Best VPS for Agno (2026): Specs for the Phidata Multi-Modal Agent Framework
Agno, formerly Phidata, is the agent framework that bet hard on multi modality before the rest caught up. By 2026 it is one of the cleanest Python options for building agents that handle text, images, audio, and structured tools without writing custom plumbing for each.
I ran Agno on three VPS tiers for a month each, testing single agent, multi agent, and the optional playground app.
What Agno Actually Needs
Three resource curves:
- The framework. Lean Python core. 200 to 400 MB resident.
- Tool stack. Each tool added has its own footprint. Vision tools pull image processing libs, audio tools pull whisper or transcription libs.
- Knowledge layer. Vector store plus embeddings if you enable it. 300 MB to 2 GB depending on corpus size.
The thing that catches people: the multi modal default means Agno pulls dependencies for vision and audio even if you only use text. Strip down your imports to keep the install small.
VPS Comparison for Agno
| Provider | Plan | vCPU | RAM | Disk | Monthly | Best fit |
|---|---|---|---|---|---|---|
| Hetzner Cloud | CX22 | 2 | 4 GB | 40 GB NVMe | 5.83 EUR | Prototyping, single agent |
| Hetzner Cloud | CCX13 | 2 | 8 GB | 80 GB NVMe | 14.86 EUR | Production single agent, knowledge enabled |
| Contabo VPS | VPS S | 4 | 8 GB | 100 GB NVMe | 4.50 EUR | Budget production |
| Hetzner Cloud | CCX23 | 4 | 16 GB | 160 GB NVMe | 29.74 EUR | Multi agent, heavy tool stack |
Hetzner Cloud CX22: Prototyping pick
For development and prototyping, the cheapest shared CPU plan is enough. Single agent with a remote model fits comfortably. NVMe and decent latency make the dev loop tight.
Pros for development use:
- Cheapest sensible Hetzner plan
- Snapshots are useful when you break the agent config
- Falkenstein latency is good for working with EU based model endpoints
Real downside: shared CPU means inconsistent inference timing during load. Not a production pick.
Get Hetzner: Hetzner Cloud.
Hetzner Cloud CCX13: Production single agent
Once you move from prototyping to production with a single agent plus knowledge, CCX13 is the right tier. Dedicated vCPU keeps the agent loop responsive, 8 GB RAM holds the framework plus a moderate vector store.
Get Hetzner: Hetzner Cloud.
Contabo VPS S: Budget production pick
4 vCPU and 8 GB at 4.50 EUR. For a single Agno agent in production, this works fine on the newer NVMe Contabo plans. The trade offs are familiar:
- Slow provisioning
- Outbound latency higher than Hetzner
- Cheaper tiers have variable performance
For a long lived production setup, the cost wins.
Get Contabo: Contabo VPS.
Hetzner Cloud CCX23: Multi agent or heavy multi modal
If you run Agno in a team setting with multi modal tools (vision plus audio plus knowledge), CCX23 with 16 GB RAM is the right step up. The tool stack adds up fast and you want the headroom.
Get Hetzner: Hetzner Cloud.
Things Worth Knowing
Three points the docs do not stress:
- Tool selection at agent level. Do not give every agent every tool. The system prompt grows and you pay for context you do not use.
- Storage abstraction is flexible. Pick PostgreSQL or SQLite for memory based on actual need, not the example. Postgres for production, SQLite for prototyping.
- The playground is convenient but optional. Run it in development, drop it for production unless you need a visual debugger.
What I would actually pick
If you are starting today:
- Prototyping: Hetzner CX22
- Production single agent: Hetzner CCX13
- Budget production: Contabo VPS S
- Multi agent multi modal: Hetzner CCX23
For the broader self hosting picture, see the SelfHostVPS comparison. Agno ships frequently and I refresh this page when major framework changes land.
Frequently asked questions
What is the minimum VPS spec for Agno?
2 vCPU and 4 GB RAM is the floor for a single agent with a remote model. Agno is a Python framework with a light footprint. The break point is when you turn on knowledge tools, vision, or audio support, which add embeddings and processing pipelines.
Does Agno need a GPU on the VPS?
No, unless you also run the model locally. Agno is an orchestration framework that calls whatever provider you wire in. A CPU only Hetzner or Contabo VPS is the standard setup.
Can Agno run on a 5 dollar Contabo box?
Yes for prototyping and small workflows. Single agent with a remote model and a small knowledge base fits comfortably. Once you stack multi modal tools or run parallel agents you outgrow the tier.
Should I use Agno with the playground or just Python?
Playground is great for development, optional for production. If you only need the API surface, you can skip it and save resources. If you want the visual debugger, plan for an extra 300 MB resident plus the Next.js dependency.