Bytebot AI Desktop Agent (Password-Protected)

Deploys Bytebot, an open-source AI desktop agent that carries out plain-language tasks on its own Linux desktop, behind a password-protected gateway. Includes the task UI, agent, desktop and Postgres. Bring your own Anthropic, OpenAI or Gemini key.

What you get

  • A Caddy gateway with basic-auth sign-in; it is the only service with a public domain
  • Bytebot task UI with a live desktop view and a Take over button
  • Bytebot agent that plans steps with your LLM, and an Ubuntu 22.04 XFCE desktop with Firefox, Thunderbird and VS Code on the private network
  • Postgres 18 on a volume for tasks and messages
  • Configurable model lists for Claude, OpenAI and Gemini, replacing the retired defaults in upstream's images

Good for

  • Handing off repetitive browser work such as filling web forms from a file or collecting data from several sites
  • Testing web flows end to end with screenshots of each step
  • Running a computer-use agent for research and prototyping away from your own machine
  • Taking over the desktop yourself and handing control back to the agent

Cost

Measured on Railway: the desktop uses 0.8 to 0.9 GB with Firefox open and the other services about 0.35 GB together, roughly 1.2 GB in total. The template notes put that at about $12 to $15 a month idle at Railway prices, more while tasks run.

Services in this template

  • bytebot-ui · image ghcr.io/bytebot-ai/bytebot-ui:edge@sha256:492d57da67513582dd15cde892dc8b29b9a6deec62c9e2b4e2c86e168c22302f
  • bytebot-agent · from baranberkay96/bytebot-railway
  • Bytebot · from baranberkay96/bytebot-railway
  • Postgres · image ghcr.io/railwayapp-templates/postgres-ssl:18
  • bytebot-desktop · image ghcr.io/bytebot-ai/bytebot-desktop:edge@sha256:e974e7ac93c8f755e98b8a437e5497a63175c7096b51a348f5f1d9320598a1c9

Questions people ask

How do I log in, and where does the model key go?

Open the Bytebot service domain and sign in with BYTEBOT_USERNAME (default bytebot) and BYTEBOT_PASSWORD from its Variables tab. Tasks need a key: set ANTHROPIC_API_KEY, OPENAI_API_KEY or GEMINI_API_KEY on the bytebot-agent service and redeploy it. LLM usage is billed by your provider.

What persists across redeploys?

Tasks and messages persist in Postgres. The desktop has no volume, as upstream, so files saved on it are lost when bytebot-desktop is redeployed.

What drives the cost?

Mostly the desktop service, which holds a running Firefox, plus the UI, agent, Postgres and gateway. It costs more while tasks run. The template deploys on Trial and Free, but Firefox does not start on the 0.5 GB Free limit, so Hobby or higher is advised.

Can I upgrade it?

Upstream is archived and publishes only an edge tag, so the images are pinned by digest. You can change the digests in template.json and services/agent/Dockerfile if a maintained build appears.

Any licence or usage notes?

Bytebot is Apache-2.0, and the desktop image also contains third-party software under its own licences. This is a community template, not affiliated with Bytebot or the model providers. Follow Railway's Fair Use Policy and the terms of the sites the agent visits.

Full template README

Deploy and Host Bytebot with Railway

Bytebot is an open-source AI desktop agent: you describe a task in plain language and it does the work on its own Linux desktop, using the browser, apps and files like a person would. This community template runs the full Bytebot stack behind a password, with current model lists for Claude, OpenAI and Gemini.

About Hosting Bytebot

Bytebot has four parts: a containerized Ubuntu desktop (XFCE, Firefox, Thunderbird, VS Code) with a control daemon, an agent that plans each step with an LLM and drives the desktop, a Next.js task UI with a live view of the desktop, and Postgres for tasks and messages. Upstream's UI has no sign-in, and through it anyone could watch and control the desktop. This template puts a password-protected gateway in front and keeps the desktop, the agent and the database on Railway's private network. The upstream repository is archived; the template pins its last published images and replaces their retired default models with configurable current ones.

Common Use Cases

  • Hand off repetitive browser work: filling web forms from a file, collecting data from several sites, downloading and sorting documents
  • Test web flows end to end and get screenshots of each step
  • Run a computer-use agent for research and prototyping without giving it access to your own machine
  • Take over the desktop yourself at any point, then hand control back to the agent

Dependencies for Bytebot Hosting

  • Bytebot desktop, agent and UI images from ghcr.io/bytebot-ai (last upstream build, pinned by digest)
  • Postgres 18 (Railway's image)
  • Caddy 2.11 as the sign-in gateway
  • An API key for Anthropic, OpenAI or Google Gemini (yours, entered after deploy)

Deployment Dependencies

Implementation Details

Service Source Purpose
Bytebot Caddy 2.11.9 (services/gateway) Public domain; basic-auth sign-in; refuses cross-site requests; proxies to the UI
bytebot-ui bytebot-ui:edge (digest-pinned) Task UI, live desktop view, proxy to agent and VNC (private)
bytebot-agent bytebot-agent:edge + model lists and start script (services/agent) Plans and runs tasks with your LLM; migrates Postgres at start (private)
bytebot-desktop bytebot-desktop:edge (digest-pinned) Ubuntu 22.04 XFCE desktop, bytebotd computer-use API, noVNC (private)
Postgres postgres-ssl:18 Tasks and messages, on a volume

First login. Open the Bytebot service's domain and sign in with BYTEBOT_USERNAME (default bytebot) and BYTEBOT_PASSWORD from the Bytebot service's Variables tab.

Add an LLM key. The deploy needs no key, but tasks do. On the bytebot-agent service, set ANTHROPIC_API_KEY (recommended), OPENAI_API_KEY or GEMINI_API_KEY, then redeploy bytebot-agent. Without a key the model picker is empty and the agent log says what to set. The models offered come from BYTEBOT_ANTHROPIC_MODELS, BYTEBOT_OPENAI_MODELS and BYTEBOT_GEMINI_MODELS (model-id=Title, comma-separated), because Bytebot's own lists name models that providers have retired.

Run a task. Pick a model, describe the task, and watch the desktop on the task page. Use "Take over" to control the desktop yourself.

Safe defaults. The password is generated at deploy. An empty, shorter than 12 or longer than 72-character password does not stop the deploy: the site is locked with a random password and the gateway log says how to fix it. Only the gateway has a public domain.

Persistence. Tasks and messages persist in Postgres. The desktop has no volume, as upstream: files saved on it are lost when bytebot-desktop is redeployed.

Resources and plans. Measured on Railway: the desktop uses 0.8–0.9 GB with Firefox open, the other services about 0.35 GB together. Use the Hobby plan or higher. The template deploys on Trial and Free too, but with 1 GB (Trial) the browser has little headroom and with 0.5 GB (Free) Firefox does not start. LLM usage is billed by your provider.

Upgrading. Upstream is archived and publishes only an edge tag, so the images are pinned by digest. Change the digests in template.json and services/agent/Dockerfile if a maintained build appears.

This is a community template, not affiliated with Bytebot, Anthropic, OpenAI, Google or Railway. Bytebot is Apache-2.0. Follow Railway's Fair Use Policy and the terms of the sites the agent visits.

Why Deploy Bytebot on Railway?

Railway runs the desktop, the agent, the UI and Postgres as one project with private networking, so only the sign-in gateway faces the internet. You get your own computer-use agent in a few minutes, and you can stop or redeploy it from one dashboard.

View Bytebot AI Desktop Agent (Password-Protected) on railway.com

Similar templates

TemplateCategoryDeploymentsDeploys
9Router
Self-hosted AI router for Claude, Gemini, Cursor & OpenAI tools.
AI/ML9,734
flowise-railway
Flowise - low-code LLM apps builder
AI/ML8,512
LibreChat
Open-source ChatGPT clone: multi-model, agents, search, multi-user auth
AI/ML6,762
Hermes Agent by Nous Research
Template for Hermes Agent by Nous Research
AI/ML6,413
OpenClaw
[Homebrew support] The AI that actually does things
AI/ML5,167
Hermes Agent
Self Host Hermes Agent. Autonomous AI agent w/ self-improving, tools & more
AI/ML4,092