Meow Coding meow-coding — master
meow-coding master* TODO 6/6

One window · many sessions

Run your coding sessions side by side. None of them wait for you.

Meow Coding is a desktop app built around a first-party coding agent. Every session gets a pane and stays mounted, so the ones you are not looking at keep working. The transcript, tool calls, permissions, cost and file edits are all in the app — nothing is a black box.

> Split the parser and keep the tests green.
Build kimi-k3 Default master Enter to send · Shift+Enter for a new line
01

The app

Windows · Linux · macOS
lightmeow-app-screen-light-mode.png
Meow Coding in light mode: a session list on the left, a chat with tool cards in the middle, and a file panel on the right.
Two projects, several sessions, one chat open. The sidebar keeps every session reachable while the others keep running.
darkmeow-app-screen-dark-mode.png
Meow Coding in dark mode: the same layout with the dark theme applied.
Same window, dark theme. Both themes ship; the accent is the same blue in each.
02

Inside a session

what the agent can do

A session is a chat with the Meow agent. It reads and writes your files, runs commands, searches the web, drives a browser, and asks before it does anything you have not allowed.

Parallel

Sessions that keep running

Create, rename, stop or delete a session from its sidebar row. Move one to the background and it keeps going while you work elsewhere; a delegated sub-session nests under the session that started it.

Agent

Native Meow chat

Streaming output, markdown rendering, tool-call cards, image attachments and per-session undo and redo — no terminal to babysit.

Control

Permissions you set once

Per-tool allow, ask or deny with wildcard patterns, plus a read-only plan mode for analysis you do not want touching files.

Recovery

Undo a whole turn

Every turn snapshots the files it touched. Undo restores the files and trims the transcript; redo puts both back.

Context

Compaction before the wall

As a session approaches its context limit, older turns are summarized with a verbatim tail kept intact — so long sessions keep their thread.

Cost

What each run cost

Tokens and dollars per session and per model, cache reads included, priced from the models.dev catalog.

Steering

Talk to it mid-turn

Messages sent while a turn is running are queued in their own card — send now, edit or remove them — so you can redirect without stopping the work.

Panels

Six panels, one grid

Files, Git, Tasks, Terminal, Browser and Sub-agents dock beside the chat. They fill a shared grid, resize together, and one expands over the whole pane when you need it.

Policy

Hooks and memory

PreToolUse, PostToolUse and Stop hooks run your own policy outside the context window, and a per-project memory store keeps durable notes across sessions.

03

38 tools

every call is visible in the transcript
bash · bash_output · kill_shellRun a command, follow a background job, stop itshell
monitorWait on a background shell or a delegation without pollingshell
read · write · editRead files, create them, or apply a targeted editfiles
apply-patchApply a unified diff across several files at oncefiles
glob · grepFind files by pattern, search their contentssearch
gitStatus, diff, log, commit — without leaving the chatvcs
todowriteKeep a visible task list for the current sessionplan
questionAsk you to choose before it commits to a directionplan
taskSpawn an isolated subagent for a focused jobagents
delegate_session · delegate_resultCreate a persistent sub-session (on any model) and read its resultagents
delegate_cliHand a focused task to the Antigravity CLIagents
skillLoad a bundled or custom skill into the turnagents
webfetch · websearchRead a page, or search for oneweb
browser_*Fifteen tools over the built-in browser or a real Chrome profile: navigate, click, type, read, hand offweb
lspAsk a language server for definitions, references, symbolscode
officeCreate and edit .docx, .xlsx and .pptx filesdocs
revertUndo the files this session changedsafety

MCP servers add their own tools to the same list — stdio or HTTP — and a custom tool is a JavaScript module dropped in userData/tools.

04

Skills & commands

6 skills · 5 commands

Type / to run a command. Five built-ins cover the everyday moves, and six skills — five of them Anthropic's front-end and design skills — load into the turn when the work calls for them.

/init /review /new /compact /frontend-design frontend-design canvas-design theme-factory brand-guidelines web-artifacts-builder katalon-studio

Write your own in Markdown: $1..$N arguments, @path file contents, !cmd shell output. Subagent roles live in .meow/agents/*.md and can only narrow what your own permission rules already allow.

05

Settings

ten tabs, no config files required
Sub-agents
Roles, models, tools
Antigravity CLI
Headless delegation worker
Providers
Keys, models, connections
MCP
stdio and HTTP servers
Permissions
Allow, ask, deny
Commands
Your own slash commands
Context
Compaction thresholds
External delegation
Claude Code → Meow
Personalize
Theme, fonts, density
Updates
Background updates

Providers cover Anthropic, OpenAI, Google and any OpenAI-compatible endpoint, plus several Codex accounts you can switch between — each routed through its own scoped local proxy. API keys and OAuth tokens live in the OS keychain, and the app updates itself in the background, announcing a new version with a toast and a status-bar pill.

06

Built on open source

with their licenses

The spec → plan → execute workflow comes from obra/superpowers as a design influence; its skills and slash commands are no longer bundled, so project and user skills fill that role.

07

Install

Windows · Linux · macOS

Download the installer for your platform from the releases page, or build it yourself: npm install && npm run build. The app updates itself in the background and tells you when a new version is ready.

active Build kimi-k3 master browser: built-in v0.54.3