bdd6362c1a
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-Claude) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
27 lines
1.6 KiB
Markdown
27 lines
1.6 KiB
Markdown
# AgentLens Launch -- LinkedIn Post
|
|
|
|
---
|
|
|
|
**Open-sourcing AgentLens: observability for AI agents that traces decisions, not just API calls**
|
|
|
|
If you're building AI agents, you've probably hit this: your agent does something unexpected, you open your observability dashboard, and all you see is a list of LLM API calls with latencies and token counts. Helpful for cost tracking. Not helpful for understanding why the agent chose that path.
|
|
|
|
I spent the last two weeks building AgentLens to address this. It's an open-source observability tool that traces agent decisions -- tool selection, routing, planning, retries, escalation, memory retrieval -- and captures the reasoning and alternatives at each decision point.
|
|
|
|
The idea is simple: if you can see what your agent considered and why it chose what it chose, debugging and improving agent behavior gets a lot more tractable.
|
|
|
|
What's included:
|
|
|
|
- Python SDK with OpenAI auto-instrumentation (pip install vectry-agentlens)
|
|
- Next.js dashboard for exploring decision flows and timelines
|
|
- Self-hostable via Docker Compose (PostgreSQL + Redis)
|
|
- MIT licensed
|
|
|
|
This is v0.1.0. It works, but it's early. The decision taxonomy is still evolving and there are rough edges. I'm sharing it now because I'd rather get feedback from people actually building agents than polish it in isolation.
|
|
|
|
Live demo: https://agentlens.vectry.tech
|
|
Repository: https://gitea.repi.fun/repi/agentlens
|
|
PyPI: https://pypi.org/project/vectry-agentlens/
|
|
|
|
If you're working with autonomous agents, I'd genuinely like to hear: what does useful observability look like for your use case? What decision types matter most to you?
|