Slack Integration

Developer platform for Slack apps, interactive workflows, and automated handoffs in channels.

Start local setup ↓Check model fit

Beginner to intermediateSetup

Hosted platformLocal runtime

3Related paths

Recommended path

Slack local AI quick start

1. Create a Slack app and enable Socket ModeCreate the app, enable Socket Mode, generate an app-level token, and grant only the bot scopes needed for mentions and replies. Socket Mode avoids exposing a public request URL.
2. Subscribe to an event and start the bot processSubscribe to app_mention or the narrowest required event, install the app to the workspace, and run the Bolt or SDK process where it can reach the local model endpoint.
3. Acknowledge first, then generate asynchronouslyAcknowledge the Slack event inside three seconds, queue the model call, and post the answer afterward. Test mentions before adding slash commands or consequential tools.

Open official setup docs Check what fits your hardware

Best for

Teams routing AI summaries, approvals, and notifications into Slack with clear workspace permissions.

Model support

Slack has no local-model feature. It provides the surfaces, and your own app process holds the model connection: events arrive over the Events API with a public HTTPS request URL, or over Socket Mode, which uses a WebSocket and needs no public URL at all.

Choose models by role

Channel replies and summaries

Fast instruction model

Choose a model that answers quickly enough for useful chat, while still acknowledging Slack before generation. Smaller models are often a better operational fit for routine replies.

Tool-backed workspace actions

Tool-capable model with approvals

Use tool calling only when the bot must choose actions. Keep posting, record changes, and other consequential operations behind explicit human approval.

How to use this integration

Answer questions in a channel from an internal knowledge base without sending it to a vendor
Summarise threads with a model on your own hardware, posting results back asynchronously
Trigger local automation from a slash command and report the outcome in Slack
Keep regulated conversations inside your infrastructure while using Slack as the interface

Connecting a local model

Use Socket Mode when the model runs on your own network: the app opens an outbound WebSocket via apps.connections.open, so nothing inbound has to be exposed. Your handler then calls the local endpoint and posts the result back to Slack.

Tradeoffs

The three-second rule shapes the whole design: Slack expects an HTTP 2xx within three seconds of an event, which is faster than most local models finish generating. You must acknowledge first and post the answer asynchronously. Socket Mode, the option that keeps your network private, carries its own constraints: apps using it are not allowed in the public Slack Marketplace, it requires granular permissions, and it is capped at ten open connections.

Source

Slack Socket Mode docs