Realtime Gateway

Runs Voice Mode's realtime sessions through a Vercel AI Gateway key, so voice bills to one gateway account and the bb host never needs an OpenAI key. Voice Mode asks installed connection plugins for a session; this one mints a single-use client secret from the gateway and returns its realtime URL with the secret already attached to the WebSocket handshake. Pick the model (gpt-realtime-mini by default), set an optional team scope, and run bb realtime-gateway test to confirm the handshake before relying on it. Needs a Vercel AI Gateway key with usage billing; no OpenAI key on this host.

Overview

View source

Realtime Gateway lets the Voice Mode plugin in bb run its voice sessions through a Vercel AI Gateway key. Your sessions bill to that gateway account, and the bb host never needs an OpenAI key of its own.

What you get

Voice sessions in Voice Mode billed through the Vercel AI Gateway.
One gateway key on the bb host, instead of a provider key per vendor.
A live bb realtime-gateway test probe that mints a secret and opens one real session, so you can confirm the handshake before relying on it.
A model setting you control: openai/gpt-realtime-mini by default, or any other gateway realtime id such as openai/gpt-realtime-2.

How it works

Voice Mode discovers installed plugins that publish a realtimeConnection rpc and asks one for a session. Realtime Gateway answers by minting a single-use client secret from the gateway's client-secrets endpoint, then returns the gateway realtime URL with that secret attached to the WebSocket handshake.

The gateway's realtime endpoint rejects an Authorization header, so the credential travels in the Sec-WebSocket-Protocol header instead, alongside the protocol marker the gateway expects. Secrets are created server-side and spent by the session, so they never reach the browser. A fresh secret is minted per session, which makes Voice Mode's refresh request a no-op.

Pair the plugin with Voice Mode's Connection setting on auto, or set it to plugin to pin this provider. On plugin, Voice Mode reports an error when this plugin cannot mint a secret rather than falling back to another provider without telling you.

Requirements

bb 0.46 or newer.
A Vercel account with AI Gateway access and usage billing enabled.
A gateway API key, entered in Settings → Plugins → Realtime Gateway or through bb plugin config realtime-gateway. The AI_GATEWAY_API_KEY environment variable works as a fallback.
The Voice Mode plugin, installed from https://github.com/steveafrost/bb-plugin-voice-mode

No OpenAI key is needed on the bb host.

Limits

This is one provider path. Voice Mode's own OpenAI key path still works and is unaffected.

The plugin depends on Vercel's realtime protocol as it stands today, version 0.0.1, because the gateway authenticates the handshake through the subprotocol header. If Vercel changes that protocol, sessions fail until the plugin is updated.

More from Steve Frost

Codex ThemeBrings the OpenAI Codex / ChatGPT desktop app's look to bb: system-font typography at 430 weight, the Codex surface and menu geometry, its send arrow, and a roomier left rail. Pick Catppuccin or BB's own colors, and turn on optional changes in the plugin settings. Light and dark.New Letta AgentsResume your local Letta agents from the sidebar without losing their saved conversations. See agent activity and dispatched tasks beside each agent. Keep agent providers out of the model picker unless you enable them. Requires a separate local Letta Docker installation with the supported filesystem layout, Docker access, and the BB ACP adapter; the adapter uses npx to install its pinned Letta ACP package.New PulseSee what needs you from Slack, Gmail and Calendar on one live board, kept current by scheduled agent runs. Hourly, Daily and Weekly runs write items sorted into Now, Next, Later and Nice to Know, and Done or Snooze sticks across runs. Each item can carry prepared replies that copy to your clipboard and open the source; nothing is sent for you. A robot button hands an item to an agent thread that drafts the work for your review. Your agents need their own Slack, Gmail and Calendar access (MCP servers, CLIs or plugins), and every run uses your model quota.25 Rad BinderTurn bb into an 80s/90s school-binder cover, with neon confetti, striped edges and chunky toy-commercial shadows. The sidebar becomes a deep-purple binder spine where the open thread is a hot-pink sticker. The composer, your messages and inline cards get thick outlines and hard offset shadows, and headings switch to bold italic capitals. Works in light and dark, makes no network requests, and uses Futura and Avenir Next from macOS, falling back to your system sans elsewhere.New

More in Agents & Providers 79

Add, choose, configure, route, or coordinate who runs a thread.

View all

Browser AutomationLet BB control the in-app browser and browsers on connected machines, with inline previews for headless sessions.BB5.8K

Bots SidebarKeep specialist bots, their memory, and their conversations together across projects. Organize chats in activity-aware trees, complete finished work, and move a thread between bots from its header. Route new project conversations to an owner and coordinate bots with @ mentions or attributed messages. Uses your configured BB providers with no additional service or account. Project-folder creation requires POSIX paths.

Tobias Lütke388

HandoffMoves a whole session between agents and machines in both directions: hands a bb thread's rendered transcript to another provider (optionally on another enrolled machine, carrying uncommitted work as a patch), or adopts a Claude Code, Codex, Gemini CLI or OpenCode session from your terminal as a bb thread in the same directory.

Vedran Burojević355