Data & Analytics

Benchmark Heaven publishes cost and capability comparisons across AI models. Its own description is "The most detailed cost-capability analysis in AI. Every model. Every Benchmark. Actual Costs." One published comparison, JevBench v1.3.0, states that 52 Jev-class systems were tested on 534 decisions, with a named leader at a specific score. The site presents itself as a benchmark and cost reference rather than a model host: it reports results and prices instead of running inference itself.

Metabase is a business intelligence tool aimed at the people who have questions rather than the people who write SQL. It connects to a database, reads the schema, and presents a point-and-click query builder alongside a native SQL editor for users who want one. The product's distinguishing choice is that it does not require a modelled semantic layer before it is useful. Connect a database and the tables are immediately explorable, with questions saved as reusable cards and assembled into dashboards. Alerts and subscriptions push a saved question to email or Slack on a schedule. Metabase is open source with a paid enterprise edition that adds features such as granular permissions, sandboxing, and audit logging. It is a good fit for a team that needs self-service reporting over an existing database; it is a weaker fit for heavy financial modelling or for organisations that need a governed semantic layer with versioned metric definitions.

Plausible is a web analytics tool that reports on traffic without cookies and without collecting personal data. The tracking script is small, and because no cookie is set, the product does not require the consent banner that cookie-based analytics trigger in many jurisdictions. The interface is deliberately narrow. It reports page views, referrers, entry and exit pages, countries, devices, and goals, and it does not attempt to build user profiles or cross-site graphs. A site can publish its dashboard publicly, which some teams use as a transparency signal. The product is open source and can be self-hosted, though the hosted service is the main commercial offering. It is a good fit for a site that needs to know which pages are read and where visitors arrive from; it is a poor fit for anyone who needs funnels, cohort retention, or per-user behavioural analysis, which is a different class of tool.

Langfuse is an observability platform for applications built on large language models. It records traces of model calls - the prompt, the response, the latency, the token count, the cost - so a team can see what an application actually did rather than inferring it from logs. The tracing model follows the shape of an LLM application. A trace is a single request; spans inside it represent retrieval, tool calls, and model invocations. Because the structure is explicit, a slow or expensive step is visible rather than buried in an aggregate. Beyond tracing, Langfuse provides evaluation and prompt management. A dataset of inputs can be run against a prompt and scored, either by a model or by a person, so a change to a prompt has a measured effect. The platform is published as open source and can be self-hosted, which matters for teams whose prompts contain sensitive data.

PostHog is a product analytics platform that bundles several tools a product team usually buys separately: event analytics, session replay, feature flags, experiments, and surveys. The argument for bundling is that these tools answer related questions and are more useful when they share one event stream. The analytics core is an event pipeline. A product sends events with properties, and the platform builds funnels, retention curves, and paths from them. Session replay records the interface alongside those events, so a drop-off in a funnel can be watched rather than only counted. Feature flags and experiments read from the same user identity, which is what closes the loop: a flag can gate a change, an experiment can measure it, and the funnel shows the result. The platform is published as open source and can be self-hosted, though the self-hosted deployment is a heavier operational commitment than the hosted service.

Grafana is a visualization and dashboarding platform. It queries data sources - Prometheus, Loki, PostgreSQL, and many others - and renders the results as panels arranged on a dashboard. The product's longevity comes from that separation: Grafana does not store most of the data it displays, it queries systems that do. The query model is per-panel. A panel selects a data source, runs a query, and applies a visualization. Because data sources are pluggable, the same dashboard tool works across a metrics database, a log store, and a SQL database without moving data between them. Alerting is built on the same queries. A rule evaluates a query on a schedule and fires when a condition holds, which means an alert and the dashboard panel that motivated it are defined in the same language. The core is open source and self-hostable, and the hosted service is the commercial layer.