scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-23 05:34:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 25

docs(lens): add lens readme

Browse source
This commit is contained in:
Ishaan Jaff 2026-10-08 14:14:06 -07:00
parent cebf30b714
commit 1ef1233be3
No known key found for this signature in database
1 changed files with 45 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

45
litellm/proxy/lens/README.md Normal file
View file

@ -0,0 +1,45 @@
# LiteLLM Lens
**The gateway that helps your agents improve**
![LiteLLM Lens](../../../.github/assets/lens-launch/hero.gif)
Tell Lens what to look for in your agents, let it analyze your traces, then read the findings and go deeper into any trace
## The agentic swarm developer
We're building for a world where one team runs hundreds of agents in parallel. A single swarm can produce 200K+ traces, and every one of those calls already flows through LiteLLM. The gateway is the one chokepoint that sees all of your agent traffic, so it's the natural place to learn from it
## The problem
Nobody can read 200K traces by hand. Tracing platforms make it worse: your data sits in someone else's system behind their rate limits, so tools like Claude Code or Codex can't dig into it the way they dig into your code
## How Lens works
### 1. Tell Lens what to look for
For each agent, describe what a good run and a bad run look like. Add specific questions if you have them, like "Find tool failures the agent does not recover from"
![Questions and checks](../../../.github/assets/lens-launch/questions-and-checks.png)
### 2. Lens analyzes your traces
Agents review your traces, group similar problems together, and surface findings. Each finding tells you what happened, what to do next, and which runs back it up
![Finding detail](../../../.github/assets/lens-launch/finding-detail.png)
### 3. Read the findings, then go deeper into the trace
Every finding links back to the exact step in the trace that caused it. You get the step tree on one side and the full input and output on the other
![Trace detail](../../../.github/assets/lens-launch/trace-detail.png)
### Agent-first tracing APIs
Traces live in ClickHouse that you host next to your LiteLLM gateway. Query them with SQL, or point Claude Code or Codex at the tracing API and let them investigate. No external rate limits, and your data never leaves your infrastructure
## Get started
Lens lives in the LiteLLM dashboard under Observability, Lens (`/ui/lens/`). It needs a Lens worker running next to your gateway. The [worker README](../../../deploy/lens/README.md) covers local installs, existing deployments, Helm, the API, and what a scan actually does
Want help rolling it out? [Sign up for early access](https://forms.gle/3GC1Ner4vjthGWi18), or read the [launch post](https://docs.litellm.ai/blog/litellm-lens-launch) and the [Lens docs](https://docs.litellm.ai/docs/proxy/lens)