Agent tool / platformFree source · check hosted pricingLiveFrom public docs
Langfuse
Langfuse (part of ClickHouse)
Open-source LLM engineering platform: tracing, evaluation and debugging, self-hosted or cloud.
Overview
Langfuse records traces of LLM and agent applications so you can monitor, evaluate and debug them. Self-host with Docker Compose, a VM or Kubernetes (Helm, with Terraform templates for AWS, Azure and GCP), or use Langfuse Cloud.
Production criteria
| Approval gatesCan a person or policy hold an action until it is approved? | Not claimed Not documented by the vendor or not yet reviewed. |
|---|---|
| Restart recoveryDoes work resume after a crash or redeploy without repeating finished steps? | Not claimed Not documented by the vendor or not yet reviewed. |
| Evidence trailCan you inspect what ran, with what inputs, and who approved it? | Built in Traces of LLM and agent application runs. |
| Kill switchCan an operator withdraw permission or stop further actions? | Not claimed Not documented by the vendor or not yet reviewed. |
| Policy controlsDo declared rules decide which tools and actions are allowed? | Not claimed Not documented by the vendor or not yet reviewed. |
| Sandbox isolationDoes code or tool execution run inside an isolated boundary? | Not claimed Not documented by the vendor or not yet reviewed. |
Marks reflect the documentation linked below as of 2026-10-29. How we mark criteria.
Limits and scope
- Traces show what your instrumentation sends; coverage depends on how you instrument.
Also in Evidence & observability
- Witness — A provenance prototype that keeps the path from evidence to claims, with signed append-only records.
Free tools, a product pilot, or a focused paid diagnosis.
Choose a tool from the directory, explore Mothership’s technical pilot path, or check fit for a software diagnosis when one issue needs investigation.