Braintrust → LLMeter Migration Guide

You don't need an eval platform to watch LLM cost

Braintrust is excellent at LLM evaluation — prompt iteration, regression testing, scoring against datasets. If the only panel you actually use is the cost dashboard, LLMeter does that half with a read-only API key and zero SDK. Free for 1 provider.

No credit card required. Free plan never expires.

When does the switch make sense?

Braintrust is an eval platform. LLMeter is a cost tracker. Pick by what you actually do.

You log in for cost numbers

If your Braintrust dashboard time is mostly spent on spend per model / customer / environment — that is the part LLMeter handles natively, without the SDK.

You're not running evals

Eval infrastructure has a real maintenance cost (datasets, scorers, regression suites). If your team is not using it, you are paying for unused weight.

You want zero-instrumentation setup

LLMeter pulls billing data from the provider directly. No SDK to wire in. No log wrappers. No deployments to push first.

Keep Braintrust if you run prompt experiments, regression eval suites, or score models against curated datasets — those features have no LLMeter equivalent.

3 steps to switch

You can run them side-by-side first. The migration is additive.

1

Decide if you actually use evals

If 90% of your Braintrust usage is the cost panel, LLMeter does that half without instrumentation. If you actively run experiments / regression eval suites, keep Braintrust.

2

Sign up for LLMeter + paste a key

Create a free account, generate a read-only key in your provider dashboard, paste it. No SDK, no log wrapping, no dataset upload.

3

Remove Braintrust SDK from cost-only services

For services where Braintrust was instrumented purely for the cost dashboard, you can remove the SDK and reclaim the bundle size and latency. Keep it in services that run real evals.

Braintrust vs LLMeter

Different products, overlapping cost panel. Here is what changes.

FeatureBraintrustLLMeter
Primary use caseLLM evaluation + prompt iterationLLM cost monitoring + budget alerts
Setup methodSDK install + logger.wrap callsRead-only API key (no code)
Time to first dashboardHours (instrument app + run evals) 30 seconds
Code changes requiredYes (SDK + log wrapping) None
Cost trackingAvailable as side feature Core product (only feature)
Eval / scoring frameworkYes (primary feature)No (out of scope)
Sees prompts/completionsYes (logged for eval) No (only billing data)
Open sourceSDK yes, platform hostedAGPL-3.0 (full stack)
Pricing entryFree limited → paid tiersFree for 1 provider → $19/mo Pro
Best fitTeams doing prompt evaluationTeams that only need cost numbers

Eval platform vs cost tracker

Pay for the tool that matches the work you do.

Braintrust (eval platform)

  • Prompt experiments + regression eval
  • Scoring framework + dataset management
  • Cost tracking (side feature)
  • Requires SDK + log wrapping
  • Captures prompts and completions

LLMeter (cost tracker)

  • Cost per model / customer / environment
  • Budget alerts + anomaly detection
  • Read-only API key (no SDK)
  • Never sees prompts or completions
  • No eval framework (use Braintrust if needed)

Migration FAQ

Is LLMeter trying to replace Braintrust?

Not directly. Braintrust is an eval platform — it shines at prompt iteration, regression testing, and scoring models against datasets. LLMeter is a focused cost tracker. The overlap is the cost dashboard: if that is the only Braintrust feature you log in for, LLMeter does it without the SDK.

Do I have to choose between them?

No — they are complementary. Many teams run Braintrust for eval and LLMeter for cost in parallel: Braintrust gives quality numbers, LLMeter gives spend numbers, both stay narrow at what they do. The question is whether you are paying for Braintrust's cost features that you are not using.

Does LLMeter capture prompts and completions like Braintrust does?

No. Braintrust logs prompts and completions so it can run evals against them. LLMeter uses read-only API keys to pull billing data — it never sees a single prompt. If you need to inspect the content of LLM calls, LLMeter is not the right tool.

How accurate is the cost data compared to Braintrust?

LLMeter pulls cost data directly from the provider, so it matches your invoice exactly. Braintrust computes cost from the requests its SDK observes — usually identical, but it can drift if your code makes calls outside the wrapped paths. LLMeter's source of truth is the provider, not the instrumented layer.

What about per-customer cost attribution?

Both can do it. In Braintrust you tag spans with metadata at log time. In LLMeter you pass metadata (user_id, tenant_id) when making the OpenAI / Anthropic call and the provider attaches it to the usage record LLMeter reads — same outcome, no SDK in the middle.

Is LLMeter free?

Yes — the Free plan covers 1 provider with 30-day retention, forever. The Pro plan ($19/mo) adds unlimited providers, alerts, anomaly detection, and 1-year retention. Team plans start at $49/mo.

Cost-only, no eval-platform weight.

Run LLMeter alongside Braintrust for a week and see if the cost panel is all you need. Free forever for 1 provider.

No credit card required. Free plan never expires.