"We already use Langfuse for everything" is the most common objection we hear when talking to teams about cost optimization. It's a reasonable thing to think — Langfuse is a genuinely good tool, and it does show cost as a data point. But "we track cost" and "we optimize cost" are different capabilities, and Langfuse is built for the first job, not the second.
1. Langfuse is an AI-native tracing and evaluation platform — deep on prompt versioning, quality scoring, and debugging what a model did.
2. Preto sits at the gateway position and focuses on cost specifically — ranked savings recommendations, budget enforcement, savings tracking.
3. They're not competing for the same job. Most teams with real LLM spend end up running both.
What Langfuse Actually Is
Langfuse was acquired by ClickHouse in January 2026, and rather than the acquisition slowing development, Langfuse had already moved nearly its entire product to MIT license in mid-2025 — tracing, prompt management, evaluations, datasets, and the playground are all open. Only thin enterprise compliance features (SCIM, audit logs, project-level RBAC, UI customization) remain commercial. Cloud pricing runs Hobby (free, 50K units/month), Core ($29/month), Pro ($199/month, 3-year retention), and Enterprise ($2,499/month), metered on ingested traces, observations, and scores rather than seats.
What it's genuinely excellent at: capturing what happened during a model call in detail, versioning prompts as they change, running evaluation suites against your own quality criteria, and giving engineers a place to debug why a specific output was wrong. This is deep, valuable work, and it's mostly done by observing traffic after the fact through SDK hooks — the architecture is built for analysis and iteration, not for shaping a request before it's sent.
What Preto Actually Is
Preto sits in the request path — a one-line proxy swap, not an SDK integration — which means it can act on a request before it reaches the provider, not just log it afterward. That gateway position is what makes budget enforcement possible: a workspace that exceeds its monthly cap can be alerted or hard-blocked at the proxy, something an after-the-fact observer architecturally can't do. On top of that position, Preto runs ranked, dollar-denominated savings recommendations and tracks the money actually recovered as a first-class dashboard metric — not just total spend.
Side by Side
| Capability | Langfuse | Preto |
|---|---|---|
| Trace capture & debugging | Yes — deep, this is the core product | Not the focus |
| Prompt versioning | Yes | Not the focus |
| Quality evaluations | Yes — datasets, scoring, eval suites | Not the focus |
| Cost tracking (spend visibility) | Partial — cost shown per trace | Yes — core dashboard metric |
| Ranked savings recommendations | No | Yes, with dollar estimates |
| Budget enforcement (hard block) | No | Yes, at the proxy layer |
| Savings-tracked as a metric | No | Yes — "money saved this month" |
| Integration model | SDK hooks, mostly after-the-fact | Proxy — in the request path |
| Self-hosting | Yes, full MIT product | Contact for enterprise |
Already running Langfuse?
Add Preto for the cost side — routing, budget enforcement, and savings recommendations Langfuse doesn't cover. They read the same underlying traffic without conflicting.
Start Your Free Cost AuditLangfuse for quality. Preto for cost. Use both. We complement each other.
The Objection, Actually Answered
Because "everything" in Langfuse's case means tracing, evals, and prompt management — not cost optimization. Ask directly: does Langfuse tell you which specific requests should route to a cheaper model, with a dollar estimate of the savings? Does it block a request when a team exceeds its monthly budget? Does your dashboard show a "money saved" number anywhere? If the honest answer to all three is no, Langfuse is doing its job well — it was never built to do this one.
Where Teams Actually Run Both
The pattern that shows up most often isn't "replace Langfuse with Preto" — it's layering them. Langfuse stays in place for prompt versioning and evaluation, because that workflow (iterate on a prompt, score outputs, ship the version that scores best) is genuinely what it's built for. Preto sits at the gateway for the cost side: routing decisions, budget caps, and the ranked recommendation list that tells an engineer specifically what to change and what it's worth.
The two tools read the same underlying traffic without stepping on each other, because they're answering different questions. Langfuse answers "is this output good?" Preto answers "is this the cheapest way to get a good output, and are we staying inside budget while doing it?" A team that only asks the first question can ship a high-quality feature that quietly costs 5x more than it needs to. A team that only asks the second can cut costs into a quality regression nobody notices until a customer complains. Most mature setups need both questions asked, continuously, by tools built for each one.
When Langfuse Alone Is Enough
If your LLM spend is small enough that cost isn't yet a real line item — under roughly $1,000-2,000/month — the case for a dedicated cost tool is weaker, and Langfuse's cost-per-trace view is probably sufficient visibility for now. The calculus changes once spend crosses into five figures a month and a 20-40% avoidable waste rate (the typical range across production traffic) starts to mean real money, at which point ranked recommendations and budget enforcement stop being a nice-to-have and start being the difference between a controlled cost curve and a surprise on next month's invoice.
For more on where gateway-position tools sit relative to observability platforms generally, see gateway vs. proxy vs. router and the full 2026 LLMOps landscape, which covers Langfuse's ClickHouse acquisition and where the rest of the category stands.
Frequently Asked Questions
What's the difference between Preto and Langfuse?
Can I use Preto and Langfuse together?
Is Langfuse still actively developed after the ClickHouse acquisition?
Does Langfuse do cost optimization or budget enforcement?
Keep Langfuse for quality. Add Preto for cost.
One line of code, no conflict with your existing observability stack — Preto adds ranked savings recommendations and budget enforcement on top of what you're already running.
Start Your Free Cost AuditFree forever up to 10K requests. No credit card required.
