Native macOS AI IDE · early prototype

Pair-program with a flock of specialists.

Duckies AI puts a team of AI reviewers, teachers, and implementers right in your editor — reviewing the lines you're looking at, explaining the ones worth understanding, and shipping the changes you approve. And they remember you.

4 ducks on duty · runs on your Mac
Duckies AI — Review
Duckies AI reviewing a TypeScript file with an inline annotation card from the Backend duck

AI help shouldn't live in a chat box.

Today's assistants sit off to the side. You copy code out, paste answers back, and the moment you close the tab they forget who you are, what your team's standards are, and what you were doing. Duckies works where you work.

// the chat-box way

Context-switch to a sidebar. Copy the code in, paste the answer out.

One generalist model guessing at your intent.

Zero memory — every session starts from scratch.

// the Duckies way

Specialists annotate the exact lines you're looking at, in the editor — no copy-paste, no context-switch.

A team of engineers each bring their own point of view, with a lead who routes your request to the right one.

Tiered memory that learns you and your standards, and carries across every session and project.


A whole team, not one generalist.

Real code review is a conversation between specialists — someone who owns the data layer, someone who owns the UI, someone who watches the words. Duckies gives you that team, each with their own point of view.

🦆

Engineering Lead

Triage & routing

Reads what you're asking and pulls in the right specialist — or a couple of them.

⚙️

Backend Engineer

Data · performance · APIs

Correctness, data flow, and the performance cliffs before they reach production.

🎨

Frontend Engineer

UI · state · accessibility

Components, state, styling, and the accessibility details that get skipped.

✍️

Technical Writer

Docs · naming · clarity

Comments, naming, and prose that makes the next reader's job easier.

Security, reliability, and more specialists join the team on the roadmap.

Learn it, review it, or ship it.

🎓

Learn

A mentor duck walks you through code you didn't write — what it does, why it's built this way, and what's worth questioning.

🔍

Review

Line-anchored feedback tagged blocker, suggestion, or nit — each with a one-click "apply the fix."

Implement

Approve a suggestion and watch a duck implement it — every tool call live, with a diff to review before you keep it.


Ambient, specialized, and yours.

Tiered memory

The ducks learn you.

Four tiers of memory — your standards, this project, this session, and each duck's own expertise — feed into every review and reply. The ducks pick up durable facts from your conversations automatically, but nothing is stored until you approve it.

The Duckies memory panel with a review queue and stored memories
Teachable moments

Learn the lines worth learning.

Ask Duckies to scan a file and it drops lightbulbs on the genuinely interesting spots — a clever pattern, a subtle risk, a decision worth understanding. Click one to learn more, tuned to your skill level.

Teachable-moment lightbulbs in the editor gutter with an explanation popover
Guided learning

A mentor for unfamiliar code.

Open something you didn't write and the mentor duck lays out what you're looking at, why the author may have done it this way, and the questions worth asking — walking you through it one note at a time.

The Learn-mode mentor panel with teaching-note cards

Feedback pitched at your level.

Duckies reads your level from how you work — the vocabulary you use, the questions you ask — and calibrates on its own. Same code, the right depth of feedback for whoever's reading it.

learned from your conversations — no dial to flip

🎓

Review above your level

A junior can open a staff engineer's PR and actually review it — the ducks explain the unfamiliar patterns and the why behind each decision, turning an intimidating diff into a guided read.

🎯

No feedback fatigue

A principal gets tradeoffs and architecture, with the fundamentals they already know left unsaid — so review stays dense with signal instead of restating the basics.


The roadmap.

A working prototype today, on a path toward always-on, local-first, deeply personal AI pairing.

Now — working today

  • A specialist team with a lead that routes your request
  • Learn · Review · Implement modes
  • Inline severity reviews with one-click apply & diff
  • Tiered memory — manual and auto, review-before-save
  • Skill level learned from your conversations, feedback auto-calibrated
  • Teachable-moment lightbulbs, tuned to skill level
  • Native macOS app; offline-self-contained UI

Later

  • A marketplace to "hire" specialist ducks — each declaring who they are and the models + effort they're tuned for, so a principal backend reviewer isn't paired with a model that can't keep up
  • A production-grade backend — models and services abstracted behind each duck, well past a single local CLI
  • Deeper project understanding across a whole codebase
  • More languages, frameworks, and platforms

Code review that finally remembers you.

A flock of specialists in your editor — reviewing, teaching, and shipping alongside you.

// early prototype · a native macOS app that runs on your machine