Nerdit

Nerdit runs your apps on your own machine the way a PaaS would: build from a folder, run as a supervised service, get an HTTPS URL, restart on crash, roll back on a bad deploy. When an app needs an AI model, Nerdit serves one locally on your GPU or proxies an external API, and the app sees the same OpenAI-style endpoint either way.

Your code and your data stay on your hardware.

nerdit serve llama3.1:8b --gpus 1   # local model, OpenAI-compatible endpoint
nerdit deploy ./my-app              # build, run, HTTPS URL, model wired in

How an app declares what it needs

nerdit.toml in the app folder:

[deploy]
name = "my-app"
port = 8000

[ai.default]
provider = "ollama"      # or "api" for an external endpoint
model = "llama3.1:8b"

The app starts with OPENAI_BASE_URL and OPENAI_API_KEY set, is reachable at https://<host>/my-app, and comes back after a reboot. Switching the model from local to an external API is a config change; the app code does not move.

Install

curl -fsSL https://get.nerdit.ai | sh

One signed archive with the CLI, the daemon and a pinned Caddy build. No Python needed. It installs a service unit and starts the daemon.

User guides and API documentation are maintained separately for the Nerdit website. This repository contains the source, tests and contributor instructions.

From source:

git clone https://github.com/nerdit-ai/nerdit && cd nerdit
python -m venv venv && venv/bin/pip install -e ".[dev,mcp]"
(cd src/nerdit/daemon/web && npm ci && npm run build)   # dashboard, required
venv/bin/nerdit init

What you get

• Deploy from a folder, a Git URL or a template. Dockerfile, Python and Node buildpacks. Versioned images, --rollback, --dry-run, --wait.
• Services with restart policy, backoff, HTTP and TCP health checks, stable ports, and zero-downtime cutover on redeploy.
• URLs and TLS from an embedded Caddy: path or subdomain routing, your own domains, mDNS on the LAN, an internal CA you can pin.
• Models via Ollama or vLLM on your GPU, or an external API. Same injected env vars in both cases.
• Databases: managed Postgres and Redis, credentials minted for you and bound into apps through [db.*].
• Secrets: write-only, encrypted at rest, rotatable, with a shared scope.
• Agents: an MCP server with 49 tools, scoped tokens, idempotent writes, structured errors and an audit log. A coding agent can deploy, wire and diagnose apps without shell access.
• Backups of the control plane, with an offline restore — plus, for a managed database, a file-level volume tar and a logical, application-consistent dump (nerdit db dump) you can restore live.

No telemetry

The daemon sends nothing anywhere. The one analytics hook, [posthog], is off by default and points at your own PostHog project if you turn it on.

Platforms

Linux (Ubuntu 22.04 or 24.04, systemd), macOS on Apple Silicon, Windows through WSL2. NVIDIA GPUs need Linux or WSL2; macOS runs models on CPU. AMD support is experimental and behind a flag.

Nerdit Cloud

Everything above is free and local. app.nerdit.ai is a separate paid service that adds remote access to your node and hosted URLs for apps you choose to share. It gates nothing local: a daemon that never links keeps deploying, proxying and serving models.

License

Apache-2.0, see LICENSE. "Nerdit" and the logo are trademarks and are not covered by the code license, see NOTICE.

Contributing

Issues and pull requests are welcome. CONTRIBUTING.md explains how contributions land in releases.