Skip to content

BetterWright 2.8.0

Choose a tag to compare

View all tags
@SSHdotCodes SSHdotCodes released this 14 Sep 01:36
· 27 commits to main since this release
v2.8.0
c416685
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

BetterWright 2.8.0 adds automatic local AI setup and Cerebras support to the built-in harness, alongside browser-provider fallback chains.

  • Run betterwright --local to select a reviewed model and quant for your hardware, install a private inference runtime, verify downloads, and configure the harness after image/tool-call checks. Compatible DFlash2/MTP acceleration is benchmarked automatically. See the local AI guide for memory, driver and platform requirements.
  • Use CEREBRAS_API_KEY with --model cerebras/qwen-3.8-27b for Cerebras vision and tool calling. betterwright models cerebras lists models. The harness preserves reasoning across turns and handles screenshot messages. See the Cerebras guide.
  • Configure ordered browser-provider fallback chains. Failed connections release newly allocated remote sessions before trying the next provider, and connected provider accounts supply API keys across launch paths.

Upgrade with bun install -g betterwright@2.8.0. Run betterwright setup if the managed browser is not installed. Local model setup is an explicit, optional betterwright --local step.

The local inference paths were exercised on Apple Metal, NVIDIA Ampere/Ada/Hopper/Blackwell, and AMD MI300X. Other hardware still needs physical validation as documented. Cerebras's public catalog and mocked API/tool-loop tests passed; authenticated Cerebras inference was not tested because no API key was supplied.

Full changelog

Assets 2
Loading