Agents collaborate asynchronously, escalate intelligently, and recover from failures — no need to build custom orchestration logic. Pick the architecture that fits your workflow. Switch between them at runtime.
Select an architecture
BUILT-IN ARCHITECTURES
Each architecture is a proven multi-agent pattern, built into the platform as a first-class runtime option. No libraries to install, no frameworks to learn. Select an architecture in the Dashboard or SDK and your agents adopt it immediately.
The agent reasons about the current situation, decides which tool to call, observes the result, then reasons again. Ideal for tasks that require real-time interaction with external systems where the next step depends on the last result.
A planner agent creates a full execution plan upfront. An executor agent works through each step sequentially. If a step fails, the planner revises. Best for structured, multi-step workflows where the full scope is known in advance.
The agent generates multiple candidate reasoning paths simultaneously, evaluates each path's promise, and pursues the most viable branches. Excels at problems requiring creative exploration where the optimal approach isn't obvious upfront.
The agent executes, evaluates its own output against success criteria, generates self-critique, and retries with improved reasoning. Each iteration incorporates lessons from previous attempts. Ideal for tasks with clear quality metrics.
Define agent workflows as directed acyclic graphs where independent branches execute in parallel and dependent steps wait for upstream completion. Maximum throughput for complex workflows with clear dependency structures.
Agents follow formally defined state transitions with guard conditions on every transition. Integrates with the Business State Machine engine to ensure agents cannot skip steps or invent invalid workflow paths. Maximum process control.
Multiple specialized agents divide work by expertise — one researches, one analyzes, one writes, one reviews. A supervisor agent coordinates handoffs. Each agent has its own tools, models, and authority level via AURA's OrgKernel.
A lightweight router agent classifies each incoming request and routes it to the appropriate tool or sub-agent. No complex reasoning chains — just fast, accurate dispatch. Best for high-volume scenarios where speed matters more than deep reasoning.
The agent accumulates context across interactions — conversation history, prior decisions, learned preferences, and organizational patterns. Each new action is informed by the full memory of past interactions. Ideal for long-running relationships and adaptive workflows.
Rewind any agent execution to any prior state and fork to explore alternative paths. Understand not just what happened, but what would have happened if the agent had made a different decision at any step.
A unified interface for calling local and remote agents interchangeably. Build agent workflows that span teams, departments, and deployment boundaries — without changing a single line of orchestration code.
Cross-namespace service calls with multi-region, multi-cloud routing. Enterprise-grade infrastructure for organizations running agent workloads across multiple environments, regions, and cloud providers.
SELECTION GUIDE
You don't need to commit upfront. Architectures are switchable at runtime — start with one, observe performance in the Dashboard, and switch if a different pattern fits better. Here's a starting point.
Use a tight reasoning-action loop where each step depends on the previous result. The agent reasons, acts, observes, and repeats.
Create the plan upfront, then execute each step sequentially. If a step fails, the planner revises without starting over.
Explore multiple reasoning paths in parallel, score each branch, and pursue the most promising approach.
Let the agent draft, self-critique, and refine iteratively. Each pass gets better based on explicit quality feedback.
Model the workflow as a DAG — independent branches execute in parallel, dependent steps wait for upstream completion.
Assign each agent a specialty — researcher, analyst, writer, reviewer — with a supervisor coordinating handoffs between them.