Home
rho-agent is a configurable AI agent for software development, debugging, and operations workflows. It provides a structured agent loop with built-in tool handlers for shell execution, file inspection, database access, and external service integration — all governed by permission profiles that control what each agent can and cannot do. Every session is traced to disk and can be resumed, inspected, or monitored live.
Key capabilities¶
Permission profiles control shell access, file write permissions, and database mutation behavior per agent. A readonly agent can safely inspect production systems. A developer agent can edit files and run arbitrary commands. An eval agent has unrestricted access for sandboxed benchmark execution. The Daytona backend (--backend daytona) routes shell and file tool execution to a remote cloud sandbox.
Native tool handlers give agents direct access to shells, files, databases (PostgreSQL, MySQL, Oracle, Vertica, SQLite), and Excel files — without relying on external plugins or MCP servers.
Delegation lets agents spawn focused child agents for subtasks.
Observability — every session writes an append-only trace.jsonl with token usage, costs, and tool execution. Sessions can be listed, resumed, and inspected offline. Attach custom observers for live export to external systems.
Monitor — watch, pause, steer, and cancel running agents with rho-agent monitor.
Prompt templates with YAML frontmatter and Jinja2 variable substitution make it easy to define repeatable, parameterized agent tasks.
Documentation¶
| Guide | Description |
|---|---|
| Quickstart | Get running in minutes |
| Installation | Environment setup, install options, and verification |
| Architecture | System design, session lifecycle, and tool routing |
| CLI Reference | Commands, flags, and usage examples |
| Python SDK | Create and run agents programmatically |
| Prompt Files | Template prompts with frontmatter and variables |
| Tools | Complete tool handler reference |
| Profiles | Permission profiles and custom profile YAML |
| Daytona | Remote sandbox execution via Daytona |
| Observability | Session traces, offline inspection, and observers |
| Monitor | Watch, control, and steer running agents |
| FAQ | Common questions and troubleshooting |