One workspace for the whole AI development loop.

One conversation surface for ChatGPT/Codex, Claude, GitHub Copilot, Cursor, Google AI, Grok, and API keys.

Match the mode to execution, prototyping, approval, or read-only exploration.

Connect several accounts per provider; ChatGPT and Claude subscriptions can rotate automatically when one hits its limit.

Pick account, model, reasoning effort, and Fast when supported — and switch mid-conversation without losing context.

Compact older turns, in the background if you like, while keeping the working thread.

Start a conversation without a project, with private files and isolated permissions, and send its plan to a project later.

An orchestrator delegates to subagent profiles you define, including across providers.

Run independent subagent tasks at the same time in Claude and Codex conversations.

Choose provider, model, effort, and instructions for reusable roles, and address them with # in the conversation.

Approve a plan in the Plan tab and hand it to a new conversation or an isolated worktree.

Agents read, message, and report back to other conversations, with your confirmation the first time and a limit on agent-to-agent turns.

Parallel read-only reviewers and a verifier check the changes or the plan; you decide on each finding before it reaches the agent.

Terminal, browser, notes, memory, conversation history, debugger, and artifacts — on by default, each with its own switch.

A web_search tool in every runtime, through Tavily, Brave Search, your ChatGPT subscription, or DuckDuckGo.

Open VS Code, terminals, and an embedded browser with history suggestions and tab shortcuts beside the conversation, in reorderable tabs kept per conversation.

Read, write, edit, bash, glob, and grep inside the authorized workspace.

Connect HTTP or stdio MCP servers, sign in to OAuth servers with one or several accounts; connection details are encrypted at rest.

Install Codex, Claude Code, and Agent Plugins plugins from Git. Their skills, MCP servers, agents, and hooks work the same with every model, after you approve what they run.

Ask for approval, Approve for me, or Full access — chosen per conversation.

Search GitHub, clone, or create a repository from a conversation and start working in it. New repositories are private by default.

Give parallel jobs separate Git worktrees and branches.

Work directly in the active checkout when isolation is unnecessary.

Send tasks to new conversations without mixing files or conversation state.

Inspect changes, pull requests, and validation status before delivery.

Package repeatable guidance at the right scope.

A memory collection shared across standalone chats and providers, with search, editing, and export.

Relevant project memories are recalled for each message; automatic saving is opt-in.

Keep durable project notes beside the work.

Agents can search and read their own conversation history.

Charts, calculators, comparisons, and small tools the agent builds right in its answer, in an isolated frame.

Add images, PDFs, and text or code files by attaching, pasting, or dropping them.

Dictate messages, transcribed on your computer with whisper.cpp or, with your own key, by OpenAI or Groq.

Use vision when the selected provider and model support it.

Review screenshots and generated images from tools in the conversation.

Generate visual assets with your ChatGPT or Google AI (Gemini) subscription, an OpenAI API key, or inside bot environments.

Each bot gets its own Chromium browser with its own profile and site logins, plus terminals, files, and programs you can watch or take control of. Videos it shares play in the conversation, and the Files drawer lets you browse, download, or clean up its files.

Install the bot server on this computer's Docker or an Ubuntu or Debian VPS over SSH, and pair several servers at once; bots keep working while your computer is off.

Group up to eight bots that share a desktop, home folder, model accounts, skills, and MCP servers, and rotate between Claude or Codex accounts at their limits. Restart one bot without interrupting the others, or move a bot to another environment with its conversation, memory, files, and browser profile.

Schedule routines whose runs keep conversations of their own, let selected bots talk in threads apart from yours, and let a bot split work into parallel tasks or delegate it across models by difficulty.

Each computer, or a workspaces environment of the server, can let a bot start and follow development conversations in chosen projects, under an approval ceiling.

Run your own chats and workspaces on the bot server — one environment per scope of your work, each conversation with its own Linux computer — or connect another computer you own. Move conversations between this computer and the server.

Answer a bot's questions and approval requests in one place, with system notifications, alert sounds, and a mark on the Bots tab when it finishes or needs you.

Follow and direct bots and server chats from iOS or Android: read as they stream, open and control a bot's screen when it needs help with a sign-in or CAPTCHA, inspect routine runs and bot-to-bot threads, manage shared files, review and edit plans, approve requests, create bots, check subscription usage, dictate, and get notified when they need you.

Agents publish prototypes, reports, and dashboards; every change becomes a new version.

Browse every artifact with a preview, its versions, and the conversation it came from.

Personal links, approved access requests, or a link with an optional access code and expiry — revocable at any time.

People comment on a passage or a spot of the page; agents can read and answer them.

Pages such as forms collect one response from each person who opens them; export them as CSV or JSON, or let an agent read them.

See input, output, and available cache metrics.

Estimate API-equivalent cost where pricing data exists.

Break activity down by conversation, provider, and model.

Compare usage distribution across connected backends.

Show subscription limits when the provider exposes them.

Available for macOS, Windows, and Linux.

Keep files, notes, tools, and task state within reach.

Use shortcuts for frequent navigation and actions.

Move standalone chats, workspace agents, and bot conversations into their own windows without losing drafts or live responses.

Install updates from GitHub Releases inside the app on macOS, Windows, and AppImage, and use the interface in English or Portuguese.