Curate-Me vs Langfuse
Langfuse is the leading open-source LLM observability and evaluation platform. By the time of acquisition it had 2,000+ paying customers, 20K+ GitHub stars, 26M+ SDK installs per month, and was trusted by 19 of the Fortune 50. In January 2026, Langfuse was acquired by ClickHouse as part of a $400M Series D that valued ClickHouse at $15B.
Langfuse is observability-first; its strengths are tracing, evaluations, and prompt management. Curate-Me is governance-first; our strengths are cost caps, PII scanning, HITL approvals, and managed agent runners. The two are often complementary — but if you’re picking one platform, here’s how they compare.
Feature comparison
| Capability | Curate-Me | Langfuse |
|---|---|---|
| LLM proxy / gateway | 50 providers (44 production-ready) | Not a proxy — observability via SDK instrumentation |
| Deployment model | Fully managed (SaaS) | Self-hosted (OSS) or Langfuse Cloud, ClickHouse-backed |
| Distributed tracing | W3C trace context, span-level cost | Industry-leading; nested traces, generations, spans |
| Evaluations | Not available | LLM-as-judge, custom scorers, dataset experiments |
| Prompt management | Not available | Versioned prompts, releases, A/B testing |
| Cost tracking | Per-request, per-key, per-org hierarchies | Per-trace, per-user, per-feature |
| Rate limiting | IETF-standard, per-key | Not available |
| PII scanning | 33 regex patterns + optional Presidio NER | Redaction filters in SDK |
| Content safety scanning | Prompt injection + jailbreak detection | Not available |
| HITL approval queues | Built-in, configurable thresholds | Not available |
| Managed execution environments | OpenClaw runners with full lifecycle | Not available |
| In-container governance | MCP server (50 tools) | Not available |
| Desktop streaming | Live VNC/RDP monitoring | Not available |
| A2A communication | Protocol with PII scanning | Trace propagation between services |
| Storage architecture | MongoDB + Redis (portable) | ClickHouse (post-acquisition, integral to platform) |
| Open-source | No | Yes (MIT, with EE features) |
| Future direction | Governance-first, runner-native | ClickHouse analytics integration |
Where Langfuse is stronger
- Observability depth: Langfuse is purpose-built for tracing and evaluation. Nested traces with generations, spans, and scores are best-in-class.
- Evaluations: LLM-as-judge, custom scorers, and dataset-based experiments are core features. We do not ship evaluations today.
- Prompt management: Versioned prompts with releases, A/B testing, and production traffic splitting.
- Community and adoption: 20K+ GitHub stars, 26M SDK installs/month, 2K+ paying customers, 19 of Fortune 50. The most adopted OSS LLM observability platform.
- Self-hostable: MIT-licensed, full source available. You own your data.
Where Curate-Me is stronger
- Gateway with governance: Langfuse is SDK-based observability — it sees what you tell it to see. We’re a transparent gateway with 14-stage governance applied to every request, including PII, content safety, and rate limits.
- HITL approvals: High-cost or sensitive requests held for human review with auto-replay. Required for EU AI Act Article 14 compliance. Langfuse has no equivalent.
- PII scanning and content safety: Every request scanned for secrets, PII, prompt injection, and jailbreak attempts before reaching the provider. Langfuse can redact at instrumentation time but doesn’t enforce policy.
- Managed runners: Full agent execution environments — sandboxed containers with lifecycle management, desktop streaming, fleet orchestration. Langfuse is observability only.
- In-container governance: MCP server with 50 tools running inside the execution environment for cost caps, tool restrictions, A2A communication.
- Architecture portability: Our storage stack is MongoDB + Redis. Langfuse is built on ClickHouse, which means post-acquisition the platform’s roadmap is tied to ClickHouse’s analytics direction.
When to choose Langfuse
- Tracing, evaluations, and prompt management are your primary needs
- You want the most mature OSS observability platform with a large community
- You’re comfortable with the ClickHouse-backed architecture and roadmap
- You don’t need a gateway, governance enforcement, or managed runners (private beta)
- You want to self-host and own the entire stack
When to choose Curate-Me
- You need a gateway with enforced governance, not just observability
- You run AI agents that need execution environments (not just LLM calls)
- You need HITL approvals, PII scanning, or content safety as enforcement (not just visibility)
- You want one platform instead of stitching Langfuse + a proxy + a runner platform
- You need an immutable audit trail for compliance (EU AI Act, SOC 2)
Using them together
Langfuse and Curate-Me play well together. Many teams instrument with Langfuse for tracing/evals while running their traffic through Curate-Me for governance. Both support W3C trace context, so trace IDs propagate cleanly. If you’re using both today, you can keep doing that.
Ready to try it? Follow the Quick Start to connect in under 5 minutes.