BBARIT Terminal (빠릿터미널)

The free AI vibe coding IDE — AI terminal, IDE and coding agent unified, built 100% in Rust, with every AI agent, Office editor and dev tool built in.

formerly Octo Terminal

Open all your projects at once and switch between them instantly — each with its own terminals, AI agents, and tools. Run Claude · Codex · Gemini · Kimi · Qwen · OpenCode · Ollama side by side, pair two of them into a self-reviewing dev team that ships while you sleep, edit Word / Excel / PowerPoint files without leaving the app — all from a fast native desktop app.

Free with GitHub sign-in · auto-updating · built with Tauri + Rust + React.

⬇️ Download

macOS repo: bbarit/terminal · Windows repo: bbarit/terminal-win

🎬 Watch the 2-minute demo

▶ Watch the full 2-minute demo on YouTube →

📖 Table of contents

• Why BBARIT?
• Who it's for
• See it in action
• The core — every codebase on one screen
• ⚡ BBARIT Agent — the built-in coding agent
• 🦌 Broker Agent — your autonomous AI dev/review team
• Multi-AI terminals
• Terminal & editor
• Knowledge that compounds
• Developer tools
• Remote & web access
• Built-in browser + automation
• Full feature list
• How it's built
• Quick start
• Requirements & install
• FAQ
• Support & community

🤔 Why BBARIT?

Most "AI IDEs" lock you into one model and one workflow, hide the AI behind a chat box, and still make you babysit every step. Meanwhile your real work is spread across a dozen terminals, a browser, an editor, a notes app, a Kanban board, and three SSH sessions.

BBARIT collapses all of that into one native terminal — and then goes further: it ships ⚡ BBARIT Agent, its own built-in coding agent that codes, tests and verifies before it's done, puts every other AI coding agent in the same window, gives them real tools (PTY, git, editors, databases, MCP, a browser), and lets two of them work as an autonomous team that designs, builds, tests, reviews, and merges — with a knowledge base that gets smarter every session.

It's not a chat wrapper. It's a cockpit for shipping with AI.

👤 Who it's for

• Solo builders & indie hackers who want an AI team without hiring one — ship features while you focus on product.
• Vibe coders who think out loud to an AI and want it to actually do the work, not just suggest.
• Terminal-native developers who never want to leave the keyboard, but want modern tools (editors, git UI, search) one keypress away.
• Knowledge-driven developers who want a wiki, notes, and design rules that compound across every project.
• Remote / on-the-go developers who need to check on long-running agents from a browser or phone.

🎥 See it in action

▶️ Watch the full 2-minute demo  ·  intro clip

📣 As featured by creators

"99% of AI coding tools are too hard. Octo Terminal is the 1%."

— @aiexplained412 on Instagram  ·  ▶️ Watch the reel

In the reel, the creator walks through why most AI coding tools are overwhelming — and shows how Octo Terminal (now BBARIT Terminal) strips that away: every AI agent and every dev tool in one native terminal, where you just say what you want and the AI does the work. It's the rare AI IDE built to be easy, not just powerful.

🗂️ The core — every codebase on one screen

BBARIT's killer feature is effortless multi-project, multi-codebase management. Stop juggling a dozen windows — put every repo in one place and move between them at the speed of thought.

The difference from VS Code: VS Code opens a separate window per folder. BBARIT keeps every codebase — and all of their terminals — on a single screen, so you manage all your projects and AI agents without ever switching windows.

• All your repos, side by side. Open dozens of projects at once; they sit in a list on the left. Jump between them instantly with a shortcut — no re-opening windows, no lost context.
• Every codebase's terminals, in one place. Each project's terminals (dev servers, AI agents, logs) stay live and visible together — no window-hopping to check another repo's terminal.
• Each codebase is fully its own. Every project keeps its own terminals, AI agents, theme, font, and working directory (local or SSH). Switch projects and everything follows.
• Search across all codebases. Ctrl/Cmd+P quick-open searches every project at once — not just the current one — and it's instant.
• Split, stack & monitor. Run a dev server in one pane, Claude in another, tests in a third; rearrange freely. Monitor mode gives you a CCTV-style overview of every service across every codebase, live.
• Drag, pin, restyle. Reorder projects by drag-and-drop, pin favorites to the top, give each its own look.

One window. Every codebase. Zero context-switching tax.

✨ Headline — ⚡ BBARIT Agent — the built-in coding agent

BBARIT Agent (빠릿에이전트) is BBARIT's own coding agent — not a chat box bolted onto an editor, but a real agent with real tools that reads, edits, runs and verifies your code, and won't call a job done until the tests actually pass.

• ✅ It verifies its own work. Change code but run nothing? The agent refuses to finish — it runs your tests or build, reads the failures, and fixes them. Judged by the exit code, not by the model's word.
• 🔧 Safe, surgical edits. Line-anchored editing lands changes on the exact line even as code shifts. It must read a file before editing it, and blind edits to files that changed on disk are blocked.
• ⏪ Auto-rollback & guardrails. Every file change is snapshotted — roll back anytime. Catastrophic commands (rm -rf /, mkfs, raw disk writes) are blocked, and untrusted projects can't be changed silently.
• 🧭 Understands your codebase. BM25 + semantic code search, a dependency & impact graph (what breaks if I change this?), and a task planner — so it edits with context, not guesswork.
• 🔌 Any model you want. OpenAI, Anthropic, Google, Mistral, AWS Bedrock, Azure — 37 providers and 1,000+ models built in. Log in with your ChatGPT or Claude subscription, or run Ollama fully local.
• 👥 297 experts + parallel subagents. Hand a task to one of 297 built-in expert personas, or fan work out to several subagents at once. Extend it with MCP servers and lifecycle hooks.

Talk to it in the terminal, in the chat panel, or from your phone — it codes, verifies, and rolls back safely.

🦌 Broker Agent — your autonomous AI dev/review team

Press one button (🦌) and BBARIT opens two real terminals — a Developer (e.g. Claude) and a Reviewer (e.g. Codex) — and runs a transparent mechanical broker between them that mediates the entire dev↔review loop. You give the goal; the pair designs, builds, tests, reviews, and merges. You approve.

The loop, step by step

1. You give the developer a task in plain language — just like talking to Claude normally.
2. The developer implements and tests its own work, keeps a project wiki, and appends a one-line DONE: to a feed file when a unit is genuinely finished.
3. The broker sees that signal, reads the git diff, and judges it mechanically — source-growth vs a baseline SHA, code-graph blast radius, sensitivity patterns (secrets/auth/destructive/SQL/payment), and requirements progress. No extra LLM calls, zero token cost.
4. The broker hands the change to the reviewer — and delivers the diff content inline, so the reviewer doesn't burn tokens re-opening files.
5. The reviewer reviews (Linus-style: direct, specific) and replies APPROVED or REWORK: …. It owns the rework loop, re-issuing instructions until the developer actually fixes it.
6. On APPROVED, the broker auto-commits and merges the harness's isolated branch — only if git checks pass. A failed merge is escalated to you, never forced.

What makes it genuinely different

• 🔍 Transparent, not a black box. Two visible terminals. Every message, diff, and decision is tracked in git + the Karpathy Wiki — reproducible and auditable, never screen-scraped.
• 🧠 Mechanical broker brain (LLM = 0). The mediation logic is pure code — growth, impact, sensitivity, RTM — so coordination is free, fast, and deterministic.
• 🌳 Worktree isolation. Each harness runs on its own git branch + worktree (.octo-tmp/broker-wt-…). It never touches your working tree, so you can run several features in parallel and merge them later.
• 🧩 Multi-harness manager. A full-screen workspace with a project → harness tree: spin up many dev/review pairs across projects, switch between them, each with its own live status window (Working / Done), broker log, and two terminals.
• 🔄 Hot-swap models mid-session. Change the developer or reviewer agent any time; the new agent reads the wiki + git and continues the in-progress work — no restart from scratch.
• 💾 Session persistence. Harnesses are saved and auto-restore on reopen/restart, continuing exactly where they left off — including queued auto-merges.
• 📋 Requirements traceability (RTM). Every requirement is a tracked row, verified against real code + real tests before anything is called done.
• 📈 Self-improving. Recurring review findings are promoted into permanent design rules, so the same mistakes get caught earlier next time.
• 🪜 Tiered knowledge. Project wiki → global wiki → other projects' code when stuck — learning your patterns and conventions.
• 🛑 You stay in control. Agents propose; merge & deploy require your approval. Always.

One click → a Claude+Codex pair that designs, builds, tests, reviews, and records — while you just give the goal.

🤖 Multi-AI terminals

BBARIT speaks every major AI coding CLI fluently — each in its own real PTY session, with independent model, flags, and autonomy mode.

• Per-project agent settings — different model/flags per project.
• Autonomy modes — normal · auto-accept · full-auto · yolo, switchable per terminal.
• AI Roundtable — have multiple models debate a question and compare answers.
• Orchestra — orchestrate a multi-AI workflow across models from one panel.

🖥️ Terminal & editor

A real terminal — not a web emulation. Built on portable-pty + Rust, rendered with WebGL.

• Multi-project, multi-terminal — open dozens of projects, each with its own terminals, themes, and fonts. Switch instantly with Ctrl/Cmd+Shift. Ctrl/Cmd+P quick-open searches across all projects and is blazing fast.
• Split panes — resizable tree layout, tabs, horizontal/vertical/grid/tile — run an agent in one pane, your dev server in another, a reviewer in a third.
• Monitor mode — a CCTV-style overview of every service across every project, with live activity dots.
• WebGL rendering — GPU-accelerated, Unicode 11, ligatures, true color, 75+ themes, custom coding fonts.
• First-class CJK / IME — Korean / Chinese / Japanese composition done right, with UTF-8 shells out of the box.
• Copy that just works — copy-on-select, right-click copy/paste, search, click-to-open links.

Every file viewer & editor built in — you never leave the app to open a file:

📚 Knowledge that compounds

• Karpathy Wiki — a structured, two-tier (project + global) knowledge base your agents read first and write back to. Kept small with per-topic files and auto-archiving, so it never bloats into one giant document.
• Notes / knowledge graph — a markdown vault with wikilinks, backlinks, tags, and an Obsidian-style graph view (4 layouts, minimap, glow). AI can auto-generate and auto-organize notes.
• Skills library — thousands of curated skills, GitHub skill-pack import, parameterized templates with {{variables}}, and one-click run. Includes the built-in Karpathy guidelines pack (think-before-coding, simplicity, surgical changes, goal-driven).
• Prompt library — save, favorite, and reuse your best prompts across projects.
• Self-improving harness — background review distills recurring lessons into durable rules.

🛠️ Developer tools

• Git panel — status, diff, history, branches, stashes, tags, worktrees, merge; one-click worktree for features or harnesses.
• Code search — hybrid BM25 + semantic over your source, plus a code-graph (defs / callers / impact with risk scoring) — the same engine the broker uses to judge change blast-radius.
• Tasks — global Kanban (6 columns, 5 priorities), Todo, and a Gantt planner that links tasks to projects/terminals. Imports GitHub & Linear issues → send a card to a terminal so the agent fixes it (one-click support automation).
• MCP — connect Model Context Protocol servers (form or JSON), browse/invoke tools, with a built-in browser MCP bridge and 100+ curated presets (Gmail, Slack, Linear, Notion, GitHub, Postgres…).
• Database — MySQL & PostgreSQL: connection test, schema browser, query runner; credentials stored locally.
• PM2 / process monitoring — real-time CPU/memory gauges, log streaming, and AI anomaly detection (via Ollama) for SSH servers.
• AutoResearch — a Karpathy-style autonomous ML-experiment panel.

🔌 Remote & web access

• 🌐 Web Terminal — open a full terminal, file explorer, and editor from any browser (phone or desktop) over a Cloudflare tunnel. Real PTY, not an HTTP shim.
• 📱 Telegram · Discord · Slack — mirror terminal I/O and drive sessions remotely from your phone (passwords auto-masked, per-project channels).
• 🔑 SSH projects — file tree, editor, and terminal work transparently over SSH — a remote project behaves exactly like a local one.
• 🔗 GitHub · Linear Hub — a unified, color-coded feed of all your issues, per-project.

🌐 Built-in browser + automation

A real Chromium browser inside the app, wired to your agents:

• Pick — click any element → CSS selector, XPath, computed styles → straight into an AI prompt.
• Live — real-time console errors & page changes, auto-notified.
• Page → Markdown — Readability + Turndown to convert any page to clean Markdown (and copy it).
• AI Fix — auto-diagnose console errors + suggest fixes.
• Test — auto-generate test cases from the live page.
• MCP bridge — your Claude/Codex CLI can navigate, screenshot, get_page_text, javascript_eval, manage tabs, read console/network — all through the built-in browser.

📋 Full feature list

AI agents & orchestration

• ⚡ BBARIT Agent — built-in coding agent: enforced verify loop (it must run your tests/build and pass, judged by the exit code — not the model's word), line-anchored safe edits with read-before-edit and stale-file guards, per-change snapshots with instant rollback, catastrophic-command blocking (rm -rf /, mkfs, raw disk writes), BM25 + semantic code search & impact graph, task planner & plan mode, 297 expert personas, parallel subagents & multi-agent teams, MCP client & lifecycle hooks, 37 providers / 1,000+ models (log in with your ChatGPT or Claude subscription, use API keys, or run Ollama fully local), usable from the terminal TUI, the chat panel, or your phone
• Multi-agent terminals: Claude Code, Codex, Gemini, Kimi, Qwen, Cursor, OpenCode, Ollama, Pi, Reasonix
• 🦌 Broker Agent — autonomous AI dev↔review pair: mechanical broker (no extra LLM cost), git + wiki channel, DONE-signal flow, worktree isolation + headless auto-merge, RTM requirements traceability, live status window, multi-harness manager, hot-swap models, session persistence/auto-resume, self-improving rules, human approval gate
• AI Roundtable (multi-model debate), Orchestra multi-AI workflow orchestration
• Per-project agent flags, models, autonomy modes (normal / auto-accept / full-auto / yolo)

Terminal & editor

• Native PTY (portable-pty + Rust), WebGL renderer, Unicode 11, ligatures, UTF-8 shells
• Resizable split panes (tree layout), tabs, multi-project tabs with per-project themes & fonts (75+ themes)
• Monitor mode (all-services overview), copy-on-select, right-click copy/paste, search, click-to-open links
• Monaco editor (Vim, Emmet, Prettier) + viewers: Markdown, PDF (edit), Word, PowerPoint, Excel, EPUB (read & create), SQLite, Mermaid, JSON tree, images (+crop), video, audio
• First-class CJK/IME (Korean/Chinese/Japanese), command palette, configurable keybindings (Octo / VSCode / IntelliJ presets)

Dev tools

• Git panel (status, diff, history, branches, stashes, tags, worktrees, merge)
• Hybrid code search (BM25 + semantic) + code-graph (defs / callers / impact + risk)
• Kanban board, Todo, Gantt planner (global across projects), GitHub/Linear issue import → terminal
• MCP server connections + built-in browser MCP bridge + 100+ presets
• Database: MySQL & PostgreSQL with schema browsing
• PM2 / process monitoring with AI anomaly detection, server Hub, SSH remote projects

Knowledge & skills

• Karpathy Wiki — two-tier (project + global) structured knowledge base, per-topic files, auto-archiving
• Notes / knowledge base with wikilinks, backlinks, tags, Obsidian-style graph view, templates
• Skills library (thousands curated) + GitHub skill-pack import + Karpathy guidelines pack
• Self-improving harness (background review → durable rules), prompt library
• AutoResearch (Karpathy-style autonomous ML experiment panel)

Remote, web & team

• Web Terminal (browser access to terminal/files/editor via Cloudflare tunnel)
• Telegram, Discord, Slack (mirror terminal I/O + remote control)
• GitHub, Linear unified Hub; SSH remote-project terminals; built-in Chromium browser + AI toolbar
• Generic HTTP proxy (CORS-free) for integrations

Media & productivity

• Screen capture (region select), clipboard image paste into terminal, native video editor
• Voice input (Whisper), OS notifications, completion sounds, 8 UI languages
• Auto-updater (Tauri), code-signed builds

🏗️ How it's built

Native and fast — small binary, low memory, real OS integration. Auto-updates via the Tauri updater; builds are code-signed (macOS Developer ID, Windows code-signed).

🚀 Quick start

1. Download & install for your platform (links above), launch, and sign in with GitHub (free).
2. Open a project — from local disk, over SSH, or open a server directly. All projects sit on the left; jump between them with a shortcut.
3. Start a terminal and run any AI CLI you have installed — or just use it as a great native terminal.
4. Talk to ⚡ BBARIT Agent — the built-in coding agent. Give it a task in plain language; it reads, edits, runs your tests, and won't call the job done until they pass — with a rollback point for every change.
5. Press 🦌 Broker Agent to spin up a Developer + Reviewer pair. Give the developer a goal, and watch the pair build, test, review, and queue a merge for your approval.
6. Open the side panels as you go — git, code search, notes/wiki, Kanban, and the built-in browser.

💻 Requirements

📥 Install

1. Download the latest build:
   • 🍎 macOS (Apple Silicon) → github.com/bbarit/terminal/releases/latest
   • 🪟 Windows 10/11 → github.com/bbarit/terminal-win/releases/latest
   • or grab either from bbarit.com
2. Launch and sign in with GitHub (free license).
3. Open a project folder and start a terminal — or press 🦌 Broker Agent to spin up an AI pair.

Auto-updates are built in (Tauri updater) — you'll always get the latest.

❓ FAQ

• What is ⚡ BBARIT Agent? BBARIT's own built-in coding agent (빠릿에이전트). It reads, edits, runs and verifies your code — it runs your tests or build, reads the failures, and fixes them before it's allowed to finish. Every file change is snapshotted for instant rollback.
• Which models can BBARIT Agent use? 37 providers and 1,000+ models — OpenAI, Anthropic, Google, Mistral, AWS Bedrock, Azure and more. Log in with your ChatGPT or Claude subscription, use API keys, or run Ollama fully local.
• Do I need an API key? No — for external CLIs you use the AI CLIs you already have installed; BBARIT runs them in real terminals. For BBARIT Agent, a subscription login or a local Ollama model works too.
• Which agents can the Broker Agent pair use? Any installed CLI — Claude, Codex, Gemini, Kimi, Qwen, Ollama, and more — and you can hot-swap either role mid-session.
• Will it touch my repo without asking? No — the Broker Agent works in an isolated git worktree and never merges or deploys without your approval. It only initializes git (with a notice) when needed for review.
• My project has no git — does review still work? Yes — the broker offers to initialize git so diff-based review works, and tells you before it does.
• Can I run more than one AI pair at once? Yes — the multi-harness manager runs several dev/review pairs in parallel, each worktree-isolated and individually monitored.
• Does it work over SSH / from my phone? Yes — SSH projects work like local ones, and the Web Terminal + Telegram/Discord/Slack mirroring let you drive sessions from a browser or phone.
• Is my code private with local models? Run Ollama for fully local, on-device inference — nothing leaves your machine.
• Is this the same as Octo Terminal? Yes — BBARIT Terminal is the evolution of Octo Terminal, rebuilt and renamed.

🆘 Support & community

• Issues: this repository's GitHub Issues
• Contact: [email protected]
• Web: bbarit.com

BBARIT Terminal (빠릿터미널) — the free AI vibe coding IDE, terminal & coding agent, built 100% in Rust. Maintained by Tenmiles Inc. · Built for developers who ship with AI.

🍎 bbarit/terminal · 🪟 bbarit/terminal-win · 🌐 bbarit.com · ▶️ Demo