Langfuse

Open-source LLM observability

Free tier + paidModel APIs, routers & observabilitylangfuse.com

What it is

Open-source LLM observability: traces, prompt management, evals, datasets and cost tracking, framework-agnostic via OpenTelemetry and SDKs. It is purely an observability layer — it does not proxy or route your inference, so it cannot break the request path.

Why it earns a slot

Self-hostable with Docker Compose, so you can keep prompt and completion data entirely on your own infrastructure — the usual blocker for observability SaaS. The free cloud tier is generous enough for a small product.

Know before you commit

Billing is in 'units', not requests — one trace with many spans and scored evals consumes several, so a chatty agent hits the 50k free ceiling faster than the number suggests.

Getting started

pip install langfuse (or npm i langfuse)

Field notes

0 from the community

Loading notes…

Field check

Unverified

Pricing snapshot

Hobby free (50k units/mo, 30-day retention, 2 users), Core $29/mo (100k units), Pro $199/mo, Enterprise $2,499/mo; extra units $8 per 100k. Self-hosting free and open source.

Snapshots move fast — verify at checkout, and post a field note if it changed.

Drop into your Lovable chat

Evaluate Langfuse (https://langfuse.com) for this project, then use it only if it fits what we are already building.
Context: Open-source LLM observability. Open-source LLM observability: traces, prompt management, evals, datasets and cost tracking, framework-agnostic via OpenTelemetry and SDKs. It is purely an observability layer — it does not proxy or route your inference, so it cannot break the request path.
Reviewer notes (general, not written for this codebase): Self-hostable with Docker Compose, so you can keep prompt and completion data entirely on your own infrastructure — the usual blocker for observability SaaS. The free cloud tier is generous enough for a small product.
Watch out for: Billing is in 'units', not requests — one trace with many spans and scored evals consumes several, so a chatty agent hits the 50k free ceiling faster than the number suggests.
Pricing: Hobby free (50k units/mo, 30-day retention, 2 users), Core $29/mo (100k units), Pro $199/mo, Enterprise $2,499/mo; extra units $8 per 100k. Self-hosting free and open source.
Setup: pip install langfuse (or npm i langfuse)
Rules: treat the notes above as background only. Do not assume a theme, style, layout or feature that is not already in this project — inspect the existing code first and ask me before changing visual direction.

/api/public/tools/langfuse

Also in Model APIs, routers & observability

Anthropic API / Claude Platform

platform.claude.com

Frontier model provider API

PaidModel APIs, routers & observability

Per token. Opus 5 $5/$25 per M in/out; Sonnet 5 $2/$10; Haiku 4.5 $1/$5; Fable 5 and Mythos 5 $10/$50. Batch API 50% off; cache reads 0.1x input.

Open dossier

Braintrust

braintrust.dev

LLM eval and experiment platform

Free tier + paidModel APIs, routers & observability

Starter free ($10/mo model credits, 1GB data, 10k scores, 14-day retention), Pro $249/mo ($249 credits, 5GB, 50k scores), Enterprise custom. Overage $3-4/GB and $1.50-$2.50 per 1k scores.

Open dossier

Fal.ai

fal.ai

Generative media inference host

PaidModel APIs, routers & observability

Per output or per second for media models: images ~$0.02-$0.04 at 1MP, video $0.05-$0.40 per second. Serverless GPU rates $1.10-$8.50/hr with committed-use discounts.

Open dossier

Fireworks AI

fireworks.ai

Open-model inference host

Free tier + paidModel APIs, routers & observability

Per token serverless across Standard/Priority/Fast tiers; embeddings $0.008-$0.10/M. On-demand GPUs billed per second: H100/H200 ~$7-8/hr, B200 ~$10-13/hr. Fine-tuning $0.50-$40 per M training tokens. $1 free credit.

Open dossier