---
title: "Deploy Botpress"
description: "Build chatbots with a visual flow editor and built-in NLU"
category: "Bots"
url: https://railway.com/deploy/botpress-ai
---

# Deploy Botpress

Build chatbots with a visual flow editor and built-in NLU

**[Deploy Botpress on Railway](https://railway.com/template/botpress-ai)**

- **Creator:** A3A
- **Category:** Bots
- **Total deploys:** 1

## Template content

### Postgres https://cdn.jsdelivr.net/gh/homarr-labs/dashboard-icons/svg/postgresql.svg

- **Image:** ghcr.io/railwayapp-templates/postgres-ssl:18

### botpress https://raw.githubusercontent.com/botpress/v12/master/icon.png

- **Source:** https://github.com/gridalpha/botpress-railway
- **Health check:** /status
- **Public domain:** Yes

### botpress-lang https://raw.githubusercontent.com/botpress/v12/master/icon.png

- **Source:** https://github.com/gridalpha/botpress-railway
- **Health check:** /info

## Documentation

![Botpress logo](https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcSawAc3Cr5XvOL9mhgZqXz11A6zYiS6dTtD2JFSLNBsOF-FQg72MHfvHCQ&s=10)

# Deploy and Host Botpress on Railway

Botpress is an open-source platform for building and running chatbots. A visual flow editor, a natural language understanding engine, a content manager and a webchat widget ship in one server, so a designer can draw a dialogue while a developer drops into JavaScript where it is needed. This is Botpress v12, the AGPL-licensed self-hostable server — your intents, conversations and transcripts stay in a database you own.

Deploy Botpress on Railway and the stack arrives wired together: the Botpress Server with the Duckling entity extractor beside it, a private Language Server holding the word embeddings the NLU engine needs, and PostgreSQL. Browser traffic reaches only the Botpress Server, which talks to the Language Server privately and writes every bot, conversation and message to Postgres. Nothing calls out to a hosted service, so you can self-host Botpress without chat data leaving the project.

![Botpress, its language server and Postgres on Railway](https://res.cloudinary.com/rroe4rtk/image/upload/v1787699990/botpress-architecture.png)

## Getting Started with Botpress on Railway

Set `BP_ADMIN_EMAIL` and `BP_ADMIN_PASSWORD` before deploying and the server creates that super admin on first boot, closing the sign-up form for good. Leave them blank and the first visitor to `/admin` registers the owner account instead — do that immediately, because until someone claims it, anyone with the URL can. Signing in lands you on the bot list.

Click **Create Bot → New Bot**, name it, and pick a starter such as *Small Talk*. Opening it loads the Studio, flow editor on the left and emulator on the right. Press **Train Chatbot** in the bottom-right corner and wait for the label to read *Ready* — that round trip proves the Language Server and the NLU engine both work. Type a greeting into the emulator and an answer should come straight back.

The bot is publicly reachable at `/s/` as a full-page webchat, and the same widget embeds anywhere via a script tag from **Config → Channels**. **Languages** should report the private lang server with English installed, and **Production Checklist** flags anything unhardened.

![Botpress Studio flow editor beside a working chat emulator](https://res.cloudinary.com/rroe4rtk/image/upload/v1787699993/botpress-studio-emulator.png)

![Botpress webchat answering four small-talk questions](https://res.cloudinary.com/rroe4rtk/image/upload/v1787699999/botpress-webchat-conversation.png)

![Language Management showing the private language server](https://res.cloudinary.com/rroe4rtk/image/upload/v1787700001/botpress-language-management.png)

## About Hosting Botpress

Most chatbot builders are hosted products: training data, transcripts and customer messages live on someone else's infrastructure, and pricing scales with volume. Botpress v12 inverts that — models, flows and message history run in containers you control. Teams reach for it when conversations carry personal data or when per-message pricing stops making sense.

Key capabilities:

- **Visual flow editor** with nodes, transitions, skills and sub-flows
- **On-premise NLU** — intents, entities, slot filling and spell-checking, with no external API call per message
- **Content management** for text, images, cards and carousels, separate from flow logic
- **Channels** for webchat, Messenger, Slack, Teams, Telegram and Twilio SMS, plus human handoff and analytics

This template runs the shape Botpress documents for production. The **Botpress Server** serves the admin panel, the Studio, the webchat and the messaging API, with **Duckling** — which extracts dates, amounts, emails and phone numbers — beside it. The **Language Server** is separate because it is the memory-hungry part: it holds the word embeddings and serves vectors privately. **PostgreSQL** stores bots, users, conversations, messages and trained models.

## Why Deploy Botpress on Railway

Railway removes the plumbing this stack normally needs.

- Private networking between Botpress, the Language Server and Postgres
- Managed PostgreSQL with backups, provisioned and connected
- Persistent volumes for the language models and bot data
- HTTPS on a generated domain, or your own
- Health checks and restarts on every service
- Vertical scaling from the dashboard when a bot gets busy

## Common Use Cases

- **Customer support deflection** — answer the repeat questions in webchat and hand anything else to a human with the transcript attached
- **Lead qualification** — run a scripted dialogue on a marketing site and post the answers into your CRM from a JavaScript action
- **Internal helpdesk** — an HR or IT bot in Slack or Teams answering policy questions and opening tickets
- **Data-sensitive assistants** — healthcare, finance and public-sector bots whose transcripts stay in-house



## Dependencies for Botpress

- **Botpress Server + Duckling** — built from [`gridalpha/botpress-railway`](https://github.com/gridalpha/botpress-railway) on the official [`botpress/server`](https://hub.docker.com/r/botpress/server) image. Public, port 3000, volume at `/botpress/data`.
- **Botpress Language Server** — the same image run as `bp lang`, private on port 3100, volume at `/botpress/lang`.
- **PostgreSQL** — Railway's managed `postgres-ssl:18`. Botpress creates its schema and runs its migrations on first boot.

### Environment Variables Reference

| Variable | Service | Purpose |
|---|---|---|
| `DATABASE_URL` | Botpress | Postgres connection string; replaces SQLite |
| `EXTERNAL_URL` | Botpress | Public base URL for webchat and channel callbacks |
| `BPFS_STORAGE` | Botpress | `database` keeps bots and revisions in Postgres |
| `AUTO_MIGRATE` | Botpress | Applies data migrations when the image updates |
| `BP_ADMIN_EMAIL` / `BP_ADMIN_PASSWORD` | Botpress | Creates the super admin, closes registration |
| `BP_LANG_ENDPOINT` | Botpress | Private address of the Language Server |
| `BP_LANG_DIM` | Language Server | Embedding size — `100` default, `300` for accuracy |
| `BP_LANG_PRELOAD` | Language Server | Language codes fetched at first boot |

### Deployment Dependencies

- Source repo: [gridalpha/botpress-railway](https://github.com/gridalpha/botpress-railway)
- Upstream: [botpress/v12](https://github.com/botpress/v12), AGPL-3.0
- Image: [botpress/server](https://hub.docker.com/r/botpress/server)
- Docs: [v12.botpress.com](https://v12.botpress.com/)

## Hardware Requirements for Self-Hosting Botpress

The Language Server dominates the memory bill: each model loads fully into RAM.

| Resource | Minimum | Recommended |
|---|---|---|
| CPU | 2 vCPU total | 4 vCPU total |
| RAM | 1 GB app + 2 GB Language Server | 2 GB + 4 GB |
| Storage | 5 GB embeddings + 1 GB bot data | 10 GB for several languages |
| Database | PostgreSQL 11+ | PostgreSQL 16+ with backups |
| Runtime | Docker | Docker |

One English model at 100 dimensions is about 1 GB on disk and 1.3 GB in RAM; `BP_LANG_DIM=300` raises that to roughly 3.1 GB and 3.5 GB per language.

## Self-Hosting Botpress

The quickest local trial runs Botpress on SQLite against its public NLU endpoints:

```
docker run -d --name botpress -p 3000:3000 \
  -v botpress_data:/botpress/data botpress/server:latest
```

For anything real, run your own Language Server and point Botpress at Postgres — the two containers this template mirrors:

```
docker run -d --name botpress-lang -v botpress_lang:/botpress/lang \
  botpress/server:latest bash -c "./bp lang --langDir /botpress/lang"

docker run -d --name botpress \
  -p 3000:3000 -v botpress_data:/botpress/data \
  -e DATABASE_URL=postgres://user:pw@postgres:5432/botpress \
  -e BP_PRODUCTION=true -e BPFS_STORAGE=database \
  -e BP_MODULE_NLU_LANGUAGESOURCES='[{"endpoint":"http://botpress-lang:3100"}]' \
  botpress/server:latest \
  bash -c "./duckling & ./bp"
```

Once the Language Server is up, install a language from the **Languages** page and restart Botpress so the NLU engine loads it.

## How Much Does Botpress Cost to Self-Host?

Botpress v12 is free and open source under AGPL-3.0, with no seat, bot or message limits. A paid Pro licence unlocks multi-node clustering with Redis, several workspaces and multilingual bots — none of which the community edition needs for a production bot. Botpress Cloud is a separate hosted product with usage-based pricing. Self-hosting on Railway costs only the compute, memory and storage the three services consume.

## FAQ

**What is Botpress?**

Botpress is an open-source chatbot platform combining a visual flow editor, an on-premise natural language understanding engine, a content manager and channel connectors for webchat, Slack, Messenger, Teams and Telegram. v12 is the self-hostable edition.

**What does this Railway template deploy?**

The Botpress Server with Duckling on a public domain, a private Language Server serving word embeddings, and managed PostgreSQL. Both Botpress services get a volume.

**Why does the template include a separate Language Server?**

The NLU engine turns every message into word vectors, and those models are large — about 1 GB of RAM per language. Splitting it out lets you scale that memory on its own and keeps a model reload from restarting the chatbot.

**Why does Botpress need PostgreSQL instead of SQLite?**

SQLite is fine on a laptop, but a container recreated on every deploy needs its state elsewhere. Postgres holds bots, users, conversations, messages and trained models, so a redeploy never loses a transcript.

**How do I add a language other than English to self-hosted Botpress?**

Set `BP_LANG_PRELOAD` to a list such as `en,fr,es` before deploying, or install extras later from the **Languages** page. Check the volume has room — each model is roughly 1 GB.


## Similar templates

- [Telegram JavaScript Bot](https://railway.com/deploy/5lRkWa) — A template for Telegram bot in JavaScript using grammY
- [Cobalt Tools [Updated Aug ’26]](https://railway.com/deploy/cobalt) — Cobalt Tools [Aug ’26] (Media Downloader, Converter & Automation) Self Host
- [Telegram Gateway](https://railway.com/deploy/railway-telegram-gateway) — Multi-bot Telegram webhook gateway with WebSocket event streaming

Open this page in a browser: https://railway.com/deploy/botpress-ai