Show HN: I built lightweight LLM tracing tool with CLI
TL;DR
A developer built 'lightrace', a lightweight LLM tracing tool, after a year of struggling to debug agentic applications with existing solutions.
Key Points
- The tool is 100% open source and ships with a CLI interface for quick onboarding without heavy configuration.
- Core feature: the ability to re-call individual tool invocations to isolate failures in agent pipelines.
- Planned additions include caching, step re-execution, and branching – the project is actively seeking contributors.
Nauti's Take
Lightrace is very early-stage – one point and one comment on HN says it plainly. But the core idea is sound: anyone building agentic workflows knows the pain of not knowing which tool call broke the chain.
The CLI-first approach is appealing because it assumes no extra UI infrastructure. Whether the project gains traction depends on how quickly the planned features land and whether documentation keeps pace.
Developers who want to contribute to real AI-agent infrastructure tooling should check out the repo.
Context
Debugging remains one of the biggest bottlenecks in building agentic systems: when ten tool calls run in sequence and one fails, root-cause analysis gets tedious fast. A lean, CLI-first tool that directly targets this problem fills a real gap – especially for developers who don't want to spin up a full observability stack. The concept of selectively re-invoking individual tool calls is architecturally smart and could significantly shorten debugging loops.
Sources
