Bots and personas
Beyond the attendance bot you can run additional bots, each with its own chat identity, its own rooms, and its own training text. They answer only when addressed, in their own voice.
A bot is an identity plus instructions
Each bot is a separate account on your chat server with its own access token and its own encryption key store, so compromising one never exposes another. You assign it to rooms from the dashboard; it joins and listens only there. Its behaviour is a system prompt: the training text you write. Empty text disables auto-reply for that bot.
Persona presets
Pick a preset to start from, then edit the text. Presets ship with a role and rules; every one of them is instructed not to respond unless asked a question or @-mentioned.
- A tone modifier is appended to any persona: neutral, formal, casual, friendly, concise, direct, encouraging, playful, executive, mentor or Socratic.
| Technical | debugging, architecture and review help; covers trade-offs and the failure modes to check |
|---|---|
| DevOps | infrastructure, deploys, on-call; smallest action that stops the bleeding first |
| Security | never echoes secrets; treats authentication questions as possible phishing |
| HR | benefits, leave, conduct and policy; routes sensitive matters to a human |
| Finance | expenses, reimbursements, budget; flags anything outside policy |
| Office | supplies, facilities, visitors |
| Onboarding | walks new people through accounts and expectations in small steps |
| PM | scope, dependencies, risks; surfaces the trade-off when asked to add scope |
| Sales and Customer Success | prospect and customer questions, with hand-offs |
| Standup Coach | asks one targeted question when an update is vague |
| General and Custom | a helpful generalist, or a blank prompt for your own |
Trained on your own material
The training text is free-form. Paste your escalation paths, the summary of a runbook, the policy lines people ask about, the names of the systems you run. That text, the recent thread in the room, and, for cloud questions from someone with a connected cloud account, that person’s own infrastructure context are what the bot answers from. It does not browse the public internet and it has no access to your files.
How a reply happens
- Triggers: the bot’s chat ID, its name with an @, or its name followed by a colon at the start of a message. The mention is stripped so the model sees only the question.
- The reply is posted from the bot’s own identity and recorded in the audit log as a public reply with the bot named.
- In a room with several bots, an intent router picks the single best-fit bot for a message and picks nobody for small talk or below 0.45 confidence, so one question never gets three overlapping answers.
- Persona replies go through the same failover chain as everything else.
Triggers without code
The Bot Behavior page has a Triggers tab with four kinds of admin-defined behaviour, each with an enable switch, a priority and a fire count, and a live tester that runs a message through the rules without sending anything.
| Keyword reply | match keywords, send a canned direct message |
|---|---|
| Routing | match keywords, forward the message to a specialist person or room |
| Rule | event-driven, for example N strikes notifies admins |
| Slash command | !cmd runs a defined action |
Where personas already work for you
- The support persona triages new tickets and either resolves them or asks one question.
- The on-call persona sends hand-off recaps and escalations.
- The audit log records which persona sent every message, so a person with several bots in their direct messages can always tell them apart.