Autonomous agents

Build agents that operate as durable processes, iterating a model decision loop over typed tasks until the work is done. The Autonomous Agent component is purpose-built for this. The developer declares the agent’s description, accepted task types, tools, and coordination capabilities, and the runtime takes care of execution, persistence, and recovery.

Overview

When to Use This Pattern

  • You need durable multi-step execution that survives crashes and restarts, without writing workflow steps.

  • You need agent-to-agent coordination (delegation, handoff, teams, moderation) without writing orchestration code.

  • The work produces typed task results that can outlive a single request, be queried later, handed off, or aggregated.

  • You want each agent to focus on a narrow context, with isolation between agents and shared context only where it helps.

  • Background processing, monitoring, or batch agents that run independently and produce typed output.

Akka Components Involved

  • Autonomous Agents run a model decision loop until tasks complete, with built-in coordination capabilities.

  • HTTP Endpoints trigger agents and expose task results to external clients.

  • Entities hold supporting state that agents read or write through ComponentClient or as function tools.

Sample Projects

The autonomous-agent-playground bundles a progression of demos covering every coordination capability:

  • helloworld: minimum viable autonomous agent.

  • pipeline: task dependencies enforce phase ordering.

  • docreview: task attachments for large content.

  • dynamic: same agent class configured per request.

  • research: delegation to specialist workers.

  • consulting: composed delegation and handoff.

  • support: handoff routing through a triage agent.

  • publishing: task dependencies plus a human-approval gate.

  • debate, negotiation, peerreview: moderation patterns.

  • devteam: collaborative team with self-coordinating members.

See the samples table on the Autonomous Agents hub page for a one-line description of each.

Alternatives

For autonomous-style processing built on request-based Agents combined with Timed Actions and Workflows, the following samples illustrate scheduled and pipeline patterns:

These demonstrate scheduling and external orchestration. For model-driven multi-step processes within a single component, prefer Autonomous Agents.

See Also