The llm CLI now runs shell and web tools through OpenRouter
The tools that used to need a custom harness now ship as flags — while a 0.32.1 hotfix rescues broken installs and v0 apps gain logins to 100+ services.
Copy markdown
Three tools, three flags
llm-openrouter 0.7 adds server-side Shell, WebFetch and WebSearch — flip each on with `-T Shell`, `-T WebFetch` or `-T WebSearch`. Any reasoning model on OpenRouter can now run commands and pull live web data without you wiring the tool loop yourself.
0.32.1 unbreaks fresh installs
OpenAI's Python library dropped its httpx dependency, which llm leaned on indirectly, so `pip install llm` was failing on clean machines. 0.32.1 pins openai<3 as a stopgap; 0.33 will move to httpx2. Upgrade if a new install just errored out on you.
v0 apps can log into your stack
Vercel Connect lets v0-built apps securely link to Slack, Google, Notion, GitHub, Salesforce and 100+ other services with managed auth. A weekend prototype can now read and write your real tools instead of mock data.