AI Judge changed title after evaluation, original title: "Raindrop AI releases Workshop debugger for AI agents"
Raindrop AI released Workshop, an open-source local debugging tool for AI agents. Co-founder Ben Hylak posted the GitHub repository raindrop-ai/workshop with a one-command installer. The daemon and Vite UI stream tokens, tool calls, spans, durations, and costs to localhost:5899. The MIT-licensed project supports Claude and Codex models and runs inside desktop environments such as Cloudflare containers. Early adopters at Thine, Merlin, and Raindrop report shorter development cycles without cloud forwarding.
our site: http://raindrop.ai/workshop github: https://github.com/raindrop-ai/workshop install: curl -fsSL https://raindrop.sh/install | bash
we built the first sane way to debug your agent locally.
you can see your traces. codex/claude code can too. this lets them write evals and test your agents automatically.
best part: it's completely free and open source. install with 1 line.
(github below)
Many users are excited about Raindrop's new open source tool for local AI agent debugging because it offers practical features like auto evals and trace visualization that feel long-awaited and useful in daily workflows.
No Digg Deeper questions have been answered for this story yet.
we built the first sane way to debug your agent locally.
you can see your traces. codex/claude code can too. this lets them write evals and test your agents automatically.
best part: it's completely free and open source. install with 1 line.
(github below)
Heard it works with your codex subscription. Now you can use codex to build agents as well as inspect them
I tried it out yesterday and it works great in the in app browser of the codex desktop app
we built the first sane way to debug your agent locally.
you can see your traces. codex/claude code can too. this lets them write evals and test your agents automatically.
best part: it's completely free and open source. install with 1 line.
(github below)
also, we launched on product hunt!
6 more upvotes and then our company will be worth more than instagram.
link below
ever wanted to see claude code traces, tool-call by tool-call? now you can.
This is cool you can also set it up to see your claude code usage (useful if you're checking what actually goes in the final prompt when using an orchestration framework)
This is very well done. Built in chat that plugs into your local codex/claude is perfect.
Need someone to make this for general otel too
we built the first sane way to debug your agent locally.
you can see your traces. codex/claude code can too. this lets them write evals and test your agents automatically.
best part: it's completely free and open source. install with 1 line.
(github below)
@benhylak why tf is instagram doing product hunt
This is insanely cool. 🔥🔥
One of the best products out there (by far) for anyone building agents just released an OSS version to play with.
We’ve already been using Raindrop heavily across Thine / Merlin. Excited to now use it across personal projects too.
we built the first sane way to debug your agent locally.
you can see your traces. codex/claude code can too. this lets them write evals and test your agents automatically.
best part: it's completely free and open source. install with 1 line.
(github below)
This is cool you can also set it up to see your claude code usage (useful if you're checking what actually goes in the final prompt when using an orchestration framework)
we built the first sane way to debug your agent locally.
you can see your traces. codex/claude code can too. this lets them write evals and test your agents automatically.
best part: it's completely free and open source. install with 1 line.
(github below)
because it's local, every single token, every single tool call shows streams instantly. why are you sending local traces to someone else's server?
we've been using this extensively internal, and with select customers. it's completely changed the way we build agents.
our site: http://raindrop.ai/workshop github: https://github.com/raindrop-ai/workshop install: curl -fsSL https://raindrop.sh/install | bash
here's what an engineer at @gcai thinks:
"The old loop was: run evals, wait, dig through traces, patch the issue, rerun, repeat. With Workshop, my coding agent is in the loop while the run is still happening." - Brian Rhindress, AI Engineer, http://GC.AI
because it's local, every single token, every single tool call shows streams instantly. why are you sending local traces to someone else's server?
we've been using this extensively internal, and with select customers. it's completely changed the way we build agents.
@benhylak mom: we have recursive-improve (https://github.com/kayba-ai/recursive-improve) at home
yes, codex is integrated right on the side.
Heard it works with your codex subscription. Now you can use codex to build agents as well as inspect them
I tried it out yesterday and it works great in the in app browser of the codex desktop app
@benhylak Great work Ben, your repo is live on RepoRank 🙌
https://reporank.co/repo/raindrop-aiworkshop
@benhylak This is actually incredible. I've been building out a project that involved running Pi in a cloudflare container, it's been a nightmare locally to try and get decent observability on what's happening and to get a feedback loop going. This really came at the perfect time!
we built the first sane way to debug your agent locally.
you can see your traces. codex/claude code can too. this lets them write evals and test your agents automatically.
best part: it's completely free and open source. install with 1 line.
(github below)
@alexisgauba @benhylak Still exploring and playing with it, but I thought this was particularly tasteful - just debugged an issue I was having through the in app chat.
https://github.com/raindrop-ai/workshop
first day of raindrop workshop:
1. just hit 200 stars 2. got our first pr!
(github link below)
very cool
I made a similar local mode for the Anything agent (same reason), but wish this had existed
we built the first sane way to debug your agent locally.
you can see your traces. codex/claude code can too. this lets them write evals and test your agents automatically.
best part: it's completely free and open source. install with 1 line.
(github below)
link: https://raindrop.ai/workshop?ref=producthunt
@benhylak First of all congrats! Love this.
I don’t mean this negatively but what might you say to someone like me that already has an at home server with litellm proxy integrated with langfuse to collect and analyze my traces already?
I’m open to switching but just curious the diff
@benhylak scenario: you finally repro the bug at hour 3, claude code auto-compacts mid-session, the context that was triggering it is gone. local trace tools help, but session persistence is the upstream fix. https://fazm.ai/r/wk52kw27