Apron Agents
Split coding tasks across sandboxed agents and merge small reviewed diffs behind a human gate.
Last verified:
What is Apron Agents?
Apron Agents coordinates multiple AI coding agents in isolated sandbox repositories, then merges their work through a human-reviewed interface. It's designed for developers who want AI code generation while maintaining control over quality and preventing direct access to real repositories.
Apron Agents pricing
Pricing model: Freemium
Uses existing subscriptions (Claude Code, ChatGPT plan, or Anthropic API key); includes free demo mode
Apron Agents pros
- Sandbox environment isolates agent work—real remote is never touched
- Merge cockpit UI provides live view and chunk-by-chunk review workflow
- Pluggable backends support Claude Code, ChatGPT, Anthropic API, or demo mode
- One-command setup (./run start) with no additional configuration needed
Apron Agents cons
- Requires local setup—no cloud-based version
- Manual review workflow may slow task completion compared to fully autonomous agents
- Limited to Claude, OpenAI, and Anthropic LLM providers
Frequently asked questions about Apron Agents
How does Apron prevent agents from damaging my real repository?
Agents work only in a temporary sandbox git repo created at launch. Apron never connects to or pushes to your real remote—merges are only applied to your working directory when you approve them.
Which LLM providers can I use?
Claude Code (via claude CLI), ChatGPT/OpenAI (via codex CLI), Anthropic API directly, or a built-in demo mode that requires no account.
Do I need to write custom agent definitions?
No. Apron discovers your existing .claude/agents/ definitions read-only and hot-reloads edits from its own overlay.
Can it run without human review?
Yes. Autonomous mode uses the same merge machinery but auto-merges based on test results (merges on green).