Launching today
Breadcrumb
Open-source LLM tracing for agent visibility
77 followers
Open-source LLM tracing for agent visibility
77 followers
Breadcrumb is the Plausible of LLM tracing. Self-hosted, open source, and built for developers who just want to understand what their agents are actually doing without the enterprise bloat of LangFuse or LangSmith. Three lines to get your app traced. An LLM watches every trace and automatically flags issues: wrong tool calls, looping agents, oversized models and cost spikes, all before you even know something's wrong. Ask questions about your traces in plain English and get charts back.
Really like this, debugging agents gets painful fast once things get complex.
The “silent failure” issue is spot on, things look right but something breaks underneath. Visibility here feels super valuable.
Curious how this scales with lots of nested agents and traces?
Also big +1 on self-hosted + privacy.
@wes_dieleman Absolutely!
The traces are stored in Clickhouse so even for large amounts of traces performance is good. Same goes for the UI, huge prompts should be no issue!
Hey everyone! AI agents are surprisingly easy to build. Understanding what they're doing is another story.
I recently had a complex coding agent where subagents silently stopped passing responses to each other. An error somewhere in the chain, but instead of failing loudly, the agents just worked around it. Output looked almost right. I only found it by accident after hours of debugging. With many, many tool calls and nested agents, you're mostly blind.
You can't fix what you can't see.
Breadcrumb gives you visibility into what your agents are actually doing. An LLM watches every trace and automatically surfaces issues like this: silent failures, agent loops, wrong tool calls, cost spikes, before you spend hours hunting them down. There's also an explore tab where you ask questions about your traces in plain English and get real charts back.
Open beta is live today. One click Railway deploy, fully self-hosted and open source.
A hosted version is planned (sign up here https://breadcrumb.sh/docs/setup...).
Give it a try here http://demo.breadcrumb.sh/!
Would love to hear what you're building and what you're struggling with when debugging your agents!
Thanks you so much!!! I was always wondering where do I spend most of the money. Thanks also for contributing to privacy and opensource.
@bogomep Glad to hear you like it! Let me know if you have any feedback!
Does it work with Claude Agent SDK or Claude Code?
@milko_slavov It can feed traces back to Claude code for your development workflow through MCP. Tracing agent SDK itself should also be possible with the breadcrumb typescript sdk