Langfuse alternatives fall into three groups: tracing and evaluation platforms for agents and LLM apps, AI gateways that log every request as it passes through, and product analytics tools that track AI usage next to everything else. Choose by where you want visibility to start, then compare the ten options below.
Langfuse now describes itself as open source agent evals and observability by ClickHouse, and ClickHouse's own page says Langfuse is now part of ClickHouse. If that change is why you're looking around, there's a wide spread of options, including open source ones you can run yourself.
The ten tools below are listed alphabetically and in no particular order, and every description comes from each vendor's own website, read as of Oct 2026. My guide to the best LLM observability tools goes broader; this page is built for someone leaving Langfuse.
Langfuse alternatives at a glance
I checked the free plan column on each vendor's own website, as of Oct 2026. Yes means the page names a free plan, a free version or an open source edition. Trial only means it names a time-limited free trial. Not stated means the page I read named neither, so check the vendor's own site before you sign up.
| Tool | Type | Good for | Free plan |
|---|---|---|---|
| Arize Phoenix | Open source observability and evals | Self-hosted tracing and evaluation for AI agents | Yes |
| Braintrust | Agent observability platform | Production traces, live monitoring and evals in one place | Not stated |
| Confident AI | Enterprise evaluation and observability | LLM evals in CI plus tracing and alerts | Not stated |
| Galileo | AI observability and eval engineering | Turning offline evals into production guardrails | Not stated |
| Helicone | AI gateway with observability | Routing, debugging and analysing LLM requests | Trial only |
| LangSmith | Agent and LLM observability | Step-by-step agent tracing and online evals | Not stated |
| Lunary | Enterprise AI platform | Chatbot analytics with self-hosting and PII masking | Not stated |
| Portkey | AI gateway, now Prisma AIRS | Gateway, guardrails, governance and prompt management together | Not stated |
| PostHog | Product context layer | Product analytics with generous monthly free tiers | Yes |
| Traceloop | LLM reliability platform | One-line OpenTelemetry tracing, tests and monitors | Not stated |
Ten LLM observability and evaluation tools
Arize Phoenix
Best for: Engineers who want open source tracing and evals they can run locally with one command, with a managed cloud option when the team grows.
Braintrust
Best for: Teams that want to define what good looks like before shipping, then log agent traces, monitor live performance and annotate results in custom views.
Confident AI
Best for: Engineering teams that want eval coverage on every LLM change in CI, research-backed metrics, and tracing with alerts once the system is in production.
Galileo
Best for: Companies that want the evals they write offline to become guardrails in production, from a platform built around eval engineering.
Helicone
Best for: Developers who want a gateway that routes every model call and logs it for debugging and analysis, with a 7-day free trial and no card needed.
LangSmith
Best for: Teams building agents who want to see each step the agent takes, score quality with online evals, and self-host or bring their own cloud.
Lunary
Best for: Enterprises that need chatbot analytics and PII masking from a platform that's SOC 2 Type II and ISO 27001 certified and designed to be self-hosted.
Portkey
Best for: Large organisations that want an AI gateway, observability, guardrails, governance and prompt management in one production stack, now sold as Prisma AIRS AI Gateway.
PostHog
Best for: Product teams that want AI usage in the same place as their product analytics, from a vendor that says 97% of its customers use it for free.
Traceloop
Best for: Developers already on OpenTelemetry who want tracking, testing and monitoring with one line of code, built on the open source OpenLLMetry SDK.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Ownership changes are reshaping LLM tooling
Here's what the home pages say right now. Portkey's page announces it's now PRISMA AIRS AI Gateway, generally available for all enterprises. Traceloop's page carries a banner saying it's joining ServiceNow. And Helicone's page says it's no longer part of Mintlify.
Braintrust now calls itself the active observability platform for agents, and that shift from LLM calls to agents shows up across the category. None of this rules a tool out, but if you're leaving because of an acquisition, check who owns the replacement first.
Tracing platform, gateway or product analytics?
- Tracing and evaluation platforms. You add an SDK, send traces and run evals against them. This is the closest like-for-like swap, and several are open source or self-hostable.
- AI gateways with logging. Every model call passes through a proxy that routes it and logs it. You get visibility without touching much code, plus routing and guardrails.
- Product analytics with AI usage. Useful when the question is how AI features affect your users, not why one prompt failed.
Questions to settle before you swap LLM observability tools
- Does your data have to stay in-house? If so, start with the open source or self-hosted options.
- Are you tracing agents or single calls? Multi-step agents need step-by-step traces and threads.
- Do you want evals in CI? Some tools here run eval coverage on every change before it ships.
- Do you already use OpenTelemetry? An OTel-based SDK makes switching again later far easier.
- Who owns the vendor? Two tools on this list announced a new owner or a deal on their own home pages.
If you're building retrieval into the app as well, my roundup of vector databases covers the storage side.
Do you make an LLM observability or evaluation tool? Tool makers can ask to be considered for a featured place on this list.
My plain advice on replacing Langfuse
Export a week of traces before you change anything, so you've got a baseline to compare against. Then instrument the same app with two candidates, ideally one open source and one hosted, and see which one helps you find a real failure faster. If you'd rather not add an SDK at all, a gateway gives you logs from the very first request.
Whichever you choose, write down what a good answer looks like before you start, because an eval tool can only score against the standard you give it.
Want your tool on this list? Featured spots and reviews here are paid partnerships. Here's how to partner with me.
Frequently asked questions
Is Langfuse still open source?
Its home page still describes it as open source agent evals and observability, now by ClickHouse, as of Oct 2026. Check the licence terms on its own site if that matters to your team.
What's the difference between LLM observability and evaluation?
Observability shows what happened: traces, latency, errors and the steps an agent took. Evaluation scores whether the output was good, either before release in tests or on live traffic.
Can an AI gateway replace an observability tool?
Partly. A gateway logs every request and response, which covers a lot of debugging. Dedicated tracing tools go deeper on multi-step agents and evals, so many teams run both.
Which Langfuse alternatives can I self-host?
Several. The open source tracing tools here can run locally, and others offer self-hosted or bring-your-own-cloud options for data residency, as of Oct 2026.
Do I need a separate tool for LLM evals?
Not always. Most tools on this list combine tracing and evals. A separate eval tool makes sense when you want eval coverage in CI on every change.
