Repository navigation
Small Macs are in. If your Mac has 8 or 16 GB of memory, setup.sh now sees it and sets you up with a model that fits.
| Your Mac | Model | Peak memory | Speed (M4 Pro) |
|---|---|---|---|
| 16 GB | Qwen 3.5 4B, bf16, 16K context | 10.6 GB | about 22 tok/s |
| 8 GB | Qwen 3.5 2B, 8-bit, 16K context | 4.2 GB | about 92 tok/s |
Both read images. If you own one of these Macs, a report in #57 helps everyone else decide.
Install
curl -fsSL https://raw.githubusercontent.com/nicedreamzapp/claude-code-local/main/install.sh | bashWhat changed
- Doctor.
scripts/doctor.sh(macOS and Linux) and the newscripts/doctor.ps1(Windows) check the core and every add-on you installed.--fixrepairs what it can,--reportwrites a summary you can paste into a bug report. - Fewer lost tool calls.
- Gemma 4 calls that drop their opening marker on long writes are recovered instead of coming back as plain text.
<function=Name>calls written with<param>value</param>arguments now parse.- Tool names match case-insensitively, and common aliases (Run, Execute, Shell to Bash; WriteFile, Save to Write; ReadFile, Cat to Read) map to the tool you declared. Small models stop getting every call rejected.
- Windows: voice (NarrateClaude) is in the add-on menu and the uninstaller.
- Hit your usage limit? The README and docs now have a usage limit page with the exact messages and three steps to switch to local and back.
- README: the small-Mac tier, the promo video and where this is going.
Full list: v0.6.0...v0.7.0