🇩🇪 Germany · Langfuse — Open-Source LLM Observability
Status: 🟩 COMPLETE 🟦 LIVING Section: 10 — AI and LLMs
| Vendor | Langfuse GmbH |
| Country/origin | 🇩🇪 Germany (Berlin) |
| Recommended for AUS? | ✅ Strongly yes for privacy-sensitive use — EU-origin; open-source; self-hostable; GDPR governed |
| Privacy summary | Self-hosted option = full data sovereignty; cloud option in EU; GDPR-native; open-source code; standard SaaS or your-own-infrastructure |
| Free tier | ✅ Free (open-source self-hosted; or generous cloud free tier) |
| Paid tiers | Cloud Pro (~199); Enterprise self-host & cloud |
| First released | Founded 2022; Y Combinator 2023; rapid growth 2024-2026 |
| Last reviewed | June 2026 |
| Official site | https://langfuse.com |
What it is
Langfuse is an open-source LLM application observability and analytics platform — the leading alternative to LangSmith. It provides tracing, evaluation, prompt management, and analytics for AI applications, with a distinctive emphasis on:
- Open-source — code on GitHub (Apache 2.0 license); audit it, modify it
- Self-hostable — run on your own infrastructure for maximum data control
- EU origin — built under GDPR; appeals to European and privacy-conscious teams
For Australian developers building AI applications who care about data sovereignty, Langfuse is the strongest choice in this category.
What it does
Same core capabilities as LangSmith — and increasingly more:
Tracing
- Capture every LLM call in your application
- Multi-step agent traces with nested span visualisation
- Cost and latency tracking per trace
- User session grouping
Prompt management
- Version-controlled prompts centrally
- Prompt deployment with rollback
- Prompt playground for iteration
Evaluation
- Datasets for test cases
- Automated evaluators (LLM-as-judge, custom code)
- Human annotation queues
Analytics
- Dashboards for production AI metrics
- User feedback integration
- Cost analysis by feature/user/model
Integrations
- Framework-agnostic: works with LangChain, LlamaIndex, vanilla OpenAI/Anthropic SDK, custom code
- SDK in: Python, TypeScript/JavaScript, more
What you’d use it for
Same as LangSmith — observe, debug, evaluate, and improve AI applications. The difference is how you host it and who’s the parent organisation.
How to access from Australia
Option 1: Cloud (managed)
- Go to https://langfuse.com → Sign up
- Free tier on signup
- Get API keys
- Integrate into your code
Option 2: Self-hosted (open-source)
- Visit https://github.com/langfuse/langfuse
- Clone the repository
- Run with Docker Compose or deploy to your cloud
- Free; you pay only for the hosting
For Australian organisations with strong data sovereignty needs, the self-hosted option on AWS Sydney gives complete data control while still using the polished Langfuse platform.
What it costs
Self-hosted: $0 (open-source) + your infrastructure costs
Cloud:
| Plan | Price | Best for |
|---|---|---|
| Free | $0 | 50,000 events/month; basic features |
| Pro | $59 USD/month/team | More events; better support |
| Team | $199 USD/month | Production teams |
| Enterprise | Custom | SSO, advanced compliance |
How it compares to LangSmith
The two main LLM observability platforms differ meaningfully:
| Aspect | Langfuse | LangSmith |
|---|---|---|
| License | Open-source (Apache 2.0) | Commercial |
| Self-host | ✅ Free; documented | Enterprise only |
| Country | 🇩🇪 Germany (EU) | 🇺🇸 USA |
| Framework integration | All (LangChain + others) | LangChain native |
| GDPR posture | Native | Standard DPA |
| Pricing model | Cloud SaaS or free | Per-user SaaS |
| Polish | Strong; rapidly improving | More mature |
Strong case for Langfuse:
- You want self-hosted for data sovereignty
- You’re building with OpenAI/Anthropic SDK directly (not LangChain)
- You’re in EU or care about EU data handling
- Open-source matters to you (you want to audit/modify)
- Cost-sensitive (free self-host)
Strong case for LangSmith:
- You’re using LangChain heavily
- You want the most polished hosted experience
- You don’t have infrastructure capacity for self-hosting
- You’re already in the LangChain ecosystem
Why open-source observability matters
For AI infrastructure, open-source has specific advantages:
- No vendor lock-in: You can switch hosting providers or stop using the service entirely
- Audit the code: You can see exactly what data is captured and how
- Modify for your needs: Add custom features, integrate with internal systems
- Long-term sustainability: Open-source projects can outlast their original companies
- Cost control: Self-hosted = predictable infrastructure costs
For Australian organisations building serious AI products, the option to self-host is genuinely valuable.
Privacy considerations
Self-hosted
- Maximum data control: No third party sees your AI traces
- Australian data residency: Run on AWS Sydney or any Australian infrastructure
- Compliance flexibility: Easier to meet APP 11 security obligations on your own infrastructure
- Customisable retention: Control exactly how long data is stored
Cloud (Langfuse-hosted)
- EU data centres by default
- GDPR-native architecture
- SOC 2 Type II in progress (verify current status)
- Standard SaaS DPA available
For most Australian use cases involving any personal information being processed by AI, self-hosted Langfuse on Australian cloud is the strongest choice.
Australian considerations
- Self-host on AWS Sydney: Combines Langfuse’s capabilities with full Australian data sovereignty
- Cost-effective: Open-source self-hosted = significant savings vs SaaS at scale
- Australian AI startups using Langfuse: Growing community of Australian AI builders choosing open-source observability
- Government and enterprise: Self-hosted Langfuse can be deployed in air-gapped or sovereign cloud environments
Setting up self-hosted Langfuse
The process for technical Australian teams:
- Provision a server (AWS Sydney, Azure Australia East, etc.)
- Clone the Langfuse repository
- Configure environment variables (database, secrets)
- Deploy with Docker Compose or Kubernetes
- Point your AI application’s Langfuse SDK at your self-hosted instance
Documentation is good but requires technical expertise. Plan for a day of setup; longer for production-grade deployment.
Gotchas
- Self-hosting requires DevOps skills. Not plug-and-play; you need someone comfortable with Docker, databases, and server administration.
- Self-hosted needs maintenance. You’re responsible for updates, security patches, backups.
- Cloud free tier may be sufficient for many. Don’t self-host if you don’t need to.
- Cost scales differently: Self-hosted is more economical at high volume; cloud is more economical at low volume.
- Production scaling considerations: A heavily-used self-hosted Langfuse needs proper database scaling, monitoring, etc.
Recent changes (LIVING)
- Langfuse v3 (2024-2025): Major upgrade with better performance and features
- Multi-modal observability (2025): Beyond text to image, audio, video AI traces
- Improved evaluations framework (2024-2026): More sophisticated quality assessment
- Y Combinator + Series A funding (2024): Significant resources to grow the platform
See also
- langsmith — main commercial competitor
- helicone — simpler observability alternative
- open-weights-vs-closed — open-source AI tooling philosophy
- australian-privacy-considerations — privacy law context
- vercel-ai-sdk — application framework integration
Sources
- Langfuse documentation: langfuse.com/docs
- Langfuse GitHub: github.com/langfuse/langfuse
- Y Combinator W23 batch announcement
- Series A funding coverage (TechCrunch, 2024)
- Developer community discussions (2024-2026)