Prompt Caching with Deep Agents
Deep Agents is a general-purpose agent that supports prompt caching across major model providers, reducing API costs by up to 80% through token reuse. It automatically sets explicit cache breakpoints, uses provider-side implicit caching, and structures prompts to maximize cache efficiency, enabling cross-provider cost savings. However, provider-specific features like explicit breakpoints and TTL vary, affecting overall savings. The harness minimizes cache degradation by maintaining consistent prefixes, and its effectiveness depends on conversation length and provider-specific features.
More from Langchain
Why managed agents are the next big thing in agent building
Managed Deep Agents gives developers a managed way to build, run, and deploy Deep Agents with built-in runtime, streaming, sandboxes, evals, memory, and auth.
langchain.com··Mads Kristian Nylund·AI InfrastructureAI-agents
LangSmith BYOC on AWS is generally available
LangSmith Bring Your Own Cloud is now generally available on AWS, giving Enterprise teams managed observability, evaluation, and deployment inside their own VPC.
langchain.com··Mads Kristian Nylund·AI DevelopmentAI InfrastructureAI Platforms
What is an AI agent?
Learn what AI agents are, how they work in an LLM loop, and where workflows fit so you can build reliable, production-ready autonomous systems.
langchain.com··Mads Kristian Nylund·AI DevelopmentAI InfrastructureAI-agents