Skip to main content
This quickstart gets a Comms agent ready to answer subscriber messages.

1. Create an agent

Open the dashboard and create an agent from Agents. Set:
  • Name: a customer-facing label, such as Example Concierge.
  • Instructions: what the agent is allowed to answer and when it should say it cannot verify something.
  • Knowledge: the agent context, policies, product facts, and preferred tone.
  • Channels: SMS, iMessage, or email.
Comms currently exposes one model in the product: osis-comms-1.

2. Attach a phone line

Agents can use a dedicated line or a shared line. For a shared line, set a claim word such as:
example-agent
When a subscriber texts that claim word to the shared number, Comms assigns that subscriber-line relationship to the agent. During testing, entering a claim word can switch the relationship to another agent.

3. Add knowledge

Use the knowledge page to write the agent context. Good context includes:
  • What the business does.
  • What the agent can answer directly.
  • What the agent must verify with tools.
  • What the agent must never disclose.
  • How the agent should phrase uncertain answers.

4. Connect MCP tools

If the agent needs private or live data, connect an MCP server from Sources and add it to the agent. For subscriber conversations, use subscriber-safe MCP tools. These tools should expose records for the verified texter or subscriber only, not broad admin data.

5. Test a reply

Use the agent playground for fast testing, then send an actual SMS or iMessage to the configured line. Example inbound text:
What reservations do I have?
If the agent has a phone-scoped booking MCP tool, Comms injects the verified sender phone number into the tool call and prevents user-typed email, name, or reference numbers from widening the lookup.

6. Go live

Once the playground response looks right, send a real message to the configured line and confirm the agent uses the right identity, knowledge, and tools.