Observability startup Raindrop AI’s new open supply, MIT Licensed "Workshop" device, launched right now, offers builders one thing that they've probably wished, maybe subconsciously, because the agentic AI period kicked off in earnest final 12 months: an area debugger and analysis device particularly designed for AI brokers, permitting devs to see all of the traces of what their agent has been doing in a single, light-weight Structured Question Language (SQL) database file (.db)
It features as an area daemon and UI that streams each token, device name, and determination to an area dashboard—sometimes hosted at localhost:5899—the second it happens. By visiting their localhost, builders can then see all the things their agent was as much as — together with errors or errors — and establish what went mistaken, when, and ideally, discern why. It's all saved in a single .db file, which takes up comparatively little reminiscence, in response to a X direct message VentureBeat acquired from Ben Hylak, Raindrop's co-founder and CTO (and a former Apple and SpaceX engineer).
This real-time telemetry eliminates the latency of conventional polling and addresses a rising developer concern relating to the privateness of sending native traces to exterior servers.
The device is on the market for macOS, Linux, and Home windows. It may be put in by means of a one-line shell command that automates binary placement and PATH configuration for bash, zsh, and fish shells. For builders preferring to construct from supply, the repository is hosted on GitHub and makes use of the Bun runtime.
The product: establishing a self-healing eval loop
The platform’s standout characteristic is the "self-healing eval loop," which permits coding brokers like Claude Code to learn traces, write evals towards the codebase, and repair damaged code autonomously.
In a sensible software, if a veterinary assistant agent fails to ask crucial follow-up questions, Workshop captures the total trajectory. Claude Code then reads this hint, writes a selected eval, identifies the logic error within the immediate or code, and re-runs the agent till all assertions go.
Compatibility and ecosystem integration
Workshop is appropriate with a broad vary of programming languages, together with TypeScript, Python, Rust, and Go.
It integrates with standard SDKs and frameworks such because the Vercel AI SDK, OpenAI, Anthropic, LangChain, LlamaIndex, and CrewAI. It’s also designed to work seamlessly with numerous coding brokers, together with Claude Code, Cursor, Devin, and OpenCode.
Licensing and neighborhood implications
Workshop is launched underneath the MIT License, making certain it stays free and open-source for all customers. This permissive licensing is meant to foster neighborhood contribution and permit enterprise customers to keep up information sovereignty.
Hylak famous on X that the device was constructed to supply a "sane" strategy to debug brokers domestically, altering how their workforce and early clients construct autonomous methods.
To rejoice the launch, Raindrop provided limited-edition bodily merchandise to customers who put in the device and executed a selected "drip" command.
