ATOM Documentation

← Back to App

Intent Routing: CHAT vs WORKFLOW vs TASK

The ATOM Meta-Agent is an intelligent orchestrator that determines the best execution path for every user request. Instead of just "chatting," the system uses an Intent Classifier to route tasks to one of three specialized paradigms.

The 3 Execution Paradigms

1. CHAT (Simple Response)

  • Best for: General questions, summarization, and quick facts.
  • Execution: Handled directly by the LLM Service with immediate context.
  • Output: A text-based response without persistent artifacts or external tool use.
  • Example: "What is the capital of France?" or "Summarize this email chain."

2. WORKFLOW (Blueprint Execution)

  • Best for: Repetitive, structured business processes.
  • Execution: Routed to the QueenAgent, which loads a predefined "Blueprint" or "Skill Chain."
  • Output: A sequential execution of specific tools (e.g., fetch Jira ticket → update Slack → create calendar invite).
  • Example: "Run the monthly payroll workflow" or "Sync these leads to Salesforce."

3. TASK (Unstructured Problem Solving)

  • Best for: Complex, multi-agent objectives that require planning and tool choice.
  • Execution: Handled by the FleetAdmiral. The system recruits a fleet of specialized agents (Search, Data, Code) to work together.
  • Output: Includes Reasoning Traces, multiple sub-steps, and generated Artifacts on the Universal Canvas.
  • Example: "Research the top 10 competitors in the AI space and generate a SWOT analysis report."

How It Works for Tenants

As a user, you don't need to specify the mode. Simply type your request into the Global Chat Widget or the Agent Studio, and the Meta-Agent will:

  1. Analyze Intent: "Is this a request for info, a known process, or a new project?"
  2. Context Resolution: "Which agents and data sources are relevant to this tenant?"
  3. Governance Check: "Is the user's requested agent at the correct Maturity Level to execute this?"
  4. Execution & Feedback: For WORKFLOW and TASK modes, you will see a "Reasoning Trace" showing the agent's thought process and any required approvals.

Advanced: Hybrid Execution

In some cases, the Meta-Agent will propose a Hybrid Execution.

  • It may start in CHAT mode to clarify your request.
  • Once the objective is clear, it will pivot to TASK mode to recruit a fleet.
  • Finally, it may suggest creating a new WORKFLOW blueprint for future use, essentially "teaching" the agent a new skill.

[!TIP] You can manually force a routing mode by using "slash commands" in the chat:

  • /chat: Force simple LLM response.
  • /workflow [name]: Explicitly trigger a blueprint.
  • /task: Force multi-agent fleet recruitment.