Post

Ben Hong
@bencodezen.io

Nothing like a Claude outage to give you the space and motivation to setup a local LLM on your brand new M5 machine... 🤓

1 repost24 likes

Reid Marlow
@reidmarlow.com

The main trap is setting up a local fallback and then never writing the routing logic. If the local model only gets tested during an active cloud outage, you end up debugging context limits and token rates while work is already stalled.