Pug is an open-source product analytics platform. Capture events, identify the people behind them, and explore behavior through funnels, retention, trends, segmentation, and user-flow analysis, all surfaced on customizable dashboards. Built in Go on PostgreSQL, ClickHouse, and NATS. Features Event ingestion: a NATS-backed capture pipeline with automatic geo, user-agent, bot-detection, and web-attribution enrichment. Web analytics: an overview auto-built from your events: visitors, sessions, pageviews, bounce rate, and traffic sources, with no tile to wire up. Live view: every visitor on a real-time world map, down to the person and the event they just fired. Profiles: identify and alias the users behind events, with a ClickHouse-backed profile and activity API. Insights: trends, funnels, retention, segmentation, user flow (Sankey), and top-K breakdowns, with filtering, breakdowns, and period-over-period comparison. Dashboards: compose insight and markdown tiles on a grid with a board-level time window, accelerated by a pre-aggregated rollup fast path. MCP server: Claude, ChatGPT, or any MCP client can query your insights, events, and profiles, and build dashboards for you. Privacy & compliance: GDPR/DPDP data-subject erasure of a person's events and profile. Overview: a web-analytics view auto-built from your events, with the previous period alongside. Breakdowns: pageviews by country, plus locations, devices, and events; click any value to filter the whole view. Live view: every visitor on a live world map; click one to fly to them and see the page, device, and profile behind it. Events: every raw event as it lands, with its location, platform, properties, and the person behind it. Dashboards: compose insight and markdown tiles on a time-windowed grid. Insights: trends, filters, and breakdowns; here product views split by country. Funnels: step-by-step conversion; here product view to purchase. Profiles: the person behind the events, with traits, sessions, and activity. MCP server Pug serves a Model Context Protocol server, so an AI assistant can query your analytics. Sixteen tools read projects, dashboards, insights, events, sessions, and profiles, and one creates and edits dashboards. In Claude, add Pug from the connectors directory. Other clients take the endpoint as a custom connector. Connect by signing in to Pug, or with a project's private (prv_) API key in the x-api-key header. A self-hosted server offers sign-in only once PUG_API_BASE_URL is set. Setup steps for each client are in the MCP docs. Tech stack Go: backend services and workers, exposed over Connect RPC (HTTP/2) PostgreSQL: relational store (orgs, projects, dashboards, auth) ClickHouse: analytical store for events, insights, and profiles NATS: messaging backbone for the ingestion and worker pipelines Self-hosting See the self-hosting guide. A Pug instance is this repo's backend, published as container images on ghcr.io/pug-sh, plus the dashboard from pug-sh/app, served as static files. Quick start # Build the binary into bin/pug make build # Start dev infrastructure (PostgreSQL, NATS, ClickHouse) make infra # Run migrations (Postgres, NATS, ClickHouse) ./bin/pug migrate # Seed the demo project with events, profiles, and dashboards # (resets the local Postgres and ClickHouse databases; see "Demo data" below) ./bin/pug seed # Start the dev server + workers together ./bin/pug dev docs/development.md walks through each step, including signing in as the demo org's admin. Environment variables are documented in .env.example. Google and generic OIDC sign-in are documented in docs/authentication.md, with a ready-to-copy config.example.json. Demo data ./bin/pug seed fills a "Pug & Pals" demo project with ~6 months of history, so dashboards, insights, and profiles have something to show. Profiles are seeded only for users that produced events, so the data is internally consistent. Local databases only. By default seed migrates Postgres and ClickHouse all the way down and back up, dropping every table, not just the demo rows. There is no confirmation prompt and no environment check: it connects to whatever DATABASE_URL and CLICKHOUSE_URL resolve to, and an already exported variable beats .env. Point it at a disposable local or demo database. Pass --no-reset to keep the schema; it then deletes the demo project's own events and profiles before re-seeding them, leaving other projects untouched. Volume is tunable with --count (default 500,000 events) and --batch (default 10,000 events per insert). Run it from the repo root; it reads the same .env as the rest of the CLI. Two read-only accounts are seeded, plus an admin if you ask for one: woof@pug.sh owns the demo org and its data. It has no password. snoop@pug.sh is a viewer with no password; open /demo with PUG_DEMO_ENABLED=true to sign in as it. admin@pug.example is the demo org's admin, seeded only when PUG_DEMO_ADMIN_PASSWORD is set (16 characters to 72 bytes, no surrounding whitespace), with that password. The address can't receive mail. Nothing logs the password, and later runs never change it, so change it in the dashboard, to 16 characters or more, and update the variable to match, since the seeder restores the admin's role only while they match. A default pug seed resets the database, though, and so recreates the admin from the current value. For a live stream of traffic instead of a one-shot backfill, set PUG_DEMO_ENABLED=true: ./bin/pug dev then also runs the demo worker. It backfills an empty project once, so after pug seed it skips straight to playing new sessions out in real time. If you seed with a smaller --count, lower PUG_DEMO_SEED_COUNT to match: the worker reads a project holding fewer events than that as an interrupted backfill and warns rather than topping it up. Development make test # run tests (race detector enabled) make cover # run tests and write coverage.out make lint # lint Go code make sqlc # regenerate sqlc queries after editing SQL make rpc # regenerate protobuf code after editing .proto files make templ # regenerate templ email templates Architecture Per-subsystem documentation lives in docs/architecture/ (insights, ClickHouse, profiles, ingestion, email, telemetry). Contributor guidance and conventions are in CLAUDE.md. Copyright (C) 2025 pug.sh contributors. Pug is licensed under the GNU AGPL v3.0, or (at your option) any later version. Contributors sign a CLA before their first pull request is merged; see CONTRIBUTING.md.
AI NEWS JOURNAL
Hacker News Show HNAI NEWS / JOURNAL
Show HN: Pug.sh – open-source product analytics
Article URL: https://github.com/pug-sh/pug Comments URL: https://news.ycombinator.com/item?id=50576620 Points: 1 # Comments: 0
번역 서비스가 일시 중지되어 원문을 표시합니다.