Hero Agent

What Hero Agent is

Hero Agent is Hero Code, the autonomous AI software engineering agent — and v0.8.0 (stable, 2026-11-05) is current. Describe a project in plain English and it designs, writes and debugs real multi-file code for you, with a full IDE around it: 27 languages with syntax highlighting, a terminal, Git, a debugger, an extensions marketplace, voice input and 18 built-in skills.

• Browser vs desktop — the browser app at agent.herostudios.dev is the live version today: no install, no setup. The Windows desktop shell is built locally (0.8.0) but not published yet, so there is nothing to download; macOS and Linux installers are planned.
• What stays local — in the browser product your workspace files stay on your machine.
• What goes to the cloud — conversations, project copies, memories and metering live in Hero's D1 database, tied to your herostudios.dev account (details on the privacy notice).
• It is the shared AI backbone of the ecosystem — CraftPilot's assistant is the first planned consumer.

Prerequisites

Browser path (recommended): nothing to install — a modern browser and, optionally, a herostudios.dev account for cloud sync and the shared plan. From source (optional): the same toolchain the docs list, so every surface agrees:

• Node.js 24+
• Python 3.12+ (local backend)
• pnpm (workspace package manager)

# from the hero-agent repository
pnpm install
pnpm dev

Install & first run

Open agent.herostudios.dev, sign in (or stay signed out), and create a project — the whole first run takes about a minute.

What you'll see. Build mode streams typed progress cards while files land; Plan mode shows typed plan cards before a single line is written.

• Press Ctrl+N (or the New chat button) to start a conversation.
• Pick a mode in the composer: Chat, Build, Review or Plan.
• Pick a slot from the HERO sequence (the ✦ sequence picker).
• Describe your project — in Build/Plan mode the agent writes real files that appear in the file explorer instantly.
• Run things: open the terminal, commit with Git, or right-click any code → Ask AI to explain, fix or improve it.

Chat modes — Chat, Build, Review, Plan

The composer has four modes. You switch per message, so one conversation can go from brainstorming to shipping.

• Chat — talk it through: explanations, quick questions, refactors discussed before they happen.
• Build — the agent writes real multi-file code into your workspace, streaming typed progress cards as files land.
• Review — a scored review pass over your code: correctness, security, performance and style findings with suggested fixes.
• Plan — lays out typed plan cards describing the work before any file changes; approve a plan, then switch to Build.

The HERO sequence — seven slots

Models come from the HERO sequence: you pick a specialist, and the server resolves that slot to the best available engine. You never see vendor model ids.

• HERO NOVA ✦ — everyday answers, fast; the default companion.
• HERO FORGE ⚒ — builds real software: multi-file projects, correct imports.
• HERO SAGE ◈ — thinks before answering: architecture, hard bugs, trade-offs.
• HERO LENS ◉ — finds what's wrong: scored review findings with fixes.
• HERO SCRIBE ✎ — writes the words: docs, READMEs, comments, naming.
• HERO ATLAS ⬡ — remembers everything: long files and full-project context.
• HERO SWIFT ➤ — instant and lean: short, snappy answers.

Keyboard shortcuts

Five shortcuts cover most of the workflow — the full 68-entry keymap lives behind Ctrl+K.

• Ctrl+N — new chat
• Ctrl+B — toggle the ChatsBar (the left conversation rail)
• Ctrl+K — command palette (every action in the app)
• Ctrl+M — Memory Panel (edit, delete, export, opt out)
• Ctrl+Shift+C — chat dock (detach the conversation while you edit)

Troubleshooting

The four failure classes worth knowing before you file a report.