---
title: "Deploy nanoMuse"
description: "Self-hosted personal AI agent with memory, goals and a phone app"
category: "AI/ML"
url: https://railway.com/deploy/deploy-nanomuse-personal-ai-agent
---

# Deploy nanoMuse

Self-hosted personal AI agent with memory, goals and a phone app

**[Deploy nanoMuse on Railway](https://railway.com/template/deploy-nanomuse-personal-ai-agent)**

Machine-readable deploy manifest (JSON, validated by TemplateCI): https://railway.com/deploy/deploy-nanomuse-personal-ai-agent/manifest.json

- **Creator:** onepush
- **Category:** AI/ML

## Template content

### nanomuse https://api.iconify.design/mdi/robot-happy-outline.svg

- **Image:** ghcr.io/onepush-dev/nanomuse:1.0.9
- **Health check:** /api/health
- **Public domain:** Yes

## Documentation

# Deploy and Host self hosted nanoMuse (AI/ML) on Railway

This template deploys nanoMuse, a self-hosted personal AI agent with memory, goals and a credential vault, as independent community packaging of nano-muse/nanoMuse v1.0.9 (GPL-3.0-or-later, Copyright 2026 nanoMuse contributors). It is not affiliated with or endorsed by nanoMuse or Meta Platforms, Inc.

## About Hosting nanoMuse open-source software on Railway (self hosted nanoMuse template)

This template runs the unmodified upstream nanoMuse 1.0.9 server image, which is GPL-3.0-or-later open-source software, as one Railway service named nanomuse.

## Why Deploy nanoMuse, the self-hosted personal AI agent on Railway

A hosted chatbot with no memory store forgets you between chats, and you cannot see or move what it has learned about you. nanoMuse keeps its memory, goals, vault and threads on one volume you control, so the state stays with you.

Railway is a singular platform to deploy your infrastructure stack. Railway will host your infrastructure so you don't have to deal with configuration, while allowing you to vertically and horizontally scale it.


### Railway vs Other Hosting Providers and VPS for nanoMuse self hosting

Where nanoMuse's always-on process and /data volume live is the main decision. This table compares Railway with three common alternatives.

| Provider | What You Get with Railway | What You Get with the Other Provider |
|---|---|---|
| DigitalOcean | A one-click template with a generated access token, a persistent /data volume, and a public HTTPS domain. | A virtual machine you control, where you install Docker, manage the firewall, and back up the volume yourself. |
| AWS | A managed service with the volume, domain, and restart policy wired up, and one replica kept for local state. | Broad services with more control, but you configure networking and storage by hand. |
| Hetzner | A managed deploy path with no server administration or host patching. | Low-cost machines for people comfortable administering Linux who want direct host control. |

Choose Railway for the fastest path to a persistent nanoMuse, or a VPS if you want to run the host yourself.

## Common Use Cases for hosted nanoMuse

- **Phone-friendly personal agent:** Pair your phone once with the access token and keep a personal agent open in the browser from anywhere, with no desktop required.
- **Long-term memory across sessions:** Let the agent remember what it has learned about you, so you do not repeat context every time you start a new thread.
- **Ongoing goals:** Give the agent goals it can keep working on over time, checking progress on the same persistent state rather than starting fresh each visit.
- **Credential vault:** Store the keys and logins the agent needs in the encrypted vault on the volume, instead of pasting secrets into every chat.
- **Single-volume backup:** Keep memory, goals, vault, threads and the workspace together on one volume, so one copy covers everything your agent knows.

## Dependencies for nanoMuse Docker hosted on Railway

This template runs one service from an unmodified upstream image, with one persistent volume and a generated access token.

### Deployment Dependencies for Managed nanoMuse Service

The nanomuse service is public on port 8080, with a volume at /data, a health check at /api/health, and restart policy Always. All state sits on that volume, so keep one replica.

Template variable (generated, zero inputs):

- NANOMUSE_SERVER_TOKEN: access token for the web app and API. Open https://your-domain/#token=NANOMUSE_SERVER_TOKEN once to pair a browser.

Optional extras, not template variables: NANOMUSE_LLM_API_KEY, NANOMUSE_LLM_BASE_URL, NANOMUSE_LLM_MODEL, NANOMUSE_VAULT_KEY.

### Implementation Details for nanoMuse

The image ghcr.io/onepush-dev/nanomuse:1.0.9 is the unmodified upstream ghcr.io/nano-muse/nanomuse:1.0.9 image plus a Railway start wrapper. The wrapper runs as root so it can change ownership of the /data volume, which Railway mounts as root, then drops to the image muse user (uid 1000) with setpriv and runs nanomuse serve --host 0.0.0.0 --no-qr. Baked environment values are NANOMUSE_DATA_DIR=/data, NANOMUSE_WORKSPACE=/data/workspace, NANOMUSE_SERVER_HOST=0.0.0.0, NANOMUSE_SERVER_PORT=8080, NANOMUSE_LLM_PROVIDER=openai, NANOMUSE_LLM_BASE_URL=https://api.deepseek.com and NANOMUSE_LLM_MODEL=deepseek-flash. There is no PORT template variable. The LLM key is not a deploy input; the server boots without a model. Shell and python tools run unboxed inside the container. This packaging uses the non-browser server image.

## How does nanoMuse compare against other tools

### nanoMuse vs a hosted chatbot with no memory store (Hosted Chatbot Alternative)

- **Memory and goals:** nanoMuse keeps long-term memory, goals and a credential vault on one volume you control, while a plain hosted chatbot usually forgets a conversation once it ends.
- **Data location:** your memory and workspace live on the /data volume in your own deployment, whereas a hosted chatbot keeps that history on the vendor's servers.
- **Convenience:** a hosted chatbot wins on zero setup, while nanoMuse needs a deploy, a token pairing step and a model key you supply.

### nanoMuse vs running a one-shot chat UI yourself (One-Shot Chat UI Alternative)

- **Persistence:** nanoMuse keeps threads, goals and the audit log across restarts on the /data volume, while a one-shot chat UI typically starts fresh each session.
- **Access control:** every /api route requires the bearer token, and you pair a browser once at https://your-domain/#token=NANOMUSE_SERVER_TOKEN.
- **Simplicity:** a one-shot chat UI is lighter and easier to reason about, which suits occasional prompts with no agent behavior.

### nanoMuse with Docker vs Railway (Railway Alternative)

- **Control:** running the image with Docker gives you full control over networking, storage paths and host hardening.
- **Setup:** Railway gives you a public domain, a health check on /api/health, restart policy Always and a volume at /data from one template, while Docker needs those wired up by hand.
- **Replicas:** keep one replica for local persistent state on either platform, so neither gives horizontal scaling for this app.

## How to use nanoMuse on Railway?

```
1. Open the service Variables tab and copy the generated NANOMUSE_SERVER_TOKEN value.
2. Open the public domain for your service in a browser.
3. Visit https://your-domain/#token=NANOMUSE_SERVER_TOKEN once to pair the browser with the server.
4. Paste your model key under Connections in the app, or add NANOMUSE_LLM_API_KEY as a service variable.
5. The defaults target DeepSeek at https://api.deepseek.com with model deepseek-flash, so change them only if you use another provider.
6. Keep one replica, because local persistent state lives on the /data volume.
7. Confirm the service is healthy by checking /api/health, which Railway uses for its health check.
```

## How to self host nanoMuse on other VPS Services (nanoMuse self hosting guide)

### Clone the Repository

Clone the upstream project at the pinned release tag so you run the code the packaging was built against.

```bash
git clone --branch v1.0.9 https://github.com/nano-muse/nanoMuse.git
cd nanoMuse
```

### Install Dependencies

Upstream: run the documented nanomuse container or install the project; see the upstream repository.

### Configure Environment Variables

Set a long random access token and a persistent data directory.

```bash
export NANOMUSE_SERVER_TOKEN=your-domain-token-here
export NANOMUSE_DATA_DIR=/data
export NANOMUSE_LLM_MODEL=deepseek-flash
```

Keep one instance writing to that data directory, and serve it over HTTPS at your-domain.

### Start the nanoMuse Application

Start the server on all interfaces, port 8080, then pair a browser at https://your-domain/#token=NANOMUSE_SERVER_TOKEN.

```bash
nanomuse serve --host 0.0.0.0 --no-qr
```

Railway handles managed hosting, the HTTPS domain, the persistent volume and restarts for you, so this manual route is only needed on your own VPS.

## Pricing and License of nanoMuse (nanoMuse pricing)

nanoMuse is GPL-3.0-or-later open-source software, and it is free to self-host.

## nanoMuse hosted vs self hosted comparison (Pricing, features, costs, and more)

This template packages the GPL-3.0-or-later software, which is not affiliated with or endorsed by nanoMuse or Meta Platforms, Inc.

### Monthly cost of self hosting nanoMuse on Railway

Usage-based. The bill depends on RAM, CPU, volume size and the services you run, so no exact numbers are given here.

### System Requirements for Hosting nanoMuse on a VPS

Requirements depend on how much memory and storage the agent uses over time.

This packaging is the non-browser server image with a phone-friendly web app. It does not include desktop apps or chat-app relays. Paste a real model key under Connections, or set NANOMUSE_LLM_API_KEY, before you expect the agent to answer. Keep one replica unless you have a clear reason to change it, because local state is not shared across instances.

## Frequently Asked Questions (FAQs)

### How do I sign in to my nanoMuse server?

Open https://your-domain/#token=NANOMUSE_SERVER_TOKEN once in your browser, using the token generated for your deploy, to pair that browser.

### Where do I put my model key?

Paste your key under Connections in the app, where it is stored in the vault on the volume, or add it as the NANOMUSE_LLM_API_KEY service variable.

### Which model is the default?

The default is DeepSeek's deepseek-flash model, reached through an OpenAI-compatible base URL at https://api.deepseek.com.

### Is this affiliated with Meta?

No. nanoMuse is an independent community project and is not affiliated with or endorsed by Meta Platforms, Inc.

### What should I back up?

Back up the /data volume, which holds memory, goals, the vault, threads, the audit log, and the agent workspace at /data/workspace.


## Similar templates

- [Chat Chat](https://railway.com/deploy/-WWW5r) — Chat Chat, your own unified chat and search to AI platform.
- [Weaviate](https://railway.com/deploy/weaviate-1) — Vector database with hybrid search and an API key-protected API
- [Möbius Personal AI OS](https://railway.com/deploy/mobius-v01-personal-ai-os) — Self-hosted AI agent that builds and runs apps for you, with memory.

Open this page in a browser: https://railway.com/deploy/deploy-nanomuse-personal-ai-agent
