Deploy PicoClaw

Deploy and Host PicoClaw on Railway

PicoClaw is an ultra-light personal AI assistant from Sipeed, written in Go, with 30,000+ GitHub stars. Its core runs in about 10–20 MB of RAM and boots in under a second, yet it connects to 30+ AI providers and chat apps like Telegram, Discord, WhatsApp and Slack, with tools for files, scheduling, web search and MCP. This template runs PicoClaw on Railway with a password-protected web dashboard and a persistent volume. ChatGPT Plus and Claude Pro each cost $20/month. PicoClaw on Railway costs about $1–2/month plus the API usage of the model you pick.

What This Template Deploys

It's one tiny service, with no database, no Redis and no sidecars. The image is Sipeed's official launcher plus a small wrapper that binds to Railway's port and sets your dashboard password before the first request is served.

About Hosting PicoClaw

Personal AI agents are usually heavy. OpenClaw-style gateways need Node.js and hundreds of megabytes of RAM. PicoClaw takes the opposite approach: one Go binary built to run on $10 single-board computers, with the same idea of an assistant you message from your chat apps, which can use tools and run scheduled tasks for you.

On Railway, you get a public HTTPS dashboard, a health check that reads the password store on the volume (so it fails if storage breaks, not just if the process dies), and state that survives every redeploy. Because the footprint is so small, it costs very little to keep online around the clock.

The dashboard is password-protected from the first second. Upstream PicoClaw lets the first visitor set the dashboard password, which is risky on a public URL. This template generates PICOCLAW_DASHBOARD_PASSWORD at deploy time and seeds it before anything is served. Copy it from the Variables tab. If you change it later in the UI, your new password survives redeploys.

Early-stage software. Sipeed describes PicoClaw as early development, not yet recommended for production before v1.0. It's great for personal use; keep API keys scoped and don't connect it to anything critical.

Typical cost: ~$1–2/month on Railway, plus model API usage, which is usually a few dollars a month for light personal use on a small model. A ChatGPT Plus subscription is $20/month.

Deploy PicoClaw in Under 3 Minutes

1. Click Deploy on Railway. The service starts in about a minute.
2. Copy PICOCLAW_DASHBOARD_PASSWORD from the service's Variables tab.
3. Open your Railway URL and sign in to the dashboard.
4. Add a model provider (OpenAI, Anthropic, Gemini, OpenRouter, DeepSeek, Groq, Ollama and more) and paste its API key.
5. Optionally connect a channel such as a Telegram or Discord bot, then message your assistant.

No SSH. No config files. No Node.js runtime.

Common Use Cases

• Cheap self-hosted ChatGPT alternative. Pay a dollar or two for hosting plus API usage instead of a $20 subscription.
• Assistant in Telegram, Discord or WhatsApp. Message your agent from your phone for answers, drafts and research.
• Reminders and recurring tasks. Use built-in scheduling for daily briefings, check-ins and recurring jobs.
• Web research. Search with DuckDuckGo, Brave, Tavily, Perplexity or SearXNG and summarise the results.
• MCP tool hub. Connect MCP servers over stdio, SSE or HTTP to give the assistant new tools.
• Team chat bot. One assistant in Slack, Matrix or Discord for a small team or community.
• Lightweight alternative to OpenClaw. The same chat-app agent idea at a fraction of the memory.
• Multi-language chat platforms. Connect LINE, QQ, DingTalk, Feishu, WeCom or IRC as well as the usual apps.

Configuration

Everything else is set in the dashboard after you sign in: model providers and API keys (stored encrypted on the volume), chat channels, MCP servers, skills and the assistant's persona. None of them need Railway variables, so changing a provider never requires a redeploy.

Connecting a Telegram bot. Message @BotFather on Telegram, send /newbot, copy the token, and add it under Channels in the PicoClaw dashboard. Discord works the same way with a bot token from the Discord Developer Portal; invite the bot to your server with message permissions. Restrict who can talk to the bot in the channel settings, so strangers can't spend your API credits.

Using a local model. Point a provider at an Ollama or vLLM endpoint you host, for example another Railway service reached over the private network. Then no third-party model API is involved at all.

Health check. Railway checks /api/auth/status, which reads the password store on the volume. If the volume becomes unwritable, the service is marked unhealthy instead of silently losing state.

PicoClaw on Railway vs. Alternatives

Dependencies for PicoClaw Hosting

• Railway account. The Hobby plan ($5/month including $5 of usage) easily covers it.
• An API key from a model provider, or an Ollama or vLLM endpoint you host.
• Optional: a bot token for Telegram, Discord, Slack or another channel.

Deployment Dependencies

• PicoClaw on GitHub: source code and releases
• PicoClaw website: documentation and channel guides
• Railway Volumes Documentation: persistence for /data

Implementation Details

The image derives from sipeed/picoclaw:launcher and carries no upstream code changes. The wrapper does two things the stock image can't on Railway: it binds the launcher to $PORT (the launcher only takes its port from a flag or config file), and it seeds the dashboard password through the launcher's own token migration before the sockets answer. That only happens while no password exists, so later UI changes stick.

All state is under /data/picoclaw. Back up the volume before upgrading across PicoClaw releases, since pre-1.0 versions can change config formats.

Frequently Asked Questions

What is PicoClaw? A tiny open-source personal AI assistant from Sipeed, written in Go. You chat with it from the web dashboard or apps like Telegram and Discord, and it can use tools, search the web and run scheduled tasks.

How much does PicoClaw cost compared to ChatGPT Plus? About $1–2/month on Railway plus model API usage, which depends on the model and how much you use it. ChatGPT Plus and Claude Pro are $20/month each.

Where do I find the dashboard password? It's PICOCLAW_DASHBOARD_PASSWORD in the service's Variables tab, generated at deploy time. Sign in with it, then change it in the dashboard if you like.

Do I lose my settings or memory on redeploy? No. Config, encrypted API keys, the password store, memory and workspace are on the /data volume, so redeploys keep them.

How is PicoClaw different from OpenClaw? Both are self-hosted assistants you reach from chat apps. PicoClaw is a separate Go project that uses a tiny fraction of the memory. OpenClaw is more mature, with a larger feature set.

Is PicoClaw ready for production? Not yet, by Sipeed's own description: it's pre-1.0. It's well suited to personal use. Use scoped API keys and keep it away from sensitive systems.

Why Deploy and Host PicoClaw on Railway?

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.

By deploying PicoClaw on Railway, you get a featherweight personal AI assistant that's always online, password-protected and reachable from your chat apps, for about the price of a coffee a month.