Slack

Put a Pepe agent in your Slack workspace so people can talk to it in channels and direct messages.

Slack

Connecting Slack lets people talk to the agent right inside your workspace. Slack delivers messages to Pepe through its Events API.

Step by step

Create the app. At api.slack.com/apps → “Create New App” → pick “Blank app” (skip “AI agent” and the other templates - they come with Slack’s own agent scaffolding, which you don’t need here; Pepe is the agent). Pick your workspace.
Grant the bot scopes. Under OAuth & Permissions → Scopes → Bot Token Scopes, add chat:write, app_mentions:read, channels:history and im:history. Add files:read too if people will send pictures or files (without it the bot can’t open them), files:write so the bot can send files back, reactions:write so it can mark the message it is working on with an eyes reaction, and reactions:read so a 👍 or ❤️ on one of its messages counts as feedback. For that last one also subscribe to the reaction_added bot event. Add channels:read as well (and groups:read for private channels) so the Channels page can show each channel’s name instead of its id, and users:read so the people who write show by name when you pick who can train.
Install the app. Still on OAuth & Permissions, click “Install to Workspace” and copy the Bot User OAuth Token (xoxb-...).
Grab the signing secret. Under Basic Information → App Credentials, copy the Signing Secret.
Register the connection in Pepe:
```
pepe setup
```
Choose the channel option, pick Slack and the agent, and enter the credentials (a ${ENV_VAR} reference is accepted for any secret). Pepe prints the callback URL to register:
```
https://YOUR_HOST/webhooks/root/slack/<slug>
```
Swap YOUR_HOST for wherever your server actually answers (a real domain in production; mix pepe serve --tunnel if you’re just testing locally).
Turn on events in Slack. In the app → Event Subscriptions → Enable Events → paste the URL from the step above (Slack fires a url_verification handshake right away, and Pepe answers it on its own - no manual step needed). Under “Subscribe to bot events”, add message.channels, app_mention and message.im (the last one is what makes a direct message actually work).
Turn Socket Mode off. New Slack apps often ship with it on by default: sidebar → Socket Mode. While it’s on, Slack routes events over a WebSocket instead of hitting your Request URL - and Pepe only speaks the plain HTTP webhook, so nothing arrives, silently, even with everything else set up correctly. Leave it off.
Save. On the Event Subscriptions page, click “Save Changes” at the bottom before leaving - pasting the URL and navigating away without saving discards it.
If Slack asks, reinstall the app to the workspace (new scopes/events require it).
Test it: invite the bot to a channel (/invite @your-bot) and mention it, or send it a direct message.

Connection fields

A connection’s config holds:

bot_token: the bot user OAuth token (xoxb-...), used as the bearer for replies.
signing_secret: verifies the X-Slack-Signature on inbound requests.

Replies are posted with chat.postMessage.

See Webhooks for the fields every connection shares (agent, mode, trainers, session_ttl_min, ephemeral, commands) and how the generic route works under the hood.

Messages from other apps

A channel that gets its work from another system (a help desk posting each new ticket, a monitoring alert) receives messages written by an app, not by a person. Pepe ignores those by default, so the bot never answers itself or other bots. To answer one, list its bot id (B...) or app id (A...) in the connection’s Answer these bots and apps field (accept_bots, comma separated). Each message Pepe ignores is logged with its ids, so you can copy the right one from the log instead of guessing. A listed app is answered without being @mentioned, so you do not need /mention off for that channel: people writing there still have to mention the bot, and only the app you listed gets through on its own.

Integrations often put the whole message in the coloured-bar attachment instead of the text. Pepe reads the attachment’s title, body and fields as the message. The bot’s own messages are never answered, even if its id is on the list.

To keep such a channel listening without an @mention, send /mention off once in it. It stays on after /new and after a restart, until someone sends /mention on.

Switching models

The /model and /models commands let people check or change which AI model answers them. They work only on an admin-mode connection with commands enabled; on support, they are treated as plain text. /models lists the models available to this connection’s project; /model shows the current one, or changes it:

/model openrouter               # ask whether to switch just this chat or everyone
/model openrouter session       # switch for this conversation only
/model openrouter global        # switch for everyone this connection talks to

Anyone in an allowed conversation may switch the model for their own conversation. Switching it globally, for everyone this connection talks to, is reserved for trainers, the same trusted list that controls memory. Set model_switch_locked: true on the connection to turn model-switching off entirely for non-trainers.

← WhatsApp Discord →