What does an AI agent need to work the web?

• Memory and CPU. A headless browser is heavy and can use noticeably more memory and CPU than chat. There is no universal verified RAM minimum — it depends on the browser backend and workload.
• Host. Give a browsing agent headroom, and match the server to the workload. The very smallest tiers may not be comfortable.
• Security. A browsing agent has a wider reach. A private tunnel and careful permissions matter more here.
• Cost. More memory usually means a higher host price, and web work tends to use more model tokens too.

Fixed host cost only. Web work also tends to raise model usage.

• Compute: where it runs: a cloud host, a rented GPU, or hardware you own.
• Model usage: how much it reads and writes, and how often it runs.
• Runner: the software that serves the model, or runs the agent loop and its tools.
• Capabilities: browser automation, web search, memory, or scheduled work.
• Required extras: security, delivery, monitoring, or managed-host fees.
• Freshness: whether each price is verified, aging, stale, or disputed.

AIStackPicker treats browser automation as a real capability with real resource and security effects. It does not apply a universal RAM floor; a backend-specific requirement will size the host once one is sourced. Until then it flags browser automation as resource-heavy in plain words.

• Why does browser automation change the recommendation?
• Do I need a VPS for an AI assistant?

• Cost to run an LLM
• Cost to run an AI agent
• Local LLM vs cloud API
• Run a local LLM
• GPU hosting vs model API
• Browse all stack parts
• Read the FAQ