Running autonomous AI agents on Apple Silicon for $1.50/month (14 errors later)(ianlpaterson.com)

2 pointsby ianlpaterson7 hours ago1 comment

ianlpaterson7 hours ago
Author here. I use Claude Code as my primary dev environment. Wanted autonomous agents I could message from Slack without sending everything through cloud APIs.
The 14 errors are the real content. OpenClaw's docs assume you're using hosted models. Running a 24B local model on a Mac Studio exposed silent failures at every layer: model registry, auth resolution, context management, sub-agent spawning. Error #14 required patching OpenClaw's compiled JS bundle directly.
Key optimization that made it usable: KV cache quantization (Q8_0) nearly doubled usable context and got generation speed from 12 to 49 t/s. Dual-slot GPU allocation prevents sub-agents from starving the main agent.
Happy to answer questions about the setup.
- naveed1257 hours ago
  You make an interesting point about home IP address, do you have any numbers around how much is too much?
  - ianlpaterson5 hours ago
    Unclear, how much is too much... what?
    naveed1253 hours ago
    > your home IP is now tied to whatever the agent does. If an agent scrapes a site too aggressively, your ISP may notice. Route agent traffic through a VPN for anything beyond light browsing.
    What is the light browsing you mentioned? 10 sites, 100 sites. At what point you felt it as too much to be considered light browsing.