agent-replay · packages

A convenience re-export package that provides the complete public API of the Agent Replay ecosystem—core engine, LLM interceptors, and shared types—from a single import.

A CLI tool for recording, replaying, debugging, and comparing AI agent traces, providing five subcommands (`record`, `replay`, `explore`, `diff`, `debug`) that operate on `.artrace.json` trace files.

Recording and replay engine for AI agent traces, providing a `RecordingEngine` to capture span-structured interactions with events and checkpoints, and a `ReplayEngine` to deterministically replay those traces in stubbed, live, or partial modes without consuming LLM tokens.

Provides callback handlers (`createLangChainHandler`) and state machine hooks (`createLangGraphHooks`) that record LangChain and LangGraph agent interactions into traces, returning handler objects compatible with each framework's interfaces.

Monkey-patches OpenAI and Anthropic SDK clients to transparently record all LLM API calls into Agent Replay traces, providing `OpenAIInterceptor` and `AnthropicInterceptor` classes that wrap `client.chat.completions.create` and `client.messages.create` respectively. Requires `@reaatech/agent-replay-core` for the `RecordingEngine`.

Shared types, interfaces, error classes, and configuration for the Agent Replay ecosystem, providing the canonical trace model shapes, LLM provider abstractions, and storage contracts as pure TypeScript exports with zero runtime dependencies.

Web-based trace viewer for Agent Replay that provides an interactive UI for exploring recorded agent traces, including span timelines, event inspection, checkpoint navigation, and diff visualization.