Raindrop AI Open Source Local Agent Debugging Tool Workshop, offering a visual dashboard, deeply integrated with Claude Code, Cursor, and others, capable of directly reading execution traces, automatically generating test cases and modifying code to form a self-healing closed loop. Supports local streaming token synchronization, tool invocation, and decision chain links, and can replay production traces via CLI. MIT open source, compatible with TS/Python/Go/Rust, integrated with multiple frameworks.

MeNews

2026-05-23 20:00:37

Abstract generation in progress

AIMPACT News, May 15 (UTC+8), according to Beating Monitoring, Raindrop AI has open-sourced a local Agent debugging tool called Workshop. This tool visualizes the debugging interface and deeply integrates code Agents like Claude Code, Cursor, and others, allowing large models to directly read Agent operation traces, automatically generate evaluation test cases, and modify business code, forming a self-healing code loop. Workshop provides a locally running dashboard that can stream real-time synchronization of each token, tool calls, and decision links during Agent operation without polling. The tool has a built-in local playback mechanism that allows developers to generate an HTTP endpoint via command line to replay captured production environment traces directly in a real local environment. In practical workflows, if developers find that the business Agent's behavior deviates from expectations, they can directly invoke the code Agent in the terminal. The integrated Workshop code Agent will proactively read logs, write test scripts for failure scenarios, and modify the corresponding code. The code Agent will then automatically trigger retries, continuously evaluating and fixing until all test assertions pass. Workshop is now open-sourced under the MIT license. The tool supports development in TypeScript, Python, Go, and Rust, and integrates with mainstream frameworks such as Vercel AI SDK, LangChain, CrewAI, and others. (Source: BlockBeats)

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.

9 Likes

Reward
9
6
1
Share

Comment

Add a comment

AirdropOrganizer

· 1h ago

The description seems to suggest installing a black box + automatic repair shop for the AI agent; the actual implementation remains to be verified.

View OriginalReply0

TokenomicsTailor

· 7h ago

The agent debugging tool has reached this level; the pressure from Cursor and Claude Code is here.

View OriginalReply0

GasFeeSensitivity

· 17h ago

Reading from the trajectory to automatically generate test cases and modify the code, this pipeline design is indeed very smooth.

View OriginalReply0

L2ArbitrageTrader

· 17h ago

Local streaming token synchronization is so crucial; finally, there's no need to send data outside.

View OriginalReply0

PerpWhisperer

· 17h ago

Visual dashboard + self-healing closed loop, this debugging experience is directly maximized, Claude Code users are ecstatic

View OriginalReply0

GateUser-0fdb3438

· 17h ago

CLI replay of production traces is simply a lifesaver; reproducing bugs no longer relies on guesswork.

View OriginalReply0

Trending Topics
View More
#
TradfiTradingChallenge
344.02K Popularity
#
PlatinumCardCreatorExclusive
125.02K Popularity
#
DailyPolymarketHotspot
1.05M Popularity
#
GateSquarePizzaDay
1.79M Popularity
#
SpaceXOfficiallyFilesforIPO
572.88K Popularity

Pinned

Sitemap

Raindrop Open-Source Local Agent Debugger Workshop to Connect Code Agent Self-Healing Loop

Trending Topics

TradfiTradingChallenge

PlatinumCardCreatorExclusive

DailyPolymarketHotspot

GateSquarePizzaDay

SpaceXOfficiallyFilesforIPO

Pinned