The Composable AI Agent Infrastructure Layer.
AgentCore is a lightweight, type-safe TypeScript framework for building composable AI agents. Engineered with decoupled primitives for model providers, intelligent gateway routing, real-time streaming, runtime guardrails, and automated evaluation—giving developers complete control without framework lock-in or runtime bloat.
Supported Provider Ecosystem
Core Building Blocks for AI Agents
Primitives designed specifically for developer control, type safety, and production scalability.
Universal Model Adapters
Swap LLM providers with zero refactoring. Standardized request/response interfaces for OpenAI, Anthropic, Gemini, and Ollama.
Enterprise AI Gateway
Multi-provider routing, load balancing, rate limiting, and automated fallbacks to keep agent uptime at 99.99%.
Safety Guardrails Engine
Intercept prompts and agent responses. Validate JSON schemas, block forbidden keywords, cap token lengths, and redact PII.
Evaluation Harnesses
Automated evaluation suites for scoring agent reasoning, tool invocation accuracy, latency benchmarks, and regression tests.
Zod Type-Safe Tools
Define agent tools using Zod schemas for runtime argument parsing, automatic function calling, and clean error handling.
Pure TypeScript Core
Built for modern Node.js and browser edge runtimes with 100% strict type safety and zero heavy dependencies.
Pipeline Architecture
How AgentCore handles prompt validation, gateway routing, LLM execution, and safety output filtering.
Input Guardrail
Policy & schema check
AI Gateway
Smart fallback routing
Model Provider
OpenAI / Claude / Gemini
Output Guardrail
Sanitized result return
Interactive Playground Simulator
Simulate prompt execution and trace real-time pipeline logs.