwick.ai); today I'm open-sourcing it under Apache 2.0. Accept customers even when you're sleeping ;)

Why it's cool: - Customize your branding/prompts/specialties/harness (yaml files) - Link your own knowledge bases (your best practices on how to code "well") - Creates or connects one or multiple repos (deploy keys) - Isolated autonomous long-running agentic dev a strong harness (compose or K8s) - Link custom MCPs like Context7 + any tool - Routines that run custom prompts periodically - Leak scanner and secrets/keys for knowledge bases and code - Yourself MCP: customers can integrate your Openvisor to their own AI tools (and are billed) - I've looked through the litterature to find the best harness implementation - Every token is counted and billed to the customer's credits

I use the auto-dev mode for my own clients: when my issues are tagged 'auto-dev', my agent team runs the dev and create the PR (auto-merge is an option).

Stack: FastAPI, Celery, Postgres/pgvector, Meilisearch, Traefik, React, docker compose on one VM (Helm chart optional). `make dev` brings it up; there's also a one-line prompt you paste into Claude Code that deploys it for you.

Honest limits: alpha. PRs need GitHub or GitLab; production demos need Sysbox; the agent harness is OpenHands with default settings, so build quality is bounded by that and we privilege reliability over speed.

If you run a consultancy or an agency, I'd like to know what would stop you from running this.