Nerdit

Nerdit helps independent builders and small teams deploy and update the tools they create with AI, on their own infrastructure. Build from a folder or Git repository, run the app as a supervised service, open its URL and connect it to persistent databases or local models when needed.

All current features are free during the public beta, with up to five linked machines per account. You provide the machines and pay for any external services you choose. Apps depend on their host staying available.

nerdit serve llama3.1:8b --gpus 1   # local model, OpenAI-compatible endpoint
nerdit deploy ./my-app --wait       # build, run and wait for health

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. Enable the HTTPS proxy to serve it at https://<host>/my-app; otherwise use its local loopback endpoint. 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.

Follow the getting-started guide for prerequisites, deployment and opening your app. 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 has no product telemetry. Optional cloud links, public relay traffic and external AI APIs send the data needed for those features. The [posthog] dashboard analytics hook is off by default and uses your own project if enabled.

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

app.nerdit.ai adds remote access, GitHub deployment and hosted URLs through an outbound connection from your machine. These features are included in the free public beta, for active accounts with up to five linked machines; no subscription is required.

Hosted shares are private to the signed-in owner by default. Public access requires explicit publication and is subject to the Terms and abuse enforcement. An unlinked daemon keeps deploying, proxying and serving models locally.

Future fleet/cluster management and a separate Managed infrastructure offer are not available as part of this delivery.

Questions or feedback: feedback@nerdit.ai.

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.