A code editor built
for the AI agent era.

Color-coded UI — see at a glance which AI is running, and which one you are talking to

Bring your own AI agent — your existing subscription works as-is over ACP, the open agent protocol

Native performance — no Electron. Rust + GPUI, low memory and 60fps

Multiplexer — Fleet mode, a new idea for the agent era

A real capture of necoder: a Claude Code (ACP) thread reading code, streaming a summary, through to the completion toast — ● Real captureClaude Code (ACP) running inside. Thinking is sped up 16×; everything else is real time.

Color tells you where you are

The color you assign to a project carries all the way through — the rail, the tabs, the caret, the AI threads. However many projects you run side by side, you never lose track of where you are.

Try it — project color

              N
            
workspace.rs
editor_core.rs
7pub mod updater;
8use theme_core::{Theme};
9// color is reserved for identity

A real screen: five projects lined up in the rail in their colors, with colored AI thread tabs and a colored caret — Five project colors. ⌘O lists every project and worktree; ⌘1..9 switches instantly.

Your own AI agents, running as they are

An AI thread mid-generation: a ⏺ Read File step, the mascot typing, and the always-visible token meter — A thread mid-generation. nyaco sits at the top right, the token meter stays at the bottom.

Claude Code, Codex and any other ACP-compatible agent work out of the box. They run on the subscription you already have — no extra API key.

and other ACP-compatible agents

The destination, always shown — you can always see which thread and which branch you are sending to

Token usage, always shown — per-thread consumption and the running total, always in view

Diff review and checkpoints — review the changes before taking them in, and roll back whenever you want

Todo board — tasks live in a Markdown file, and the agent ticks off what it finishes

Native speed in every interaction

Written in Rust and drawn straight on the GPU by GPUI. No Electron, no webview — launch, scrolling and typing all stay light.

215ms

launch → ready to type

cold start

122MB

idle memory

RSS right after launch

10ms

⌘P filtering

measured over 50,000 files

60fps

drawn on the GPU

GPUI native UI

Measured on an M1 Mac

Fleet mode — a multiplexer for agents

Run several agents side by side on one screen, and direct them while watching the whole picture. Each agent is isolated in its own worktree, so they can work on the same repository at once, safely.

Fleet mode in full: the color-coded agent list on the left, the lineage graph in the center, the split cell grid below. — Fleet mode in full. Left: the agent list. Center: the lineage graph and the split grid. All natively drawn.

Fan view of the lineage graph — Four views of the lineage graph. {{ viewCap }}

Tree view of the lineage graph — Four views of the lineage graph. {{ viewCap }}

One cell maximized: thumbnails of the other cells across the top, the real transcript in the center.

Every cell is live

The cells are not shrunken previews — they are the running agents and terminals themselves. Work in them directly, or maximize one to focus on a single agent.

Everything you expect from an editor

LSP · tree-sitter · Git · terminal · Remote SSH · IME · hot exit

The ⌘O switcher — ⌘O — switch project / worktree

The command palette — **⌘⇧P** — every command. ⌘P stays at 10ms even across 50,000 files

A diff tab — **Git** — read diffs in a tab, side by side with HEAD

completion (auto-triggered)definition F12 / references ⇧F12rename F2 / code action ⌘.format on save tree-sitter incremental parsinghunk stage / blame⌘F find · ⌘⇧F find in projectthemes are a single JSON file