Agent( )

Pattern

PrepareStep: Trim Message History

Demonstrate ToolLoopAgent prepareStep for trimming long message history. Pre-seeded 24-message conversation shows context compression on the first run with per-step trim telemetry.

IntermediateJun 2026View the full pattern on desktop
PrepareStep: Trim Message History preview

Problems solved

  • Trim long message histories before each agent step
  • Keep recent tool results while dropping stale turns
  • Observe prepareStep compression with per-step telemetry
  • Prevent unbounded context growth in ToolLoopAgent loops

Use cases

  • Long-running agents with many tool calls
  • Context-window budget management
  • Debugging what the model sees each step
  • Production agents that must cap history size

AI SDK APIs

ToolLoopAgentprepareStepstepCountIstoolgateway

Tags

aiagentsprepare-stepcontext-windowmessage-trimmingtool-loop-agentai-sdkagent-api

Requirements

Node.js
24+
Memory
512MB
Other
  • OpenAI API key

Environment variables

External services

openai

Ready to use?

View the full interactive pattern with live preview and copy the source code.

View the pattern on desktop

Related patterns

Loop Control: stopWhenDemonstrate ToolLoopAgent loop control with stopWhen. Covers stepCountIs, hasToolCall, combined conditions, and a custom StopCondition with step timeline visualization.Jun 2026
Tool Context: experimental_contextInject request-scoped context (user identity, database handles, session) into ToolLoopAgent tools via experimental_context, so the model never supplies sensitive identity in tool inputs.Jun 2026
Call Options: prepareCallDynamically configure one ToolLoopAgent per request with callOptionsSchema and prepareCall. A customer-support agent rewrites instructions, gates tools, tunes reasoning effort, and injects RAG context based on account tier and urgency.Jun 2026
Structured Agent Output: Output.choiceDemonstrate structured agent outputs using ToolLoopAgent with Output.choice. Constrain agent outputs to predefined choices for predictable, type-safe responses.Nov 2025