Currently when someone replies to broadcasts that you send, you can see those replies in our product in the Threads Inbox and reply back to individuals there.

Increasingly we have seen people want to automate some of this process, for example to be able to use external AI tools to generate a potential reply.

In order to facilitate this, we have a few things you can leverage:

• Webhooks to receive a notification when there's an incoming message

• an API endpoint to Send a Message to a phone number

With these, you can build a system (outside of Switchboard) to manage replies and automate sending of replies back however you best see fit for your program.

Example Prototypes

To serve as concrete examples for this, we have published two open-source, AI-assisted SMS reply-desk prototypes.

In these prototypes, we have an LLM that drafts a reply back to the contact. You can also update this however you like (potentially to include canned responses or scripts to give context to the LLM).

And as an example for how you can modify this process on your own, we've added an approval layer -- the LLM sends the draft to Slack, and then you can approve it, edit, or ignore it in Slack. Nothing sends automatically — each draft needs a human to sign off.

NOTE: These are prototypes, not an officially supported Switchboard product. They're meant to be forked and customized and deployed on your infrastructure, not run as-is. We're interested in feedback and can help if you have questions, but can't deploy this for you.

To get started you'll need:

• A Switchboard account with API credentials (see here for instructions)

• An API key from an LLM provider (Anthropic, OpenAI, etc.)

• A Slack workspace where you can install a bot

Choose between one of the two prototype options

They both works the same — you can pick based on your team's tech stack and where you want to host it. The Python repo has instructions for deploying to Render; the JS repo has instructions for deploying to Vercel or Cloudflare Workers. If you have questions or feedback about the installation instructions, please reach out to developers@oneswitchboard.com.

Python

sb-inbox-replier-prototype-python — the main branch covers both running it locally and deploying to Render for a stable cloud URL.

JavaScript

sb-inbox-replier-prototype-js — the main branch is local-only. Deploying to the cloud instead (Vercel or Cloudflare Workers) lives on its own branch, deploy/vercel or deploy/cloudflare.

Fork whichever repo you want and follow its README — it walks through setup, Slack configuration, and customizing the bot's persona and reply rules.