Hey Product Hunt, Mete here 👋

I built Remarc because I kept running into the same problem with AI agents: getting a product to 90% was fast, but communicating the last 10% of feedback was painfully clunky.

Real product work is a chain of feedback loops. Change this line in the plan. Clarify that paragraph. Add a hover state to this button. Rephrase the third sentence in this draft. Chat is great for broad direction, but not for pointing at the exact thing you mean. I kept ending up with long feedback dumps or handling every detail one at a time.

So I built the feedback layer I wanted for myself. Remarc is a native macOS app for capturing comments on selected text, annotated screenshots, and web elements. It keeps the context attached and organizes comments into sessions. When you hand a session off, Claude Code, Codex, Cursor, or another MCP-compatible agent can read, work through, and update each comment.

It started as a small experiment in commenting on text anywhere on my screen. As I used it, I added the forms of feedback I reach for every day: screenshots for visual QA, web selection for UI work, voice for quick capture, and sessions for keeping different projects separate.

The bigger lesson for me is that building with AI has a last-mile problem. Models can get us surprisingly far, but high-quality work still comes from taste, QA, and precise iteration. AI makes the first pass faster. It does not make the feedback loop disappear.

Remarc is open source, local-first, and has no accounts or telemetry. If you’re an opinionated builder who cares about that final layer of polish, I’d love to know: where does your feedback loop with AI agents break down?

Hey Product Hunt, Mete here 👋

I built Remarc because I kept running into the same problem with AI agents: getting a product to 90% was fast, but communicating the last 10% of feedback was painfully clunky.

Remarc

Your feedback layer for AI collaboration.

Your feedback layer for AI collaboration.