VibeSpace is a local-first coding agent that runs as a desktop app on your Mac. You authorize a workspace folder, describe the outcome you want in plain language, and VibeSpace reads code, breaks down the work, runs commands, and edits files to deliver finished results. Every step happens through visible, recoverable tool calls — you always know what it is doing, and you can approve, reject, or roll back at any time.
The core experience
- Authorize a workspace: pick a folder — VibeSpace reads, writes, and runs commands only inside that boundary.
- Describe the outcome: one sentence or a paragraph of plain language. No terminal skills or Git knowledge required.
- Watch it work: a live timeline shows every tool call (browse, search, read, edit, write, run command), and protected operations ask for your approval.
- Get the result: all changes land as a reviewable ChangeSet — inspect diffs and restore with one click.
Principles
- Local-first: tasks, sessions, and API keys stay on your Mac. Model requests go straight from your machine to the provider — VibeSpace runs no relay server.
- BYOK (bring your own key): eleven built-in providers; paste your own API key and pay the provider directly.
- Visible execution: every agent action is a record you can see, approve, and revert.
- Reliable before autonomous: start with conservative permission modes and hand over automation as trust grows.
- Artifacts, not answers: the goal is finished work, not explanations about how work could be done.
What VibeSpace can do
- Implement features, fix bugs, and refactor modules in an existing codebase;
- Run tests and scripts, then iterate on the output;
- Review code for quality and security, producing structured reports;
- Run routine tasks on a schedule (daily summaries, weekly checks);
- Split a vague requirement into subtasks handled by an expert team.
What VibeSpace does not do (yet)
- Windows or Linux (macOS only);
- Cloud accounts, cross-device sync, or team collaboration;
- MCP (Model Context Protocol) or a plugin marketplace;
- A remote package store or automatic module downloads;
- Background execution after the app quits (see Scheduled tasks).
UI tour
| Navigation item | Purpose |
|---|
| New Task | Start an agent task and watch its run timeline |
| Skills | Manage reusable skill packages |
| Kanban | Organize tasks and project progress on a board |
| Schedules | Create one-time / daily / weekly automated tasks |
| Experts | Browse built-in experts and expert groups |
| Projects | Organize workspaces and tasks by project |
| Modules | Manage feature modules (.vpack packages) |
| Settings | Providers, proxy, appearance, data, shortcuts |
Key concepts at a glance
| Concept | Meaning |
|---|
| Workspace | The folder you authorize; the boundary for all reads, writes, and commands |
| Project | An organizational unit for workspaces and tasks |
| Task | One complete conversation and unit of work, containing multiple runs |
| Run | One execution of the agent for a given input, with its timeline and approvals |
| Skill | A reusable "playbook" folder that instructs the agent |
| Expert | An agent role with a dedicated responsibility and tool restrictions |
| Expert group | A lead expert that decomposes, delegates, and integrates member work |
| ChangeSet | All file changes a task made; reviewable and restorable |
| Module | An extension package (.vpack) contributing navigation, UI, terminals, and more |
