VibeAround

VibeAround is a hub for coding agents that run on your own machine. From one desktop application or command-line daemon you pick an agent, a model profile or API endpoint, a workspace, a terminal and a session, then launch or resume it — Claude Code, Codex CLI, Gemini CLI, Pi, OpenCode, Cursor CLI, Kiro CLI, Qwen Code and the Claude and Codex desktop apps. It adds these capabilities around each agent's existing configuration and project permissions rather than rewriting them.

Provider settings are saved as profiles holding keys, base URLs, models, aliases and metadata, and a local bridge translates between the API shape an agent speaks and the one a provider offers: OpenAI Responses, OpenAI Chat Completions, Anthropic Messages and Gemini Generate Content. That lets the same agent run against different providers across sessions. Preset providers include DeepSeek, Alibaba DashScope, Moonshot/Kimi, MiniMax, Xiaomi MiMo, xAI/Grok, NVIDIA NIM, Z.AI/GLM, Google Gemini, OpenRouter, Azure OpenAI and custom endpoints, and a recorder captures the original request, the bridged request, both responses and the contents of search calls.

Where a provider offers no server-side web search, a host-side search runtime substitutes for it and feeds normalised results back through the bridge, drawing on Exa, Tavily, Grok/xAI and Brave Search. Local agents can also be exposed as OpenAI- and Anthropic-compatible endpoints for development and local testing, with a header selecting the working directory a request runs in.

A running session can be handed over with a pickup command and continued from a messaging channel — Feishu/Lark, Discord, Slack, Telegram, WeChat, DingTalk, WeCom or QQ Bot — each built as a plugin on a channel software development kit rather than wired into the core. A browser hub and a web terminal give the same control from desktop or mobile browsers, and dev servers or rendered Markdown can be previewed on their own loopback origin or shared through links gated by a six-digit code that expires after ten minutes.

Execution stays local: agents run on your computer, credentials stay in local settings, and the daemon listens on loopback unless a tunnel is enabled. Supported tunnel options are Localtunnel, Cloudflare Tunnel, ngrok and Tailscale Funnel; public owner surfaces require browser pairing on top of that.

Features

• Agent launch: one place to start CLI and desktop agents with a chosen profile, endpoint, workspace, terminal and session
• Session resume: continue previous sessions on macOS, Windows and Linux
• API profiles: saved provider keys, base URLs, models and aliases, usable per session rather than as one active configuration
• API bridge: translation between OpenAI Responses, OpenAI Chat Completions, Anthropic Messages and Gemini Generate Content
• Request inspection: the original and bridged requests, both responses and search tool contents recorded per launch
• Host-side web search: a local search runtime backed by Exa, Tavily, Grok/xAI and Brave Search, substituted for provider-native search
• Agents as APIs: local agents exposed on OpenAI- and Anthropic-compatible endpoints for development and testing
• Messaging handover: a pickup command moves a live session into any connected chat channel, with workspace and agent switching by command
• Web hub and terminal: browser launch, session continuation, chat and a real terminal from desktop or mobile
• Live preview: dev servers and Markdown opened locally or shared through ten-minute, code-gated links
• Tunnels on demand: Localtunnel, Cloudflare Tunnel, ngrok or Tailscale Funnel, with the daemon on loopback by default
• Local execution: repositories, credentials and terminals stay on your machine