Claude Code
Anthropic's agentic coding CLI. Reads codebase, edits files, runs commands. 🪐
Upstream: anthropics/claude-code ↗
Resources
| vCPUs | 1 |
|---|---|
| Memory | 2048 MB |
| Volume | 5 GB |
| Workspace UI | Terminal (ttyd-wrapped) |
Create one
gentity compute create --agent claude-code --model claude-sonnet-4-5First run
Claude Code is CLI-only. Gentity wraps it in ttydso the workspace URL renders the TUI in your browser. Open the workspace and you're dropped straight into claude running under the node user with /workspace as cwd.
The container starts with --permission-mode auto — a classifier auto-approves low-risk tool calls and only prompts for truly dangerous ones. To change this, edit ~/.claude/settings.json inside the workspace.
A fresh /workspace has an empty git init applied at first boot so Claude's diff / commit flows work without setup. Your files persist across stop / start; only delete destroys the volume.
Models
| Provider | Name | Label |
|---|---|---|
anthropic | claude-sonnet-4-5 | Claude Sonnet 4.5 (recommended) |
anthropic | claude-haiku-4-5 | Claude Haiku 4.5 (fast) |
anthropic | claude-opus-4-6 | Claude Opus 4.6 (deep) |
Notes / known limits
Auto-update is pinned off (DISABLE_AUTOUPDATER=1) for reproducibility — restarting an instance won't silently bump the Claude Code version. New CLI versions ship via Gentity image rebuilds.