HappyVibe
Good vibes, real code.
Everything set up. Nothing happens behind your back.
v0.1.0 · macOS on Apple Silicon · Linux also as .deb · All downloads
A real session, recorded as it ran. Nothing staged.
A desktop app for coding with AI, where everything is already set up and you can see everything it does.
Everything is already here
Terminals, a browser, documents, git, schedules, sub-agents, MCP, skills — wired and ready. No plugins to find, no config to write, no restart. Open it and start.
Any model, including your own
29 providers, over a thousand models, six one-click sign-ins, and local models detected automatically. Your keys stay on your machine. Nothing leaves it that you did not send.
Nothing happens behind your back
You see every command before it runs, in plain language. Nothing loads itself — no skill, no extension, no plugin — until you say yes to that specific thing. And the description you approve is written by the app, never by the AI.
Reading the MDN page on pointer events to fix the drag
The agent wants to run something
Web: developer.mozilla.org
details
One call, two voices. The feed carries the agent's own sentence; the prompt carries the app's.
What's in it
Everything below is in the app today, grouped the way the app itself groups it.
- Workspaces — a project folder on your machine, with its own settings
- Sessions — run several at once; they sleep and wake where you left them
- Panes — up to four, holding chat, files, terminals and browsers
- Windows — tear any tab out into its own
- Chat — streams as it works, with tool cards and real diffs
- Editor — a file tree and a real editor beside the conversation
- Your terminal — your own shell, in a tab
- Git — human verbs on top, real git underneath
- Context — a live gauge you can open and prune
- Rewind — the conversation, the files, or both
- Scheduled runs — at the times you set
- Voice — on your machine, hold a key and talk
- AI autofill — the calls the app makes for you, each one switchable
- Keyboard shortcuts — yours to rebind
- System prompt — read the real one, add your own
- Permissions — rules by tool, path or command
- Agent tools — the full list, and whether each one is allowed
- Stats — sessions, models and cost, day by day
- Audit log — what was asked, and what you decided
- Every billed call — itemised, newest first
- Plan mode — it drafts a plan without being able to change anything
- Ask user — it stops mid-turn to ask you rather than guess
- Agent terminal — commands run where you can watch, type and stop
- Agent browser — it opens, reads and clicks a page; localhost is free, every other site asks
- Web — search, and read a public page as clean text
- Memory — durable facts about you and each project; saving and forgetting ask first
- Schedules — it can propose a recurring run, never create one
- Documents — Word, PowerPoint, Excel, PDF, OpenDocument, RTF and EPUB, read on your machine
- Plugins — a store, or bring your own marketplace
- Skills — review one before it goes live
- Prompts — slash commands, including the ones you already wrote
- MCP — a curated catalog, sign-in handled for you
- Agents — read or edit what each one is told, and watch it work
A few straight answers
- Why not just use the terminal?
- Use the terminal — it is in the app. The point is that the browser, the docs and the git history are in there too, and you can read what happened.
- Is it secure?
- It asks before it acts, and the app writes what you are approving rather than the AI. It is not a sandbox and does not claim to be.
- Who is it for?
- People who want to build things with AI and want to understand what is happening while they do.