Self-host CrewAI on a no-KYC VPS
Run multi-agent crews on your own server, 24/7, without Replit or a notebook tab open. Anonymous signup, no email required, no KYC. Pay with crypto or card, full root, no logs.
Quick start: CrewAI on Servury (Python)
Tested on Ubuntu 24. Pick a 2 GB+ plan, deploy, SSH in.
# 1. Install Python 3.11 + uv (fast Python package manager)
apt update && apt install -y python3.11 python3.11-venv pipx
pipx install uv
# 2. Scaffold a new crew
mkdir /opt/my-crew && cd /opt/my-crew
uv venv && source .venv/bin/activate
uv pip install crewai 'crewai[tools]'
crewai create crew research-crew
cd research-crew
# 3. Set your LLM key (or point at local Ollama)
echo "OPENAI_API_KEY=sk-..." > .env
# 4. Edit src/research_crew/config/agents.yaml + tasks.yaml to define your crew
# 5. Run it (one-shot)
crewai run
# 6. Run forever as a systemd service:
cat > /etc/systemd/system/crew.service <<EOF
[Unit]
Description=Research crew loop
After=network.target
[Service]
WorkingDirectory=/opt/my-crew/research-crew
ExecStart=/opt/my-crew/.venv/bin/crewai run
Restart=always
[Install]
WantedBy=multi-user.target
EOF
systemctl enable --now crew
What people run on a CrewAI VPS
Researcher agent crawls the web, analyst summarizes, writer drafts. Output a daily intel report into Notion or Slack.
Scraper finds prospects, qualifier filters, copywriter drafts outreach. Approve or auto-send.
News scraper + sentiment agent + risk analyst running on a schedule. Read-only signals to a private Telegram channel.
Topic-finder + outliner + writer + editor. Generate weekly drafts you review and ship.
Watcher agents that poll RSS / changelogs / forums for keywords, summarize, then notify.
Triage incoming alerts with a triager agent + remediation suggester. Pair with controlled tool access.
Frequently asked questions
Why a VPS instead of running CrewAI on my laptop?
Two reasons: 24/7 uptime (your laptop sleeps, the VPS does not), and a stable IP (rate limits, API keys, scrapers all behave better from a fixed cloud IP). Also, CrewAI runs are slow with API calls; you do not want to babysit them.
How much RAM do I need?
2 GB is fine if your LLM is hosted (OpenAI/Anthropic/OpenRouter). If you also self-host the model on the same VPS via Ollama with a 7B model, jump to 16 GB. CrewAI itself is light Python.
Can I run CrewAI without OpenAI?
Yes. CrewAI works with any LiteLLM-compatible backend: Anthropic, Ollama, vLLM, Groq, Together, OpenRouter. Set OPENAI_API_BASE to point at your local LLM endpoint.
How do I orchestrate scheduled runs?
systemd timers or plain cron. Or use the systemd service example above for a continuous loop. CrewAI does not bring its own scheduler, but the OS already has good ones.
Where do agents store memory?
CrewAI supports vector memory backends (Chroma is the default). Run Chroma alongside the crew for persistent multi-run memory.
Can I monitor what the crew is doing?
CrewAI has built-in tracing (CrewAI Crews dashboard, or AgentOps, or simple log forwarding to Loki/Grafana). The systemd service's journal is the lazy first step.
Is it safe to give an agent shell access?
Treat agents like junior devs you barely trust: scope tool access, sandbox file ops, never give "run any command" without a human-in-the-loop checkpoint.
Does Servury log my CrewAI traffic?
No. No application-level logging on customer servers. Anonymous signup, crypto/card payment, no logs.
