Skip to content
Kernel Service

Creating Agents

An Agent is a persistent AI identity in Kernel Service. Unlike a one-shot API call, an agent has:

  • A persona (role, communication style)
  • Objectives (what it’s trying to achieve)
  • Memory (context from previous tasks)
  • Permissions (which tools it can use)

Creating an Agent

Terminal window
curl -X POST https://api.kernelservice.com/api/v1/agents \
  -H "Content-Type: application/json" \
  -H "X-Tenant-ID: $TENANT_ID" \
  -H "X-API-Key: $API_KEY" \
  -d '{
    "name": "Alex - Account Manager AI",
    "external_user_id": "user-42",
    "persona": {
      "role": "Account Manager",
      "communication_style": "professional, relationship-focused",
      "description": "Manages client relationships and identifies growth opportunities"
    },
    "objectives": [
      "Maintain regular contact with top 20 accounts",
      "Identify upsell opportunities",
      "Ensure client satisfaction above 90%"
    ],
    "allowed_tools": [
      "crm.accounts.get",
      "crm.contacts.get",
      "email.draft",
      "calendar.book"
    ]
  }'

Persona Design Tips

The persona directly influences how the LLM behaves:

  • role — defines the agent’s identity (“SDR”, “Account Manager”, “Support Agent”)
  • communication_style — tone and approach (“professional and warm”, “concise and data-driven”)
  • description — detailed context about the agent’s specialisation

Submitting Tasks to an Agent

Once created, submit tasks with the agent_id:

Terminal window
curl -X POST https://api.kernelservice.com/api/v1/tasks \
  -d '{"agent_id": "AGENT_ID", "objective": "Check in with our top 5 accounts"}'

The agent’s persona, objectives, and memory are automatically injected into the LLM context.