Hey Product Hunt! 👋 I’m Niteesh, creator of Copak.

Like many of you, my daily coding workflow is now split across multiple AI agents: Claude Code in the terminal, Cursor in the IDE, and ChatGPT/Claude web interfaces for high-level architectural brainstorming.

The single most frustrating part of this workflow isn’t the models IS it’s Context Amnesia.

Every time you open a fresh terminal or switch agents, you hit the same wall:

- You have to re-explain the goal and what you just finished.

- You have to remind the model of rejected approaches so it doesn’t revert your architecture.

- You have to manually inspect git status and copy diff hunks.

- You constantly run the risk of pasting a dirty .env or sensitive API key.

Most solutions try to solve this with "autonomous agent memory" or vector databases. In practice, that creates drift, hallucinations, and token bloat.

I built Copak around a simpler, deterministic principle: developers know the intent; machines know the repo state.

Copak pairs your explicit session notes (goals, next tasks, architectural decisions, known bugs) with an automated, sanitized project snapshot. It scrubs your credentials through a multi-pass regex engine, bounds your project rules (AGENTS.md, CLAUDE.md, .cursorrules), clamps everything to a strict token budget ceiling (1k, 3.5k, or 10k), and outputs a provider-tailored prompt packet.

Press ⌘O to open your repo, ⌘⇧H to preview, and ⌘↵ to copy.

It’s 100% local, native macOS (Swift), zero telemetry, and completely open source under the MIT license.

I’d love for you to try it on your current project and let me know: What agent do you spend the most time context-switching between? What else should Copak ingest or sanitize?

Looking forward to your feedback and thoughts! 🚀