Pattern Overview
This section contains documentation for various AI agent patterns that help solve common problems when building AI applications.
Available Patterns
Tool Budget
Controls tool usage with budget constraints to prevent expensive operations from exhausting resources.
Learn more about Tool Budget →
Embedded Explaining
Adds explanation requirements to tools for better observability and quality by requiring agents to justify their tool usage decisions.
Learn more about Embedded Explaining →
Tap Actions
Transforms opaque AI agent operations into transparent, real-time insights by intercepting, aggregating, and presenting human-readable summaries of agent activities.
Learn more about Tap Actions →
Agent Switch
Provides flat complexity scalability for categorical use cases by hiding the selection of specialized agents behind a single tool with categorical parameters.
Learn more about Agent Switch →
On-Demand Context Retrieval
Efficiently handles medium-sized documents by using a specialized agent to extract only relevant information, reducing token usage and improving performance in multi-turn conversations.
Learn more about On-Demand Context Retrieval →
Countdown Timer
Provides real-time awareness to LLMs by wrapping tools to include time information, enabling better task prioritization and quality vs. speed trade-offs.
Learn more about Countdown Timer →
Countdown Turns
Provides visibility into turn consumption, enabling agents to plan strategically and balance quality with resource constraints by making turn limits visible to the agent.
Learn more about Countdown Turns →
Surrender Task
Enables AI agents to fail gracefully when tasks become impossible or impractical, preventing resource waste and providing clear feedback to users and systems.