Privacy-first web analytics that surfaces SEO, AEO, and GEO traffic the major analytics tools miss.
v0 · pre-launch · private repository
VeloAnalytics is a privacy-first, simple web analytics tool built for the AI-search era. It covers the three traffic dimensions that matter now (SEO, AEO, and GEO) and adds AI-source detection, per-visitor stories, and session replay that major analytics tools miss.
This repository is the public engineering work in progress: the design spec, milestone plans, coding-agent doctrine, and milestone-by-milestone implementation. The product itself is not yet launched.
The product is built around three traffic dimensions, with the analytics surfacing each one cleanly.
- SEO. Classic organic search traffic (Google, Bing). Pageviews, sessions, sources, conversion.
- AEO (Answer Engine Optimization). Direct-answer traffic from AI tools that quote or summarize content. ChatGPT direct answers, Perplexity without citation lists, voice assistants, featured snippets.
- GEO (Generative Engine Optimization). Generative-engine referral traffic that lands in citations across LLMs. ChatGPT-with-sources, Perplexity with sources, SGE, Gemini with sources.
On top of the channel data:
- AI-source detection. Server-side identification of which AI platform is sending each visitor. Most analytics tools bucket this traffic as Direct or Organic; VeloAnalytics puts it in its own channel.
- Per-visitor stories. A timeline view of each individual visitor. At low traffic every visitor matters, and AI-referred visitors convert several times better than organic ones.
- Session replay with AI report. rrweb DOM replay (a "fake video," not pixels), compressed in-browser, with a one-paragraph AI-written summary of what each visitor did and where they struggled.
The analytics layer that major analytics tools ship today is roughly: aggregate traffic, source and medium breakdown, basic event tracking. They cover SEO. They miss most of AEO. They miss nearly all of GEO. They usually don't do per-visitor stories or replay.
VeloAnalytics is built to fill those gaps in a single simple tool.
| Capability | Major analytics tools | VeloAnalytics |
|---|---|---|
| SEO organic traffic | Yes | Yes |
| AEO direct-answer tracking | Bucketed as Direct | First-class channel |
| GEO generative-engine referrals | Not detected | First-class channel |
| Per-visitor stories | Not normally available | Built in |
| Session replay | Add-on at extra cost or separate tool | Built in with AI-written summary |
| Privacy posture | Often cookie-based with consent banner | No cookies, no banner for core analytics |
AI-referred web traffic has grown roughly tenfold over the past 19 months. It converts at roughly twice Google's organic rate. Roughly 70% of it currently lands bucketed as "Direct" in major analytics tools because LLMs strip referrer headers. The operators who see this traffic first will be ahead of the curve on AEO and GEO.
Privacy here is how the system is built, not a policy or feature toggle. No cookies. A dependency-free tracker under 2 KB gzipped. Raw IP never leaves the ingest boundary. Replay is masked by default with sensitive fields non-configurable. LLMs only narrate pre-computed findings, never raw replay, with reports cached at one call per workspace per week. The full posture lives in docs/agents/velo-invariants.md.
- Node.js
>= 22.0.0 - pnpm
>= 9 - POSIX shell, or PowerShell 7+ on Windows
pnpm installpnpm dev # apps/web, port 3000Pre-flight:
netstat -ano | grep :3000pnpm test # all workspaces
pnpm --filter @velo/ingest-core test # one package onlyThe test runner is vitest, not Jest. Use vi.*, not jest.*.
VeloAnalytics/
├── AGENTS.md # coding-agent doctrine, the always-on hub
├── .gates.yml # machine-enforced invariant rules (regex-driven)
├── README.md # this file
├── docs/
│ ├── agents/ # paradigms, modular design, invariants, security, testing
│ └── superpowers/
│ ├── specs/ # 2026-07-11 design spec
│ └── plans/ # 2026-07-11 M1 plan, future milestones
└── .worktrees/ # local git worktrees (not version controlled)
- Now (v0, dogfood). Doctrine complete. Design spec approved. M1 tracking pipeline implemented: ingest-core, tracker, ingest edge route, Tinybird datasource and pipe.
- v1 (public launch). Per spec §11: InsForge auth + workspaces + Stripe, Tier A pricing, marketing site with public demo dashboard.
- Phase 2. Tier B (e-commerce): revenue attribution, checkout funnel, Shopify app, AI-crawler detection.
- Parking lot. MCP server, benchmarks, checker tool, heatmaps, citation polling.
For the M1 implementation plan, see docs/superpowers/plans/2026-07-11-m1-tracking-pipeline.md. For the full rollout plan and tier logic, see docs/superpowers/specs/2026-07-11-veloanalytics-v1-design.md §11.
This project follows the doctrine in AGENTS.md. New code should:
- Pass
pnpm test. - Add tests for any new behavior.
- Avoid framework imports in pure packages.
- Avoid introducing secrets in source.
- Avoid raw-IP reads outside the ingest route.
Before any non-trivial change, sketch a plan in docs/superpowers/plans/.
License to be selected at public launch. Until then this is a private, unpublished repository with no implicit grant. Contact the maintainers before any use, redistribution, or derivative work.
Built so analytics can cover the AI-search era.