![\"Rebon\"](\"assets/app-icon/icon.png\"){kind=icon}
\n rebon
\n\n An agent CLI for coding and more — a terminal harness that drives an\n LLM through a real agent loop with tools, sessions, permissions, and a\n full TUI.\n
\n\n ![\"npm\"](\"https://img.shields.io/npm/v/@rebon/cli?label=%40rebon%2Fcli\")
![\"Apache-2.0\"](\"https://img.shields.io/badge/license-Apache--2.0-blue\"){kind=icon}
| \n \n | \n \n rebon\n\n An agent CLI for coding and more — a terminal harness that drives an\n LLM through a real agent loop with tools, sessions, permissions, and a\n full TUI.\n \n\n | \n
![\"Rebon](\"assets/demo/core-loop.webp\")
|\n| Each `shift+tab` switches permission mode, four in all: default, plan, accept edits, auto. In auto it stops asking and finishes the rest by itself. | ![\"Rebon](\"assets/demo/modes-auto.webp\")
|\n| `/agent` sends a job to a second agent. It runs in the background, so you don't wait for it and can keep asking about something else. `/tasks` shows how long it ran, what it spent, and what it came back with. | ![\"Rebon](\"assets/demo/background-agent.webp\")
|\n| Broke something and want out of it? `/rewind` lists every message you sent earlier. Pick one and both the files and the conversation go back to how they were before you sent it. | ![\"Rebon](\"assets/demo/rewind.webp\")
|\n\n## What is in this repository\n\nThe `rebon` command-line agent and everything it needs to build: the crates, the\nplugins, the packaging chain and the release workflow.\n\nRebon's other surfaces ship on their own schedules and are **not** in this tree.\nComments here name them because the code they share was shaped by them:\n\n| Surface | Where it lives |\n| --- | --- |\n| Desktop app | ` | Project directory on that remote. Requires `--remote`. |\n\n## Config & data\n\n- `~/.rebon/config.json` — providers, models, credentials, defaults.\n- `~/.rebon/sessions/` — saved transcripts.\n- `~/.rebon/skills/`, `~/.rebon/agents/`, `~/.rebon/memory/` — user assets.\n- `$REBON_LOG_DIR/rebon.log` — TUI log file (defaults to\n `%TEMP%/rebon/logs/rebon.log` on Windows, `$TMPDIR/rebon/logs/rebon.log`\n elsewhere). `--acp` mode logs to stderr instead.\n\nProject-level overrides live under `.rebon/` and `.claude/` in your working\ndirectory.\n\n## What you get\n\n- **Local TUI** — prompt input with history, paste, image paste, `@`-mention\n and slash-command pickers, queued submissions, mode cycling; streaming\n transcript with markdown, tool grouping, thinking blocks, plan approval,\n permission modal; full dialog stack for onboarding, settings, resume,\n rewind, quick-open, history search, global search, tasks, background\n tasks, teams, agents.\n- **Agentic tool loop** — Bash / PowerShell, Read / Write / Edit, Glob /\n Grep, Sleep, TaskCreate / TaskUpdate / TaskList / TaskGet / TaskStop,\n Agent, SkillTool, AskUserQuestion, SendMessage, EnterPlanMode /\n ExitPlanMode, ToolSearch, Team{Create,Delete,Files,Mailbox,Manager},\n Worktree, plus MCP tools (stdio, Streamable HTTP, legacy SSE).\n- **Providers** — Anthropic, OpenAI, and OpenAI-Responses, with streaming,\n compact / context-prune passes, and automatic session titles.\n- **Permissions, hooks, sandbox** — fine-grained allow / deny for shell,\n filesystem, web-fetch, and skill invocations; user-defined hooks; sandbox\n config with violation reporting.\n- **Skills, agents, memory** — bundled and user skills; spawnable\n worker agents and a coordinator for background tasks; a persistent memory\n layer with recall and surfacing.\n- **Remote hosts** — run a session against a project on another\n machine over ssh. A server build is installed on the far end and the\n agent executes there — shell, filesystem, git — while this machine\n keeps the UI and the transcript. `rebon remote add`, then\n `rebon --remote