Best VPS for LobeChat (2026): Specs That Actually Fit the Agent Marketplace
LobeChat is the most polished React based open source chat UI in 2026. It has a real plugin marketplace, multi agent orchestration, and a good MCP story. The hosting decision is interesting because the default mode is genuinely light, while the server database mode bumps the resource profile up significantly.
I ran LobeChat for eight weeks across two setups, one solo on a small Hetzner box and one for a team with the server database mode.
What LobeChat Actually Needs
Two scenarios to size for:
- Client storage mode. Next.js server only. 200 to 400 MB resident. Fine on a 1 vCPU and 2 GB VPS.
- Server database mode. Adds PostgreSQL plus authentication and sync. Plan for 1.5 to 2 GB RAM at idle, more with active users.
The thing that catches people: the marketplace pulls plugin manifests on every visit. Cache busting in browsers means even returning users trigger fetches. On a 5 dollar VPS with many concurrent users you will feel that.
VPS Comparison for LobeChat
| Provider | Plan | vCPU | RAM | Disk | Monthly | Best fit |
|---|---|---|---|---|---|---|
| Hetzner Cloud | CX22 | 2 | 4 GB | 40 GB NVMe | 5.83 EUR | Solo user, client storage mode |
| Contabo VPS | VPS S | 4 | 8 GB | 100 GB NVMe | 4.50 EUR | Small team, server mode |
| Hetzner Cloud | CCX13 | 2 | 8 GB | 80 GB NVMe | 14.86 EUR | Team with multiple plugins |
| DigitalOcean | Premium AMD 2 GB | 2 | 4 GB | 80 GB NVMe | 21 USD | Solo US user |
Hetzner Cloud CX22: Solo client storage mode
LobeChat is one of the few self hosted chat apps where the smallest shared CPU plan actually works. CX22 with 4 GB RAM and 2 vCPU runs the Next.js server comfortably for one or two users in client storage mode. NVMe means cold starts after an update are quick.
Pros for solo use:
- Cheapest sensible plan that still gets NVMe
- Falkenstein latency under 80 ms to most providers
- Snapshots make rollback easy after an upgrade
Real downside: shared CPU means if your neighbor is loud, you feel it. For solo light use, that is acceptable. For a team, move up a tier.
Get Hetzner: Hetzner Cloud.
Contabo VPS S: For server mode on a budget
If you want the server database mode with multi user sync, you need a Postgres. Contabo’s VPS S at 4.50 EUR gives you the resources for the Next.js plus Postgres stack with room to spare. The trade offs are the usual:
- Slow provisioning
- Outbound latency to model APIs is higher than Hetzner
- Cheaper tiers can have noisy neighbors
For a long lived team setup that you configure once, the price is hard to beat.
Get Contabo: Contabo VPS.
Hetzner Cloud CCX13: For teams that grow
Once you have a real team using LobeChat with many active agents and plugins, CCX13 is the right tier. Dedicated vCPU and 8 GB RAM hold the Postgres plus Next.js stack with headroom for plugin worker spikes.
Get Hetzner: Hetzner Cloud.
DigitalOcean Premium AMD 2 GB: For US solo users
If you are in the US and want a US east host for low latency to OpenAI and Anthropic, the smallest Premium AMD plan handles a solo LobeChat deployment in client storage mode. Pricier than equivalents in Europe, but the snapshot story and the dashboard are friendly.
Get DigitalOcean: DigitalOcean.
Things Worth Knowing Before You Start
Three points the docs hide:
- Auth is your responsibility in server mode. LobeChat supports several providers, but you have to wire them. Clerk, NextAuth, or your own Keycloak.
- Plugin sandboxing is partial. Treat third party plugins as untrusted. Audit before you let them touch your data.
- The marketplace catalogue updates from a remote source. If you are running a fully air gapped setup, you will lose the marketplace catalogue.
What I would actually pick
If you are starting today:
- EU, solo, client mode: Hetzner CX22
- EU, team, server mode, low budget: Contabo VPS S
- EU, team, server mode, daily use: Hetzner CCX13
- US solo: DigitalOcean Premium AMD 2 GB
For the wider self hosted AI picture, see the SelfHostVPS comparison. LobeChat ships frequently and I refresh this page when major architecture changes land.
Frequently asked questions
What is the minimum VPS spec for LobeChat?
2 vCPU and 2 GB RAM is enough for a single user with a couple of agents and a remote model API. The Next.js server is lean and most of LobeChat's storage lives client side by default. Once you switch on the server side database mode and add plugins, plan for 4 to 8 GB.
Does LobeChat need a GPU on the VPS?
No, unless you also self-host the model. LobeChat is a Next.js app plus a plugin layer. All inference happens at the provider you configure. A small CPU only VPS is the normal setup.
Can LobeChat run on a 5 dollar Contabo box?
Yes for a solo user. The server side database mode plus a busy team is where it starts to feel tight. The chat UI itself runs fine on small hardware.
Should I use client storage or server database mode?
Client storage is the default and works for one user across one browser. The moment you want sync across devices or multi user, switch to the server database mode. That move adds a PostgreSQL dependency, which is what bumps the RAM requirement on the VPS.