Function: plan()

@providerprotocol/agents

---

@providerprotocol/agents / execution / plan

Function: plan()

plan(options): ExecutionStrategy

Defined in: src/execution/plan.ts:102

Creates a plan-then-execute strategy for agent execution.

The plan strategy implements a two-phase approach:

1. Planning Phase: LLM generates a structured plan with steps and dependencies
2. Execution Phase: Execute each plan step in topological order respecting dependencies

If a step fails and allowReplan is true, the strategy can generate a new plan to recover from the failure (replanning is not yet fully implemented).

This strategy is ideal for complex multi-step tasks where the execution order matters and steps may have dependencies on each other.

Parameters

options

PlanOptions = {}

Configuration options for the plan strategy

Returns

ExecutionStrategy

An ExecutionStrategy instance implementing the plan-then-execute pattern

Example

import { createAgent, plan } from '@providerprotocol/agents';

// Basic plan-based agent
const agent = createAgent({
  llm: myLLM,
  tools: [readFileTool, writeFileTool, searchTool],
  strategy: plan(),
});

// With step limit and replanning
const robustAgent = createAgent({
  llm: myLLM,
  tools: [apiTool, dbTool],
  strategy: plan({
    maxPlanSteps: 10,
    allowReplan: true,
  }),
});

// Execute a complex task
const result = await agent.generate(
  'Read config.json, update the version, and write it back'
);
console.log(result.state.plan); // Shows executed plan steps

See

• loop for simpler tool-loop execution
• react for reasoning-enhanced execution