reaatech/agent-replay
These packages provide a deterministic recording and replay system for AI agent interactions, capturing LLM calls, tool invocations, and routing decisions as structured traces. You would adopt them to debug agent behavior and run integration tests without incurring repeated LLM API costs or latency. The system uses runtime interceptors and framework-specific integrations to capture state, allowing you to replay, step through, or diff agent execution paths against recorded snapshots.
Packages
7 packages
@reaatech/agent-replay
Provides a unified entry point that re-exports the core engines, LLM interceptors, and shared types for the Agent Replay ecosystem. This package allows you to access the full API surface for recording, replaying, and debugging agent execution traces through a single import.
- status
- published
- published
- 3 days ago
@reaatech/agent-replay-cli
Provides a CLI for recording, replaying, exploring, and diffing AI agent execution traces stored as `.artrace.json` files. It supports multiple replay modes, including stubbed execution for testing and live comparison for debugging agent behavior.
- status
- published
- published
- 3 days ago
@reaatech/agent-replay-core
Provides a suite of classes—including `RecordingEngine`, `ReplayEngine`, and `DiffEngine`—to capture, deterministically replay, and analyze AI agent interactions without consuming LLM tokens. It enables state-based debugging, regression testing, and semantic comparison of agent traces stored as serialized JSON.
- status
- published
- published
- 3 days ago
@reaatech/agent-replay-integrations
Provides callback handlers and state machine hooks to record LangChain and LangGraph interactions into Agent Replay traces. It exports factory functions that generate integration objects compatible with existing framework interfaces, requiring the `@reaatech/agent-replay-core` package to process the captured data.
- status
- published
- published
- 3 days ago
@reaatech/agent-replay-interceptors
Monkey-patches OpenAI and Anthropic SDK clients to automatically record LLM requests and responses into Agent Replay traces. It provides class-based interceptors that require an instance of `RecordingEngine` from `@reaatech/agent-replay-core` to function.
- status
- published
- published
- 3 days ago
@reaatech/agent-replay-shared
Provides the canonical TypeScript types, interfaces, and error classes for the Agent Replay ecosystem. It serves as the shared schema definition for trace models, LLM abstractions, and storage contracts used across all related packages.
- status
- published
- published
- 3 days ago
@reaatech/agent-replay-web-ui
Provides a web-based interface for visualizing and inspecting recorded agent traces, including span timelines, event logs, and diff comparisons. It is designed to consume trace data generated by the @reaatech/agent-replay-core engine.
- status
- awaiting publish
Comments
Sign in with GitHub to comment and vote.
Loading comments…