No reviews yetBe the first to leave a review for Glance
Wispr Flow: Dictation That Works Everywhere — Stop typing. Start speaking. 4x faster.
Stop typing. Start speaking. 4x faster.
Promoted
Maker
📌
Hey! I built Glance, an open-source dashboard where AI generates the widgets as React components — not config files, actual JSX.
What makes it different:
1. AI writes real code, not config
When you ask for a widget, the AI writes a React component with fetch logic, error handling, the whole thing. It's stored in SQLite and transpiled at runtime. No YAML, no templates, no pre-built widget library to choose from.
2. Agent refresh
This is the part I'm most excited about. Widgets can request a refresh from your AI agent. Click refresh → webhook fires → agent wakes up, does whatever's needed, pushes new data.
Example: I have a Claude Code usage widget. There's no API for Claude CLI usage stats — they only exist in the /usage command output. So when I refresh, my agent spawns a PTY session, runs the command, parses the output, and updates the widget. That widget can't exist in any other dashboard.
3. Bidirectional
The AI can write widgets, but it can also read them. Ask "what needs my attention?" and it reads your dashboard state to answer. Your dashboard becomes context for your agent.
The unsexy truth: This started because I wanted to see my Claude usage without opening a terminal. Then I realized the pattern generalizes — any data an AI agent can access, it can surface in a widget, even if no API exists.
Hey! I built Glance, an open-source dashboard where AI generates the widgets as React components — not config files, actual JSX.
What makes it different:
1. AI writes real code, not config
When you ask for a widget, the AI writes a React component with fetch logic, error handling, the whole thing. It's stored in SQLite and transpiled at runtime. No YAML, no templates, no pre-built widget library to choose from.
2. Agent refresh
This is the part I'm most excited about. Widgets can request a refresh from your AI agent. Click refresh → webhook fires → agent wakes up, does whatever's needed, pushes new data.
Example: I have a Claude Code usage widget. There's no API for Claude CLI usage stats — they only exist in the /usage command output. So when I refresh, my agent spawns a PTY session, runs the command, parses the output, and updates the widget. That widget can't exist in any other dashboard.
3. Bidirectional
The AI can write widgets, but it can also read them. Ask "what needs my attention?" and it reads your dashboard state to answer. Your dashboard becomes context for your agent.
The unsexy truth: This started because I wanted to see my Claude usage without opening a terminal. Then I realized the pattern generalizes — any data an AI agent can access, it can surface in a widget, even if no API exists.
Stack: Next.js, SQLite (local-first), shadcn/ui.