Talk to Cursor, Claude, and ChatGPT.
Dictate prompts that ramble the way a pairing session does. Edge cases, prior attempts, constraints. All of it ends up in the prompt the agent sees.
Write code and prompts 4× faster with syntax-aware voice that handles camelCase, file paths, CLI flags, and the AI-IDE prompt the way you’d actually say it. Works in Cursor, Windsurf, Claude, and every other app or website you live in on your Mac.
Free for 14 days or 1,000 words, no card required.
"
Talk the change. Get a clean commit message.
Coding agents got good. The cap is now how fast you can prompt them. Developers type at 40 words a minute and think at 400, so the agent gets a thin slice of the context that would have produced the better diff. The richest prompt, the one with edge cases, prior attempts, and the constraints you'd mention in a pairing session, is the one a developer can't be bothered to type out.
Dictate prompts that ramble the way a pairing session does. Edge cases, prior attempts, constraints. All of it ends up in the prompt the agent sees.
File reference tagged for the prompt automatically.
camelCase, snake_case, kebab-case, file names, CLI commands. Talk about a file in Cursor and Epilude tags it for the prompt automatically.
Memoize search query to cut redundant fetches
epilude/app #482
Dictate the review, the commit message, the issue. Talking is faster than typing a paragraph of context, and the context is the part that matters.
One hotkey, every editor. No Epilude window to open.
VS Code, Cursor, Windsurf, Xcode, WebStorm, Terminal. One hotkey; no Epilude window to open.
Epilude lives inside the AI-IDE stack you already run — the PR description lands in the window you opened it from.
…and every other app you write code, PRs, or docs in on your Mac. In any app you already write in, on your Mac.
This adds backpressure to the streaming tokenizer so we stop buffering the whole response in memory on slow consumers. The parser now pauses the upstream read when the output queue hits the high-water mark and resumes once it drains. I've also added a regression test that replays the 40MB payload that triggered the OOM in
“Inlib/queue.ts, swap the in-memory queue for the new bullmq adapter. Keep the public API identical. Don’t touch the worker pool config, that’s pinned by ops. Tag@yaraon the PR.”
Free for 14 days or 1,000 words, whichever comes first. No card required. Install on your Mac in 30 seconds. Open Cursor and talk.