Logfire brings LLM tracing and evaluation into full-stack observability, with PostgreSQL-compatible SQL across the surrounding application, infrastructure, traces, logs, and metrics.
Trusted by teams building production software and AI
| Feature | Logfire | Langfuse |
|---|---|---|
| Observability scope | AI and application telemetry across traces, logs, and metrics | LLM tracing, prompts, evaluation, and datasets |
| Ingest volume | 200B+ records and evaluation scores per month | 90B+ observations per month (self-reported) |
| LLM Tracing | Yes | Yes |
| Token/Cost Tracking | Yes | Yes |
| Prompt Playground | Yes | Yes |
| Human annotations & queues | Annotations on production runs + annotation queues | Annotations on production runs + annotation queues |
| Hosted evaluation | LLM judges and deterministic evaluators on live traffic | Model-based and custom evaluators |
| Continuous improvement | Scheduled proposals and model-swap recommendations | Evaluation and prompt workflows |
| Full-stack workflow | Traces, logs, metrics, and infrastructure views | Trace-oriented LLM observability |
| Database/API context | Native OTel integrations and service context | Can ingest arbitrary OTel spans within traces |
| Query Interface | PostgreSQL-compatible SQL | Product UI, metrics API, and public API |
| MCP server | SQL queries over production telemetry from your editor or AI agent | Query observations, metrics, scores, datasets, and annotation queues |
| OpenTelemetry | Native across full-stack telemetry | Native OTLP trace ingestion |
| Self-Hosting | Enterprise | Open Source |
| Free Tier | 10M logs, spans, and metrics | 50K units |
| Python SDK | Native OpenTelemetry SDK | Native OpenTelemetry SDK |
| JavaScript SDK | Native OpenTelemetry SDK | Native OpenTelemetry SDK |
| Any OTel Language | Yes | Yes |
Agent behavior depends on the application around it. Logfire makes model calls, tools, browser RUM and session replay, database queries, APIs, services, logs, metrics, and infrastructure available in the same product, instead of centering the investigation on LLM observations alone.
Logfire applies OpenTelemetry across the entire application, including model and tool calls, databases, APIs, services, logs, metrics, and infrastructure. That keeps instrumentation portable while preserving the context needed to explain production behavior.
Query telemetry with PostgreSQL-compatible SQL, including joins, aggregations, and ad hoc questions that were not anticipated by a saved dashboard. The same interface is available to coding agents through Logfire's MCP server.
Logfire ingests more than 200 billion records and evaluation scores per month. That volume includes the full-stack telemetry around the model calls, not only LLM observations, and pricing stays $2 per million records at any volume.
Run hosted LLM judges and deterministic evaluators on sampled production traffic, route low-scoring runs to human review, and turn the resulting evidence into scheduled optimization proposals and model-swap recommendations.
Start free with 10 million spans, logs, and metrics each month. No credit card required.
| Illustrative monthly volume | Logfire | Langfuse |
|---|---|---|
| 1M stored units/month | $49 Team (30 days) · $249 Growth (up to 90 days) | $101/month (90 days, unlimited users) |
| 10M stored units/month | $49 Team (30 days) · $249 Growth (up to 90 days) | $731/month (90 days, unlimited users) |
| 50M stored units/month | $129 Team (30 days) · $329 Growth (up to 90 days) | $3,331/month (90 days, unlimited users) |
Calculation scope
Langfuse Core's graduated cloud-unit pricing compared with both the lowest applicable 30-day Logfire plan and Logfire Growth for up to 90-day retention at the same stored-object count. The mapping is object-level, not run-level equivalence.
Basis: USD; public monthly list price. Prices checked .
For this illustration, each Langfuse trace, observation, or score is mapped to one Logfire telemetry record so the two published meters can be calculated from the same object count.
$29 $29 900K ÷ 100K × $8 $72 $49 $49 (1M − 10M included) ÷ 1M × $2 $0 $249 $249 (1M − 10M included) ÷ 1M × $2 $0 For this illustration, each Langfuse trace, observation, or score is mapped to one Logfire telemetry record so the two published meters can be calculated from the same object count.
$29 $29 900K ÷ 100K × $8 $72 9M ÷ 100K × $7 $630 $49 $49 (10M − 10M included) ÷ 1M × $2 $0 $249 $249 (10M − 10M included) ÷ 1M × $2 $0 For this illustration, each Langfuse trace, observation, or score is mapped to one Logfire telemetry record so the two published meters can be calculated from the same object count.
$29 $29 900K ÷ 100K × $8 $72 9M ÷ 100K × $7 $630 40M ÷ 100K × $6.50 $2,600 $49 $49 (50M − 10M included) ÷ 1M × $2 $80 $249 $249 (50M − 10M included) ÷ 1M × $2 $80 “Covers 10 million logs, spans, and metrics every month.”
The official calculator shows 20 million records and no extra seats totaling $69/month on Team.
“100k units / month included, additional: $8/100k units.”
The official calculator shows a $37 Core total at 200,000 monthly units: $29 base plus $8 usage.
“Units are either traces, observations or scores.”
Langfuse documents the formula as traces plus observations plus scores, including units created by evaluations and annotations.
Logfire puts LLM tracing, prompts, evaluation, and optimization inside full-stack observability, so model and tool calls stay connected to browser RUM and session replay, databases, APIs, services, logs, metrics, and infrastructure.
It depends on volume and retention. Logfire includes 10 million records a month; paid plans charge $2 per million additional records. Langfuse Core is $29, includes 100,000 units, 90-day access, and unlimited users, then applies graduated pricing to traces, observations, and scores. The worked rows quote Logfire Team throughout rather than setting a free tier against a paid one, so the smallest workload shows $49 against $101 and Logfire's record pricing stays the lower one as stored-object volume grows. The table keeps the 30-day and 90-day retention options visible rather than treating them as interchangeable.
Yes. Logfire accepts standard OpenTelemetry traces, so you can send the same telemetry to Logfire while you compare production investigations and move integrations incrementally.
Yes. Logfire is available as a managed cloud service or as an Enterprise self-hosted deployment, and its SDK is open source.
Yes. Run hosted LLM judges or deterministic evaluators without supplying a model key, sample live production traffic, and route low-scoring runs to a human annotation queue. Scores become telemetry you can query, dashboard, and alert on.
Get started with 10 million free spans, logs, and metrics per month. No credit card required.