SpeedIQ

Automations

AI auto-reply

AI auto-reply drafts a reply to a lead’s most recent inbound message. It can queue the draft for a human to send, or send it automatically, with rate limits, business hours, and caps so it never runs away.

Set it up

  1. 1
    Open the settings
    Go to Automations → AI Auto-Reply.
  2. 2
    Enable and pick channels
    Turn on AI replies and choose which channels it may reply on — WhatsApp, SMS, Email, and social DMs (Instagram, Facebook, TikTok, LinkedIn).
  3. 3
    Choose a mode
    Draft only (a human sends) or Auto-send.
  4. 4
    Add the action
    Create a Message received automation and add the AI auto-reply action.

Models & providers

Replies run through the Vercel AI SDK, so switching models or providers is a settings change, not a code change. The default is Gemini 2.5 Flash (Google), fast and inexpensive. A bare id is treated as Anthropic; prefix to switch provider:

example model ids
google:gemini-2.5-flash    (Google, default)
google:gemini-2.5-pro      (Google, premium default)
claude-sonnet-5            (Anthropic)
openai:gpt-4o              (OpenAI)
gateway:openai/gpt-4o      (Vercel AI Gateway, any model)
gateway:xai/grok-3         (Vercel AI Gateway)

Each provider reads its own API key from the environment: GOOGLE_GENERATIVE_AI_API_KEY, ANTHROPIC_API_KEY, OPENAI_API_KEY, or AI_GATEWAY_API_KEY. With the defaults you only need the Google key.

Any model via the Gateway
Prefix a model with gateway: and a creator/model slug to route it through the Vercel AI Gateway, one key (AI_GATEWAY_API_KEY) reaches OpenAI, Anthropic, Google, xAI, Meta, and more without adding a provider SDK.

Guardrails

  • Rate limit, at most N AI replies per recipient per hour.
  • Daily cap, a per-project ceiling on AI replies per day.
  • Business hours, optionally only reply during set hours.
  • Fall back to human, the AI defers instead of guessing when it is not confident.
  • Kill switches, a per-project enable toggle and a platform-wide switch.

What the AI sees

The reply is grounded in the recent conversation and your tone and custom instructions. It is told not to invent facts (order numbers, prices, policies) and to keep within a length cap you set. You can override the whole behaviour with your own instructions.

Billing

An AI reply is billed twice: once for generating the draft, and, in auto-send mode, again for the outbound message on its channel. Draft-only mode only bills the generation. Every draft and send is recorded in the AI Auto-Reply log with its disposition.

Start in draft-only
Run in Draft only for a while and review the drafts in the log before switching to Auto-send.