Inferable uses a Re-Act agent to reason and act. At their core, ReAct agents follow a distinctive pattern: they alternate between reasoning about a situation and taking concrete actions. This isn’t just a simple back-and-forth—it’s a sophisticated approach that enables AI systems to:

  • Think deliberately about the current state
  • Plan appropriate actions
  • Execute those actions
  • Evaluate the results
  • Adjust their approach based on outcomes

Inferable’s default agent execution model is based on this reasoning and action pattern.

​
Iteration

When you run an agent, it will execute in a loop. At each iteration, the agent will take a series of steps which involves in searching for a tool based on the context, planning the next step, and outputting the required actions to take.

The next step of the Inferable Re-Act agent is either to:

  1. Execute one or more tools
  2. Present a final result

If the agent fails to reach either one of these conclusions, the agent will be prompted to think again, based on a set of heuristics.

Was this page helpful?

Suggest edits
Using Tools with AgentsAgentic Runs