Skip to main content
Agents in Jarvis are purpose-built AI workers that handle tasks on your behalf. Each agent has a defined role, a set of tools it can call, and the ability to delegate sub-tasks to other agents — so complex work gets broken down automatically and routed to the right specialist.

​
Role hierarchy

Jarvis agents follow a four-tier role hierarchy that mirrors how engineering teams operate:
RoleResponsibility
CEOHigh-level task framing, goal setting, delegation to CTOs
CTOTechnical planning, breaking goals into engineering tasks
EngineerCode generation, debugging, infrastructure execution
WriterDocumentation, summaries, structured output
When you hand a task to an agent, it evaluates the work and either handles it directly or delegates down the hierarchy until the right specialist takes over.
You don’t need to target a specific role manually. Agents route tasks internally — submit your goal to Paperclip or Hermes and let the hierarchy handle delegation.

​
Available agents

Paperclip

General-purpose task orchestration. Start here for most goals — Paperclip routes work to the right agent automatically.

Hermes

Communication and coordination. Use Hermes when your task involves messaging, notifications, or cross-agent handoffs.

Claude Code

Software engineering tasks. Claude Code handles code generation, review, debugging, and technical analysis.

MCP tools

Explore the 526+ tools your agents can call, including ServiceNow integrations and custom Jarvis tools.

​
How agents delegate

When you submit a task, the receiving agent breaks it into sub-tasks and assigns each to the most capable agent in the hierarchy. For example:
  1. You ask Paperclip to “set up a monitoring dashboard and document it.”
  2. Paperclip (CTO role) plans the work and creates two sub-tasks.
  3. The Engineer sub-agent provisions the dashboard using MCP tools.
  4. The Writer sub-agent drafts the documentation.
  5. Paperclip assembles the results and returns them to you.
You see a single response, but multiple agents may have contributed to it.

​
Agents vs. direct model inference

Use agents when your task requires:
  • Multiple steps or decision points
  • Tool use (file writes, API calls, workflow triggers)
  • Cross-domain work (e.g., code + documentation)
  • Long-running or asynchronous execution
Use direct model inference when you need:
  • A fast, single-turn response
  • Simple text generation or summarization
  • Low-latency completion without tool overhead
See model inference for direct inference options.

​
How to interact with agents

You can reach agents through the REST API, n8n workflows, or the Jarvis UI. 
curl -X POST https://your-jarvis-host/agents/invoke \
  -H "Authorization: Bearer your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "agent": "paperclip",
    "task": "Summarize the last 10 deployment logs and flag any errors."
  }'
For a complete list of agent API endpoints, see the agents API reference.