Show HN: Inflow – invoke an LLM with your viewport just by typing (github.com)

Inflow is a lightweight Chrome extension (Show HN) that lets you summon an LLM chat overlay simply by starting to type on any webpage — no hotkeys or input fields required. After loading the unpacked extension and configuring an API provider (OpenRouter, OpenAI, Google Gemini, Anthropic or any OpenAI-compatible endpoint) and your key, an overlay appears with your initial keystrokes pre-filled. The tool automatically grabs the current viewport/page content as context, adapts its UI colors to the site, streams responses in real time (expects OpenAI-style SSE), and closes with Esc or a click outside. For AI/ML users this is a practical, low-friction way to get context-aware model assistance directly in the browsing flow—useful for code review, research, documentation or in-situ summarization. Technical trade-offs are notable: it injects a chat panel into pages and requests broad <all_urls> permissions to operate everywhere, though the developer says API keys are stored locally and keystroke detection is also local-only. The streaming parser’s OpenAI-compatible SSE requirement and the local-key model make it flexible across providers, but the permission model and page injection mean users should weigh convenience against security and privacy risks.