jazzenchen/VibeAround

VibeAround – an all‑in‑one hub for AI coding agents

What it is – VibeAround is a desktop‑first application (with a CLI package @vibearound/cli) that lets you launch, configure, and keep‑alive a variety of AI‑powered coding assistants (Claude Code, Codex CLI, Gemini CLI, Pi, OpenCode, Claude Desktop, Codex Desktop, etc.) without having to edit each tool’s own configuration files.

Key capabilities

Feature	How it works (as described in the README)
Unified agent launch	Choose an agent, model/profile, API endpoint, workspace, terminal and session from a single UI and start the agent (CLI or desktop) with one click.
API‑shape bridge	A local “VAAAB” bridge (va‑ai‑api‑bridge) translates between the agent’s native request format and the provider’s API (OpenAI, Anthropic, Gemini). This lets you use any supported provider with any supported agent.
Multiple providers per session	Save provider keys, base URLs, model aliases, etc. as profiles and run the same agent against different providers in parallel sessions.
Live request inspection	VibeAround records the original request, the bridged request, the raw response, the bridged response and any search‑tool payloads, so you can see exactly what is sent and received.
Session continuity	A /pickup command can be posted to any connected messaging platform (Discord, Slack, Feishu/Lark, Telegram, WeChat, etc.) to resume the exact same agent session from another device or chat client.
Web Hub & Web Terminal	A browser‑based dashboard lets you launch agents, switch profiles, and chat with them. The same dashboard can expose a terminal (via WebSocket) so you can drive a local CLI agent from a mobile or remote browser.
Remote tunnels	When you need a public URL, VibeAround can start a tunnel via Localtunnel, Cloudflare Tunnel, ngrok, or Tailscale Funnel; otherwise it stays bound to localhost only.
Live preview	Development servers, Markdown files, or static HTML can be turned into short‑lived share links that open in any browser or messaging app.
Host‑side web search	If a provider does not offer built‑in web search, VibeAround can invoke the va‑search‑tool runtime (supports Exa, Tavily, Grok/xAI, Brave) and inject normalized results back into the model.
Agent‑as‑API	Local agents can be exposed as OpenAI‑ or Anthropic‑compatible endpoints (/local-agent/{agent_id}/{profile_id}/v1/...) for quick integration testing without a cloud gateway.

Supported agents – Claude Code, Claude Desktop, Codex CLI/Desktop, Pi, Gemini CLI, OpenCode, Cursor CLI, Kiro CLI, Qwen Code (some in “coming soon” state). The app does not modify the agents’ own config files; it only adds a thin launch wrapper.

Supported providers – Built‑in presets for DeepSeek, Alibaba DashScope, Moonshot/Kimi, MiniMax, Xiaomi MiMo, xAI/Grok, NVIDIA NIM, Z.AI/GLM, Google Gemini, OpenRouter, Azure OpenAI, plus any custom endpoint that matches one of the four bridged API shapes.

Why you might use it

• You already have a favorite coding assistant but want to experiment with other model providers without reinstalling or re‑configuring the tool.
• You need to hop between a desktop IDE, a terminal, and a chat app while keeping the same AI session alive.
• You want to see exactly what is being sent to the provider (debugging, cost tracking, compliance).
• You prefer to keep all code execution on your own machine, only exposing a tunnel when you explicitly choose to.

Known limitation – The README notes a common error (Unable to connect to API (ConnectionRefused)) caused by leftover config entries from other tools like CC Switch; you must clean those fields before relaunching.

Roadmap hints – Usage‑and‑cost tracking is listed as “🚧 Roadmap”, and MCP/Skills management is not yet supported.

---

All information above is taken directly from the repository’s README; no additional features have been inferred.