Free agency tools / AI Agent VPS Calculator
What server does your agent fleet need?
Running Claude Code, MCP servers or autonomous agents on a VPS? Size the vCPU and RAM from how many run at once and how heavy each is — so you provision enough, but don't pay for idle cores.
How it works
- 1Enter how many agents you run in total and what share run at the same time at peak.
- 2Pick a workload level — this sets a realistic vCPU and RAM budget per concurrent agent.
- 3We add operating-system overhead and recommend a VPS size that fits your peak with headroom.
Frequently asked
Why size on concurrency, not total agents?
Agents that aren't actively working use almost no CPU. What determines the server you need is how many run at the same moment — a fleet of 100 that peaks at 10 concurrent needs far less than 100× one agent.
What do the workload levels mean?
Light ≈ 0.25 vCPU / 0.5 GB (a chat or simple tool call). Medium ≈ 0.5 vCPU / 1.5 GB (a coding agent that reads files and browses). Heavy ≈ 1 vCPU / 3 GB (running builds, test suites or a headless browser).
Does RAM or CPU usually decide the plan?
For agent work it's often RAM — model context, a Node/Python runtime and a headless browser add up fast. We size for whichever hits a tier first and show both.
Is this exact?
It's a planning estimate. Real usage depends on your models, tools and how spiky the work is — start here, then watch actual CPU and memory and adjust.