Features
PRD-driven loop generation from raw requirements
Per-task specification with detailed steps for hundreds of tasks
Deterministic Docker sandbox per agent (ralph-<agent>-<dir>-<hash>)
Multi-agent support: Claude, Codex, Cursor, Copilot, Gemini, opencode
Live step detection and stream preview
Screenshot capture during execution
Full history logs with per-iteration timing
Mid-flight steering by editing .agent/STEERING.md
Automatic commit after loop completion for review
Network policy allowlisting for sandbox (sbx policy)
YOLO mode (--dangerously-skip-permissions) for full shell access
Sandbox debugging via sbx exec to reproduce failures
Iterates through task list until completion
CLI-based interface (no web UI)
Open-source (MIT license)
Laguna XS.2 model (33B params, 3B active, on-device)
Laguna M.1 model (225B params, 23B active, via API)
256K context length for multi-step reasoning
Multi-agent orchestration with planning and tool use
Sandboxed agent execution environments
Developer surfaces: agents, TUI, IDE extensions, binaries
Custom foundation models deployed on-prem or in VPC
Data connectors to repos, databases, data warehouses
Role-based access control for humans and agents
Executive-grade governance and auditability
On-prem, VPC, and workstation deployment (defense only)
Forward Deployed Research Engineers embed with teams
Real-time observability with end-to-end traces
Air-gapped network support
Open-weight model weights available
