A tag manager built for agents.

So lightweight it ships 1% of the JavaScript of a traditional tag manager — measured, not marketed.

A real container with seven destinations — GA4, Google Ads, Meta, TikTok, LinkedIn, Hotjar and a first-party collector — compiles to 4.2KB gzipped. The vendor SDKs a traditional setup loads for the same stack weigh 908KB, before the tag manager itself. There is no UI, no templates, no interpreter shipped to your visitors: config lives in git, a compiler emits only the code your tags need, and your agent — Claude, ChatGPT, anything that speaks MCP — operates it: plan → approve → apply.

The whole install

<script src="https://cdn.tagless.foo/t/<your-site>.js" defer></script>

How it works

• No UI. Declarative config in git; the manager is a compiler. Your agent plans, you approve, it ships — Terraform-style, with the exact network payloads and the byte delta in the diff.
• No templates. Agents write integrations on the fly; the community contributes vendor contracts — specs and fixtures that CI validates generated code against.
• No vendor SDKs. Direct mode fires the tracking request itself, at parity with what the SDKs send: consent gating, advanced matching, sessions, click ids, ecommerce items.

See a compiled container

curl https://cdn.tagless.foo/t/demo.js # runtime + GA4 + Meta, ~2.5KB gz

Use it end to end

Install, the agent conversations (migrate a GTM container, add a pixel, track "that button"), the gotchas, and how to keep your agent's token bill low: read the guide.

Open source, deliberately

tagless is open source by conscious choice, not by default: it was built to prove a shape, and its author won't be its sustained maintainer. It's structured so you don't need him — the spec is the doc, the CI is the contract, new vendors are YAML plus fixtures, and Apache-2.0 means you never ask permission. If the community takes it further, that's the plan working.