Team chat
Message an agent in a direct message or mention it in a channel, and it replies in the thread.
- Slack
- Microsoft Teams
- Google Chat
- Webex
- Mattermost
- Rocket.Chat
- Zulip
Agents
Always-on agents, wherever your people already are.
Build an agent once and put it to work in Slack, Teams, WhatsApp, email, your support desk or on the phone. It acts as a named service account or as the verified person asking, inside the rules your team sets.
Customer · WhatsAppMy order #4471 arrived damaged. Can I get a refund?
Support agentSorry about that. Order #4471 qualifies for an $84.00 refund under your policy. Refunds need a lead’s approval, so I’ve asked one now.
Support leads · SlackApprove an $84.00 refund for order #4471?
Support agentApproved and issued. You’ll see it on your card in a few days.
Acted as: support service account · every step audited
Any surface
70+ apps can start a conversation with an agent.
The same agent can answer in several places at once. Each surface is a trigger you turn on, with no glue code.
Messaging and SMS
Customers and staff reach the same agent from the messaging apps they already use.
- Telegram
- SMS via Twilio
- Vonage
- MessageBird
- Sinch
An agent answers a mailbox: it reads the thread, does the work and replies as the mailbox.
- Gmail
- Outlook
- Postmark
- SendGrid
- Mailgun
- Amazon SES
Support desks
Agents pick up tickets and conversations where your support team already works.
- Zendesk
- Intercom
- Freshdesk
- Front
- Help Scout
- Gorgias
Phone calls
Give an agent a phone number. It listens, speaks its replies and lets callers interrupt it mid-sentence.
- Twilio Voice
Engineering tools
Agents respond to issues, merge requests and incidents in the tools engineers live in.
- GitHub
- GitLab
- Jira
- Linear
- PagerDuty
- Opsgenie
Beyond chat
Agents also start on a schedule, from a workflow, or as an API.
On a schedule
Run an agent every morning, every hour or at month end, unattended, with its own model choice per schedule.
From a workflow
A workflow can start an agent, send it a message and wait for its answer before moving on.
Over an API
Publish an agent as an endpoint in one of your APIs, with its own keys, quotas and rate limits.
From another AI assistant
Expose an agent as a tool on your own MCP server, so assistants your team already uses can call it.
Identity and approvals
Every agent knows who it’s acting for.
A named service account
By default an agent acts as the service account of the surface it answers on, with only the access you grant it.
Or as the person asking
Switch an agent to act as the verified person messaging it, so it can only see and do what that person can.
Approvals in the conversation
When an action needs a yes, the agent asks right in the chat: a direct message on Slack, a private chat on Telegram, a verified SMS, or a keypad press on a call.
Safe when nobody’s watching
In group channels, shared inboxes and support desks, an agent takes only the actions you allow without asking anyone.
What an agent is made of
Described in plain language. Published like software.
Persona and instructions
Write who the agent is and how it works, in plain language.
Tools
Give it your workflows, any connector action, platform utilities and actions on your own data. Each tool’s inputs come from the tool itself.
Skills and memory
Bind reusable skills, and choose whether it remembers what it learns in conversations inside RocketKind.
Model and limits
Pick the model, the guardrails and the budget it works within.
Versions
Publish a version, test it, and set it live. Roll back in one step. Every publish is in the audit log.
People when needed
Hand a decision to a person through an inbox, a signed link, email or a button in Slack, Teams, Telegram or Discord, with deadlines, reminders and escalation.