Backup brain for Claude Code. When Claude Max hits its rate limit or Claude is down, one command flips your agents to a local MLX model and they keep going.
Claude Max has a weekly limit that almost no one hits — until they're deep in a 4-hour debugging session and the agent suddenly stops. Or Anthropic has a brief outage. Or you're on a plane with no internet. claude-failover is the simplest possible answer: one shell command flips every claude -p call on your machine to a local MLX model.
Default is still Claude. The local model is only there for the moment you need it. When the cloud is back, flip back with one command.
`claude-failover gemma` and every agent is now local. Reverse it just as easy.
You still use Claude when Claude is available. Local is the safety net.
MLX models run fast on M-series. Gemma 31B is the smart default.
HQ shows which brain is running. You always know.
`brew install` + an alias. That's all the setup.
Nothing changes day-to-day. claude -p uses the cloud.
When you hit the wall: `claude-failover gemma`. Done.
It's free, open source, and runs on the hardware you already own. Head to GitHub to get started, or drop a star to help us keep building in public.
browser-agent runs the same on your firm's MacBook as it does on mine. AirGap AI is the commercial pilot — a 14-day engagement that ships browser-agent (and the rest of the local-AI stack) into a real legal/medical workflow with verified network audits. Privileged docs in, answers out, never a byte to a cloud.
This page is the front door. The full technical README — every detail, every code block, every benchmark — lives on the repository itself.