// the find
avibe-bot/avibe
The local-first Agent OS — your AI partner lives on your own machine. Drive the official Claude Code, Codex & OpenCode from your browser or any chat app.
Avibe is a local Python service that runs the official Claude Code, Codex, and OpenCode CLIs on your own machine and lets you drive them from a browser or from Slack, Discord, Telegram, WeChat, or Lark. It is aimed at developers who already pay for agent subscriptions and want those agents reachable from a phone without moving code or credentials to someone else's server.
- It wraps the vendor CLIs instead of reimplementing an agent loop, so the behaviour you get is the one you would get in a terminal. The integration surface is thin, which makes it easier to reason about than a custom orchestrator.
- Chat connectors are outbound only (Slack Socket Mode, Discord Gateway, Telegram long polling, Lark WebSocket), so the machine needs no open inbound ports for chat control, and access defaults to fail-closed.
- Model Hub only fails over before the first token arrives, so a single response never gets stitched together from two providers. Per-source quota tracking and usage priced at API rates is a useful way to see what a subscription is actually returning.
- The repo shows real release discipline: install-integrity and atomic file IO modules, a Tauri runtime host with updater signature fixtures and tests, and release-guard workflows for the Model Hub engine and git runtime.
- Routing consumer Claude and ChatGPT subscriptions through a gateway to other clients may conflict with vendor terms. The README warns at sign-in, but an account that depends on this could still be restricted. Read the vendor terms before pointing a subscription at it.
- 'Local-first' holds for code and execution, not for the remote path. Sign-in, the Cloudflare tunnel behind vibe remote, and voice transcription all go through avibe.bot. The one-line install is curl piped to bash from that same domain. The script is in the repo, which helps, but you are still trusting the domain.
- Windows is handled through WSL according to docs/WINDOWS_WSL.md, and the agent runtime depends on tmux (core/tmux_runtime.py plus a build-tmux workflow). A native Windows desktop build is still in testing, so the runtime is effectively Unix-first.
- The scope is very wide: three agent CLIs, five chat platforms plus a browser Workbench, Model Hub, Vault, the Harness, Show Pages, and a Tauri desktop app. The README itself marks Vault custody as pre-launch and the desktop app as in testing, so parts of it are unfinished in ways that matter for anything holding credentials.