Skip to content
HN On Hacker News ↗

GitHub - y-times-y/y: A malleable coding agent app. Use Claude Code and Codex to build your projects and reshape y itself live.

▲ 37 points 23 comments by HetPatel106 2w ago HN discussion ↗

Pangram verdict · v3.3

We believe that this document is fully AI-generated

99 %

AI likelihood · overall

AI
0% human-written 100% AI-generated
SEGMENTS · HUMAN 0 of 2
SEGMENTS · AI 2 of 2
WORD COUNT 520
PEAK AI % 99% · §1
Analyzed
Jun 24
backend: pangram/v3.3
Segments scanned
2 windows
avg 260 words each
Distribution
0 / 100%
human / AI fraction
Verdict
AI
Pangram v3.3

Article text · 520 words · 2 segments analyzed

Human AI-generated
§1 AI · 99%

y is a self-modifying desktop coding-agent app. It is built around a simple idea: software should become malleable while you use it. The main interface is a chat, but the app can reshape its own UI through a protected Modify surface. You can ask y to change how y itself works, keep the change if it renders safely, or roll it back if it does not. y is not a new agent model. It is a local, chat-first workspace for the coding agents you already use: Claude Code, OpenAI Codex, and other CLI-native agents over time. Why y exists Most coding-agent apps are fixed products. You can use them, configure them, maybe install plugins, but the product itself still belongs to someone else. y is different. It treats the app as malleable software:

Chat first. The default surface is a focused conversation, not an editor clone. Self-modifying. The Modify rail can edit the live Userland UI while the protected Kernel stays locked. Local agents. Claude Code and Codex run as official local CLIs with the user's own login. Parallel work. Chats can use isolated workspaces so agents can work in parallel without stepping on the same files. Diff-gated changes. UI changes compile, render, show a diff, and can be kept or discarded. Rollback built in. The app keeps known-good snapshots so broken UI changes can recover.

How it works y is split into two parts:

Layer What it does

Protected Kernel Auth, local engine adapters, app state, safety rails, filesystem boundaries, terminal bridge, Modify rail, and rollback runway.

Mutable Userland The chat UI and app surface that can be edited live by the user or by Modify.

This split lets y feel self-modifying without giving the modification agent control over the protected core. The app can change its own interface, but the Kernel still owns the trust boundaries. Claude Code and Codex y runs coding agents locally instead of proxying them through a hosted account.

Claude Code uses the official Claude Code CLI. Codex uses the official Codex CLI. Model and effort settings stay visible in the composer. Multiple chats can run against different engines. Isolated workspaces let parallel agents work without interfering with each other.

The user's local CLI auth remains the source of truth.

§2 AI · 99%

y orchestrates the experience; it does not replace the agent providers. Download The latest macOS build is published on GitHub Releases:

Current release target: macOS Apple Silicon. Development cd app pnpm install pnpm dev Useful checks: pnpm typecheck pnpm test:ui Build a local macOS bundle: pnpm build:mac The generated app artifacts are written to app/dist/ and are intentionally not committed. Privacy and analytics y keeps the coding workflow local. Project files, terminal commands, and agent prompts are not sent to y's product analytics. Product analytics are for app usage health: sign-in state, feature usage, feedback, and missing-brick requests. Missing-brick reports are structured and should describe the missing capability, not the user's private prompt or source code. Status y is under active development. The current focus is launch-readiness for macOS: packaging, auth, analytics, local app-state durability, and the self-modifying Modify workflow. License MIT. See LICENSE.