Burnstack vs Langfuse
Langfuse is an open-source LLM engineering platform: tracing, evaluations, prompt management and cost per trace. Burnstack is about the bill itself: what you were charged, across all your dev vendors.
The short version
- Choose Langfuse if you're debugging and evaluating an LLM app and want cost per user, session or feature; you want prompt management and evals in the same tool.
- Choose Burnstack if you want billed AI cost next to hosting, database and CI costs; you don't want to instrument your code to see what you spend.
At a glance
| Burnstack | Langfuse | |
|---|---|---|
| What it is | Desktop app (Mac; Windows soon) that reads your vendors' billing APIs | Open-source LLM tracing, evals and prompt management |
| How it gets data | Billing APIs | Traces from your instrumented code |
| Cost figures | What you were billed | From traces, or model prices times usage |
| Evals and prompt management | — | ✓ |
| Price | Free forever; Enterprise for teams | Free hobby tier; Core $29/month; free to self-host |
Who owns the data
Langfuse runs in its cloud or self-hosted under the MIT license. Costs come from your traces or are inferred from model prices and token usage, with caveats such as tokenizer differences and reasoning tokens (docs). It doesn't read provider billing APIs.
With Burnstack, API keys stay in your computer's keychain and your spend history is a file on your disk. You can export or delete all of it any time, and none of it is sent to us. What owning your data means.
Which bills each one reads
| Vendor | Burnstack | Langfuse |
|---|---|---|
| Vercel | ✓ | AI SDK and AI Gateway tracing, not billing |
| Neon | ✓ | — |
| Supabase | ✓ | — |
| OpenAI | ✓ | SDK tracing |
| Anthropic | ✓ | SDK tracing |
| GitHub billing | ✓ | Copilot session tracing, not billing |
| OpenRouter | ✓ | Request tracing, not billing (source) |
| DigitalOcean | ✓ | — |
| Twilio | ✓ | — |
| Stripe fees | ✓ | — |
| Cloudflare | ✓ | Not found |
| Netlify | ✓ | — |
| AWS | ✓ | Bedrock tracing, not billing (source) |
| Datadog | ✓ | — |
| Sentry | ✓ | — |
| Cursor | ✓ | — |
| PlanetScale | ✓ | — |
| Azure | ✓ | Azure OpenAI tracing, not billing (source) |
| Google Cloud | ✓ | Vertex AI tracing, not billing (source) |
| MongoDB Atlas | ✓ | — |
| Railway | ✓ | — |
Burnstack also reads Render, Heroku, Akamai (Linode), Vultr, Fastly, Runpod, xAI, Together AI, fal.ai, Runway and Upstash (new in 0.9.0). We haven't checked Langfuse's coverage of those yet, so they aren't in the table.
Pricing
Hobby is free (50k units a month), Core $29/month, Pro $199/month and Enterprise $2,499/month; self-hosting the core is free (Langfuse pricing).
Burnstack is free forever. Enterprise (several accounts per vendor, monthly reports, team features) is priced per team. Pricing.
Where Langfuse is better
- Cost per user, session and feature at the trace level.
- Evaluations and prompt management alongside cost.
- Self-hostable under MIT.
- Monitors on cost, quality and latency with Slack and webhook alerts.
Where Burnstack is better
- Billed cost from OpenAI and Anthropic, not an estimate.
- Hosting, database and GitHub costs in the same place.
- Nothing to instrument; works with any code.
Langfuse joined ClickHouse in January 2026, with no planned license changes (Langfuse). The two tools work well together: Langfuse for what happens inside your AI features, Burnstack for the whole bill.