RunOnce – Run one-off LLM scripts from Windows context menu
Windows context menu integration for LLM scripts avoids terminal context switching entirely.
Sidebar AI assistant with Inline Prompting for follow-ups.
The ability to 'fork' a chat into nested child cards that preserve exact DOM selection and prune the history array is a clever, pragmatic answer to 'context pollution'. Real-time token streaming, local-only API key storage, and CSS tricks like contain: paint show attention to performance and privacy. It's a browser-hack that feels powerful for deep research, but it will remain fragile and tightly coupled to Gemini's UI unless it expands to other providers or formalizes a more stable integration layer.
Researchers, prompt engineers, and power users who conduct deep, exploratory conversations with LLMs in the browser
Windows context menu integration for LLM scripts avoids terminal context switching entirely.
Git metaphor for agent memory is clever; execution and adoption remain unproven.
Makes token bloat visible like bundle size, but GitHub Actions badge metrics already exist.
Solves real LLM cost mystery with zero-code API interception across Claude, GPT, Gemini.
Cuts cargo test output from 61 lines to 1 — saves 60-90% of wasted LLM tokens.
Tiered context summarization beats naive token culling for long agent sessions.