See what your AI agent
actually did.
One decorator captures every LLM call your agent makes. View the timeline locally, share it with one link.
free · open source · no account ever
Click into a real trace.
No install needed. These are real shared traces · the same view someone sees when you send them a debug link.
No account required to view.
What your dashboard looks like.
Every run becomes a visual timeline you can explore, inspect, and share.
Getting started.
No account. No API key. Just a decorator.
@trace
def my_agent(q):
# your code, untouched
return answer
glasspipe dashboard. Click any run. Click Share to get a public link.✓ localhost:3000
# share: glasspipe.dev/t/F3Sdpg
When to use GlassPipe (and when not to).
GlassPipe isn't for everyone.
| What you need | Best choice |
|---|---|
| Install in 60s, see a trace, share a debugging link. No account ever. | [ GLASSPIPE ] |
| Student learning how AI agents work internally | [ GLASSPIPE ] |
| Share a debugging session like a CodeSandbox link | [ GLASSPIPE ] |
| Enterprise observability, SSO, SOC2, alerting, async, team workspaces | [ LANGFUSE ] / [ LANGSMITH ] |
[ USE GLASSPIPE IF ]
You want to install a tool, see a trace, and share a debug link without making an account. Indie devs, students, and anyone who'd rather skip the signup.[ USE LANGFUSE OR LANGSMITH IF ]
You need enterprise features, production monitoring, team workspaces, alerting, async support, or SOC2 compliance. They're built for that.We're not trying to replace them. We're built for a different moment.
Questions you'd ask.
Nothing. Not until you choose to share. When you click Share, you see an exact preview of what would upload. Redact anything, then confirm. If it looks wrong, just close it.
The pre-share preview automatically scans for common secrets (API keys, tokens, JWTs, emails) and flags them before anything leaves your machine. You can add your own patterns via GLASSPIPE_REDACT_PATTERNS. But always eyeball the preview yourself. We catch the obvious stuff; you know your codebase.
Not in v1. The local dashboard runs entirely on your machine, no server involved. The public share service runs on glasspipe.dev. Self-hosting is on the roadmap if there's demand.
Not in v1. GlassPipe focuses on synchronous Python with the raw OpenAI and Anthropic SDKs. If you're deep in LangChain or async today, Langfuse or Phoenix are good fits. We'd rather do one thing really well.
Public traces expire after 30 days. You get a delete token when you share, use it to remove a trace early anytime. No account needed.
Install GlassPipe.
Two lines of code. No account.
@trace
def research_agent(topic):
with span("plan", kind="custom") as s:
s.record(input={"topic": topic},
output={"plan": "..."})
# ... your agent code, untouched
return result
research_agent("AI agent observability")
# then: glasspipe dashboard
Currently supports synchronous Python, OpenAI, and Anthropic. MIT license.