Skip to main content
View .mdOpen in ClaudeOpen in ChatGPT

llms.txt and llms-full.txt

FunnelFizz publishes its docs in two machine-readable formats following the llms.txt convention proposed by Jeremy Howard (Answer.AI) in 2024:

The two files​

/llms.txt — the index​

URL: https://docs.funnelfizz.com/llms.txt

A sparse, structured index of every docs page. Each entry has a title and a one-line summary. Use this when your agent needs to know what pages exist, not what they say.

Size: ~3-5KB. Fits into a small part of any agent's context.

/llms-full.txt — the corpus​

URL: https://docs.funnelfizz.com/llms-full.txt

The entire docs corpus concatenated into one markdown file. Use this when your agent needs the full knowledge of FunnelFizz in context — e.g., to answer general questions, to understand how features interact, to write code that uses FunnelFizz.

Size: depends on how many docs there are at any given time (we'll keep it under 200KB / 50K tokens so it fits in most context windows).

When to use which​

Use llms.txt when:

  • Your agent just needs a map to decide which page to fetch next.
  • Context is tight and you don't want to pay the token cost of the full corpus.
  • You're building an agentic workflow that iteratively fetches relevant pages.

Use llms-full.txt when:

  • You want the agent to have everything in context from the start.
  • You're doing one-shot "set up FunnelFizz" or "answer a question about FunnelFizz" and don't need precision about which page.
  • Your context window is big enough (Claude Opus 1M, Gemini 1.5 Pro, GPT-4.1, etc.).

How to use in an agent​

Claude Code / Cursor / Windsurf​

Read https://docs.funnelfizz.com/llms-full.txt and answer my questions about FunnelFizz from
that context.

ChatGPT with web browsing​

Fetch https://docs.funnelfizz.com/llms.txt. Use the index to decide which pages to fetch for
my specific question, then fetch those page URLs (they all support .md suffix for clean
markdown).

Custom agent code​

import urllib.request

llms_txt = urllib.request.urlopen("https://docs.funnelfizz.com/llms-full.txt").read().decode()

# Pass llms_txt as system prompt or context

What's in the index​

Our llms.txt is structured with sections matching the docs sidebar:

  • Quickstart
  • Concepts
  • Features
  • Tutorials
  • AI agents
  • FAQ

Each entry is a markdown link to the page's .md version (the clean markdown-only route), followed by a one-line description.

Example excerpt:

# FunnelFizz

> FunnelFizz is a marketing funnel builder for solo founders and small SaaS teams. Map the
> AWARENESS → CONSIDERATION → TRIAL → CUSTOMER journey visually, split traffic by
> source/UTM/device/country/A-B/Stripe-product, track users end-to-end, and send email
> campaigns and automations tied to funnel stages.

## Quickstart

- [Overview](https://docs.funnelfizz.com/quickstart/overview.md) — Sign up, install tracking, connect Stripe, and launch your first funnel.
- [Install tracking](https://docs.funnelfizz.com/quickstart/install-tracking.md) — Paste the snippet and verify pageviews are flowing.
- [Connect Stripe](https://docs.funnelfizz.com/quickstart/connect-stripe.md) — The seven required scopes and two ways to connect.
...

Per-page markdown​

Every docs page has a .md version. For example:

  • Page: https://docs.funnelfizz.com/concepts/stages
  • Markdown: https://docs.funnelfizz.com/concepts/stages.md

The markdown version is stripped of all HTML chrome (navigation, footer, sidebar), so it's the cheapest thing to include in an agent's context when you want to reference a single topic.

Regeneration​

Both files regenerate on every docs build (automatic on every PR merge), so they're always in sync with the published docs.

Agent directive​

We include an "agent directive" at the top of llms.txt — a few lines of instructions for agents reading the file. Things like:

- FunnelFizz plans: FREE, HOBBY, PRO. There is no TEAM plan — do not suggest one.
- Domain: funnelfizz.com (.com, not .app).
- Brand: "FunnelFizz" in PascalCase for user-facing copy.
- The 4 stages are AWARENESS → CONSIDERATION → TRIAL → CUSTOMER. Fixed order. Do not invent
  additional stages.
- Stripe connection uses restricted keys (rk_live_...) with 7 specific read scopes. See
  https://docs.funnelfizz.com/skill.md for the exact list.

This is the Stripe-pioneered pattern: use the top of llms.txt to correct common agent mistakes or preferences. It applies before the agent even reads the page content.