argus

A local web app that orchestrates Claude Code agents from a real UI.

---

Workspaces, persistent agent personas, pinned skills, live output streams, all wrapping the claude CLI you already have. Uses your existing Max plan, no Anthropic API key required.

Named after Argus Panoptes, the many-eyed giant who watched everything.

What it does

• Workspaces point at folders on disk. Each agent in that workspace runs with the folder as its cwd and inherits any .claude/ config you already have there.
• Agents are reusable claude-code personas (name, system prompt, model, pinned skills). Create one once, run it many times.
• Runs stream stdout from claude back to the browser as Server-Sent Events. You see what the agent is doing in real time, no terminal needed.
• History is persisted in SQLite. Refresh the page, your workspaces and agents are still there.

Requirements

• Node.js 20+
• pnpm (or npm / yarn — pnpm recommended)
• claude CLI installed and authenticated on the same machine (install instructions)
• macOS or Linux (Windows untested, may work in WSL)

Setup

git clone <your-fork-url> argus
cd argus
pnpm install
pnpm dev

Open http://localhost:3000.

The SQLite database (argus.db) is created automatically on first run. It lives at the project root and is gitignored.

Usage

1. Create a workspace. Give it a name and the absolute path to a code folder you work in.
2. Create an agent. Set a name, optional system prompt (the agent's persistent role), optional model, optional skills to pin.
3. Run a prompt. Type a prompt, hit Run, watch claude output stream in.

Architecture

• Frontend: Next.js 15 App Router, React Server Components, Tailwind v4.
• Database: SQLite via better-sqlite3 + Drizzle ORM. Tables auto-created on first import.
• Agent execution: child-process spawn of claude -p ... --output-format=stream-json --verbose. Each JSONL line is forwarded to the browser via SSE.
• Skill pinning: writes an argus.pinnedSkills block into the workspace's .claude/settings.local.json before each run. Merges with whatever's already there.

Scripts

pnpm dev          # start dev server
pnpm build        # production build
pnpm start        # serve production build
pnpm db:generate  # generate Drizzle migration from schema changes
pnpm db:migrate   # apply migrations
pnpm db:studio    # open Drizzle Studio (DB inspector)

Roadmap (rough, not committed)

• v0.2: persistent run history view per agent, output search across runs
• v0.3: multi-agent parallel runs in one workspace, output side-by-side
• v0.4: skill marketplace UI (browse .claude/skills/, toggle on/off per agent)
• v0.5: webhook hooks (run an agent on a cron, on a git push, on a file change)
• v0.6: optional cryptographic attestation of each agent run (per-action receipts)

License

MIT. See LICENSE.

Why not just use Claude Code directly?

Claude Code is a fantastic CLI. Argus is for when you want:

• A persistent persona library (instead of typing the same --append-system-prompt flag every time)
• A visual workspace with multiple agents running side-by-side
• A web UI you can open on another monitor while you code in your terminal
• Output history that survives terminal session resets
• Skill pinning that's discoverable in a UI, not buried in a settings file

If none of those matter to you, the raw claude CLI is faster.